Incomplete Input Research Articles

Application of neural networks for predicting electric load

The paper shows that the operational management of the power consumption regime is reduced to solving the problem of operational forecasting of the enterprise's load. The paper analyzes the works devoted to the forecasting of electric loads of power systems and industrial enterprises. It is shown that in order to achieve the required forecast accuracy, it is advisable to use adaptive forecasting procedures and, in particular, to use artificial neural networks. The use of artificial neural networks for forecasting the load of industrial enterprises is due to their properties, such as the ability to learn, reliability with incomplete input information, and the rapid response of the learned network to input influences. The conditions for determining the configuration of a neural network are considered. The structure of a neural network for predicting the electrical load of an industrial enterprise is presented. The process of training an artificial neural network with fitting the model to data from a retrospective sample is presented. The considered models of daily load forecasting were investigated on retrospective data on the modes of electricity consumption of a chemical enterprise with normalization of input data. The graphs of the actual total load and the forecast obtained by the artificial neural network models are presented. The research has shown that the use of an artificial neural network allows for a qualitative forecast of the enterprise's load under normal operating conditions of the equipment.

Comparative Analysis of Human and Artificial Intelligence Planning in Production Processes

Artificial intelligence (AI) has found applications in enterprises′ production planning processes. However, a critical question remains: could AI replace human planners? We conducted a comparative analysis to evaluate the main task of planners in an intermittent process: planning the duration of production orders. Specifically, we analysed the results of a human planner using master data and those of an AI algorithm compared to the actual realisation. The case study was conducted in a large production company using a sample of production products and machines. We were able to confirm two of the three research questions (RQ1 and RQ3), while the results of the third question (RQ2) did not meet our expectations. The AI algorithms demonstrated significant improvement with each iteration. Despite this progress, it is still difficult to determine the exact threshold at which AI outperforms human planners due to the unpredictability of unexpected events. Even though AI significantly improves prediction accuracy, the inherent variability and incomplete input data pose a major challenge. As progress is made, robust data collection and management strategies need to be integrated to bridge the gap between the potential of AI and its practical application, fostering the symbiosis between human expertise and AI capabilities in production planning.

Physics-informed ensemble learning with residual modeling for enhanced building energy prediction

Accurate modeling of building energy use is important to support a diverse spectrum of its downstream applications, such as building energy efficiency assessment, resilience analysis, smart control, etc. As mainstream approaches of building energy modeling, physics-based modeling builds on different fidelities of physics rules yet are usually compromised in modeling accuracy due to insufficiency of physics rules to capture real-world dynamics and incomplete input information. Data-driven approaches are computationally efficient, but black box (uninterpretable) in nature. For improved modeling of building energy use, this work proposes a physics-informed ensemble learning approach in building energy prediction through residual modeling. Specifically, we first analyze the components of building energy use data. Evidence suggests that the building energy use data can be decomposed into physics-driven part, occupant-driven part, and white noise. Second, high-fidelity physics-based building models (EnergyPlus) and low-fidelity ones (RC models) are developed to capture the physics-driven part while time series methods are explored as the residual modeling approach to capture the occupant-driven discrepancies between physics-based simulation and measured building energy use (i.e., residuals). Finally, the physics-informed ensemble learning is proposed to integrate physics-based and data-driven models for enhanced accuracy and robustness of building energy modeling. Results demonstrate 40–90% increase of accuracy between modeling and field observations compared to traditional physics-based modeling methods. Moreover, when the training dataset size is small, the proposed ensemble model overperforms the pure data-driven models, demonstrating its higher robustness in extrapolation scenarios. This work makes fundamental contributions to the development of convergent modeling approaches in the building modeling field.

Towards end-to-end structure determination from x-ray diffraction data using deep learning

Powder crystallography is the experimental science of determining the structure of molecules provided in crystalline-powder form, by analyzing their x-ray diffraction (XRD) patterns. Since many materials are readily available as crystalline powder, powder crystallography is of growing usefulness to many fields. However, powder crystallography does not have an analytically known solution, and therefore the structural inference typically involves a laborious process of iterative design, structural refinement, and domain knowledge of skilled experts. A key obstacle to fully automating the inference process computationally has been formulating the problem in an end-to-end quantitative form that is suitable for machine learning, while capturing the ambiguities around molecule orientation, symmetries, and reconstruction resolution. Here we present an ML approach for structure determination from powder diffraction data. It works by estimating the electron density in a unit cell using a variational coordinate-based deep neural network. We demonstrate the approach on computed powder x-ray diffraction (PXRD), along with partial chemical composition information, as input. When evaluated on theoretically simulated data for the cubic and trigonal crystal systems, the system achieves up to 93.4% average similarity (as measured by structural similarity index) with the ground truth on unseen materials, both with known and partially-known chemical composition information, showing great promise for successful structure solution even from degraded and incomplete input data. The approach does not presuppose a crystalline structure and the approach are readily extended to other situations such as nanomaterials and textured samples, paving the way to reconstruction of yet unresolved nanostructures.

From code sharing to sharing of implementations: Advancing reproducible AI development for medical imaging through federated testing

BackgroundThe reproducibility crisis in AI research remains a significant concern. While code sharing has been acknowledged as a step toward addressing this issue, our focus extends beyond this paradigm. In this work, we explore “federated testing” as an avenue for advancing reproducible AI research and development especially in medical imaging. Unlike federated learning, where a model is developed and refined on data from different centers, federated testing involves models developed by one team being deployed and evaluated by others, addressing reproducibility across various implementations. MethodsOur study follows an exploratory design aimed at systematically evaluating the sources of discrepancies in shared model execution for medical imaging and outputs on the same input data, independent of generalizability analysis. We distributed the same model code to multiple independent centers, monitoring execution in different runtime environments while considering various real-world scenarios for pre- and post-processing steps. We analyzed deployment infrastructure by comparing the impact of different computational resources (GPU vs. CPU) on model performance. To assess federated testing in AI models for medical imaging, we performed a comparative evaluation across different centers, each with distinct pre- and post-processing steps and deployment environments, specifically targeting AI-driven positron emission tomography (PET) imaging segmentation. More specifically, we studied federated testing for an AI-based model for surrogate total metabolic tumor volume (sTMTV) segmentation in PET imaging: the AI algorithm, trained on maximum intensity projection (MIP) data, segments lymphoma regions and estimates sTMTV. ResultsOur study reveals that relying solely on open-source code sharing does not guarantee reproducible results due to variations in code execution, runtime environments, and incomplete input specifications. Deploying the segmentation model on local and virtual GPUs compared to using Docker containers showed no effect on reproducibility. However, significant sources of variability were found in data preparation and pre-/post- processing techniques for PET imaging. These findings underscore the limitations of code sharing alone in achieving consistent and accurate results in federated testing. ConclusionAchieving consistently precise results in federated testing requires more than just sharing models through open-source code. Comprehensive pipeline sharing, including pre- and post-processing steps, is essential. Cloud-based platforms that automate these processes can streamline AI model testing across diverse locations. Standardizing protocols and sharing complete pipelines can significantly enhance the robustness and reproducibility of AI models.

'Where have my patients gone?': A simulation study on real-world data processing in Clinical Data Warehouses

ObjectiveTo access Electronic Health Record (EHR) data, hospitals have implemented Clinical Data Warehouses (CDWs) using Extract Transform and Load (ETL) processes. While ETL performances are typically evaluated individually, our study examines the cumulative impact of ETLs on data availability. MethodsUsing a real multi-hospital CDW as a case study, we modeled EHR data processing from the software sources to the CDW's data store. We simulated a scenario where researchers aimed to reconstruct breast cancer care trajectories using EHR data. We calculated the size and characteristics of the data store population, and compared them to the original population. ResultsEHR data are recorded in various software depending on data category, hospital, and year, each requiring specific series of ETLs for integration in the CDW. Despite acceptable transfer rates for each ETL (range 73 %-100 %), cumulative losses led to study populations in the data store being up to 90 % smaller than anticipated when researchers required data exhaustivity for patients. Population size decreased steeply with the more data categories required. No difference was found in population characteristics between the data store and the original cohorts. Discussion & ConclusionResearchers should scrutinize data availability in CDWs as missing data could result from outsourced care, incomplete input, or underperforming ETLs. Integrating more data sources in CDWs increases the number of data routes, necessitating time for ETL implementation and maintenance, and increases data loss risks. Though commonly perceived as a “black box”, data transformation can significantly influence the reliability of populations studied in CDWs. Public interest SummaryTo access data generated during care, researchers build Clinical Data Warehouses (CDWs). CDWs are infrastructures composed of a series of processing steps to extract the data from the data source, transform it according to the needs and load it into a data store. Usually, the performances of these processing steps are evaluated one a time. However, each data point goes through a series of processing steps before being made available for research. In this study, we aim to evaluate the impact of the entire data processing pipeline on the availability of data points in a CDW by simulating a study on breast cancer and evaluating the impact on the size and the characteristics of the final cohort. The cumulative losses of the processing steps resulted in a population 90 % smaller than anticipated. The characteristics of the final population showed no difference to those of the original cohort.

CompleteDT: Point cloud completion with information-perception transformers

In this work, we propose a novel point cloud completion network, called CompleteDT. To fully capture the 3D geometric structure of point clouds, we introduce an Information-Perception Transformer (IPT) that can simultaneously capture local features and global geometric relations. CompleteDT comprises a Feature Encoder, Query Generator, and Query Decoder. Feature Encoder extracts local features from multi-resolution point clouds to capture intricate geometrical structures. Query Generator uses the proposed IPT, utilizing the Point Local Attention (PLA) and Point Global Attention (PGA) modules, to learn local features and global correlations, and generate query features that represent predicted point clouds. The PLA captures local information within local points by adaptively measuring weights of neighboring points, while PGA adapts multi-head self-attention by transforming it into a layer-by-layer form where each head learns global features in a high-dimensional space of different dimensions. By dense connections, the module allows for direct information exchange between each head and facilitates the capture of long global correlations. By combining the strengths of both PLA and PGA, the IPT can fully leverage local and global features to facilitate CompleteDT to complete point clouds. Lastly, the query features undergo refining to generate a complete point cloud through the Query Decoder. Our experimental results demonstrate that CompleteDT outperforms current state-of-the-art methods, effectively learning from incomplete inputs and predicting complete outputs.

Cardiac Disease Diagnosis Using K-Nearest Neighbor Algorithm: A Study on Heart Failure Clinical Records Dataset

This article introduces an approach to diagnose heart diseases utilizing the K-Nearest Neighbor algorithm and diverse correlation filters for selecting the most pertinent attributes. Results high- light that meticulous filter selection enhances survival predictions in patients with heart diseases. Employing K = 5 and correlation filter CF = 0.1, key attributes for classification were identified as anemia, high blood pressure, serum creatinine, and sex. Omitting the 'time' attribute led to information loss but was crucial to prevent biases and generalize predictions across various clinical scenarios. Utilizing these classification parameters, we designed an Android mobile application called “Heart Info System”, functioning as an artificial intelligence service. It employs the K-Nearest Neighbor algorithm with optimal parameters to evaluate the probability of survival in the progression of heart disease. The main activity of the application retrieves data from a Firebase database. While the study results show promise, the accuracy of the application may be influenced by inaccurate or incomplete input data. Nevertheless, this application has the potential to improve the early detection of heart diseases, paving the way for life-saving interventions.

Modeling via peridynamics for crack propagation in laminated glass under fire

Despite laminated glass is widely used in building construction, its susceptibility to cracking and collapsing during an accidental fire can intensify fire propagation, leading to substantial casualties and economic losses. In this work, we developed a meshfree framework for modeling thermomechanical cracking of laminated glass via peridynamic (PD) theory. First, to overcome the incomplete input experimental data for implementing thermal boundary conditions, we proposed both the uniform and non-uniform temperature field strategies. Second, PD interface model and the stochastic distribution of defects were implemented within this framework. Third, the challenging problem of crack path selection in thermally loaded borosilicate/steel bi-material beams was studied to validate the feasibility of the proposed model. It is demonstrated that the proposed framework can capture the temperature history and crack features of laminated glass under fire, showing great consistency with experimental results. Our findings show that with the decrease of the glass strength or the increase of the defect content, more islands form in the central region of the glazing, implying higher risks of glass fall-out in fire conditions. This model constitutes a promising tool for predicting the fire performance of laminated glass and contributes to designing safer engineering structures.

Evaluation of the Suitability of Using Artificial Neural Networks in Assessing the Effectiveness of Greywater Heat Exchangers

The use of greywater heat exchangers (GHEs) is an effective way to reduce energy consumption for heating domestic water. However, the available characteristics of this type of device are often insufficient and consider only a few selected parameters of water and greywater, which results in the need to look for tools enabling the determination of the effectiveness of GHEs in various operating conditions with incomplete input data. The aim of this paper was to determine the usefulness of artificial neural networks (ANNs). For this purpose, comprehensive experimental tests were carried out on the effectiveness of the horizontal heat exchanger, taking into account a wide range of water and greywater flow rates and temperatures of these media, as well as the linear bottom slope of the unit, which allowed for the creation of a database of 32,175 results. Then, the feasibility of implementing the full research plan was assessed using ANNs. The analysis showed that the impact of the media temperatures on the heat exchanger effectiveness values obtained using ANNs is limited, which makes it possible to significantly reduce the number of necessary experiments. Adopting only three temperature values of at least one medium allowed the generation of ANN models with coefficient values R2 = 0.748–0.999 and RMSE = 0.077–1.872. In the case of the tested GHE, the slope and the flow rate of the mixed water are of key importance. However, even in the case of parameters of significant importance, it is possible to reduce the research plan without compromising the final results. Assuming five different values for each of the four input parameters (a total of 625 combinations) made it possible to generate an ANN model (R2 = 0.993 and RMSE = 0.311) with high generalization ability on the full research plan covering 32,175 cases. Therefore, the conducted analysis confirmed the usefulness of ANNs in assessing the effectiveness of GHEs in various operating conditions. The approach described in this paper is important for both environmental and economic reasons, as it allows for reducing the consumption of water and energy, which are necessary to carry out such scientific research.

Gradient-Guided Modality Decoupling for Missing-Modality Robustness

Multimodal learning with incomplete input data (missing modality) is very practical and challenging. In this work, we conduct an in-depth analysis of this challenge and find that modality dominance has a significant negative impact on the model training, greatly degrading the missing modality performance. Motivated by Grad-CAM, we introduce a novel indicator, gradients, to monitor and reduce modality dominance which widely exists in the missing-modality scenario. In aid of this indicator, we present a novel Gradient-guided Modality Decoupling (GMD) method to decouple the dependency on dominating modalities. Specifically, GMD removes the conflicted gradient components from different modalities to achieve this decoupling, significantly improving the performance. In addition, to flexibly handle modal-incomplete data, we design a parameter-efficient Dynamic Sharing (DS) framework which can adaptively switch on/off the network parameters based on whether one modality is available. We conduct extensive experiments on three popular multimodal benchmarks, including BraTS 2018 for medical segmentation, CMU-MOSI, and CMU-MOSEI for sentiment analysis. The results show that our method can significantly outperform the competitors, showing the effectiveness of the proposed solutions. Our code is released here: https://github.com/HaoWang420/Gradient-guided-Modality-Decoupling.

Amplidiff: an optimized amplicon sequencing approach to estimating lineage abundances in viral metagenomes

BackgroundMetagenomic profiling algorithms commonly rely on genomic differences between lineages, strains, or species to infer the relative abundances of sequences present in a sample. This observation plays an important role in the analysis of diverse microbial communities, where targeted sequencing of 16S and 18S rRNA, both well-known hypervariable genomic regions, have led to insights into microbial diversity and the discovery of novel organisms. However, the variable nature of discriminatory regions can also act as a double-edged sword, as the sought-after variability can make it difficult to design primers for their amplification through PCR. Moreover, the most variable regions are not necessarily the most informative regions for the purpose of differentiation; one should focus on regions that maximize the number of lineages that can be distinguished.ResultsHere we present AmpliDiff, a computational tool that simultaneously finds highly discriminatory genomic regions in viral genomes of a single species, as well as primers allowing for the amplification of these regions. We show that regions and primers found by AmpliDiff can be used to accurately estimate relative abundances of SARS-CoV-2 lineages, for example in wastewater sequencing data. We obtain errors that are comparable with using whole genome information to estimate relative abundances. Furthermore, our results show that AmpliDiff is robust against incomplete input data and that primers designed by AmpliDiff also bind to genomes sampled months after the primers were selected.ConclusionsWith AmpliDiff we provide an effective, cost-efficient alternative to whole genome sequencing for estimating lineage abundances in viral metagenomes.

Surrogate model‐aided global sensitivity analysis framework for seismic consequences estimation in buildings

AbstractSeismic consequences estimation for individual buildings is valuable for various stakeholders, including government entities, building owners, and insurers. The robustness of estimation results in the presence of incomplete input information can typically be investigated through sensitivity analysis. However, the estimation process's complexity and the sensitivity analysis's computational burden hinder its practical application, which requires a more efficient procedure to facilitate broader use. This paper proposes a novel framework for sensitivity analysis of seismic consequences estimation to improve the efficiency and reliability of such analysis. The proposed approach encompasses three key components: (1) stochastic ground motion modeling (SGMM)‐based seismic consequences estimation to evaluate the economic, environmental, and social consequences given specific buildings by considering different hazard levels, (2) the training of surrogate model (Gaussian process model) for structural analysis to reduce the computational cost of the evaluation process, and (3) variance‐based global sensitivity analysis to investigate the importance of parameters of concern in the estimation process. The entire procedure is implemented in Python, adhering to object‐oriented programming, and does not rely on external software. Then, the proposed methodology is applied to two distinct three‐story steel moment‐resistant frames (SMRFs) subjected to four different hazard levels to demonstrate its effectiveness. The SGMM method can generate specific ground motions for each hazard level, mitigating the potential for result bias from using ground motions with unrealistic characteristics. Furthermore, the SGMM method is particularly suitable for automated analysis processes reducing the laborious task of screening ground motions from the database. Comparative analysis with surrogate‐free estimation reveals that the surrogate‐based analysis delivers reliable results with significantly reduced computational cost. The results of analyzing different structures under varying hazard levels reflect the variability in sensitivity analysis of consequences estimation, highlighting the necessity of the proposed flexible and efficient framework. Furthermore, the proposed framework's advantages, limitations, and future research needs are discussed.

Relevance vector machine with hybrid kernel-based soft sensor via data augmentation for incomplete output data in sintering process

A ratio of CO and CO2 (CO/CO2) is a key indicator of sintering carbon consumption, which is difficult to be determined in real-time. Therefore, the establishment of its soft sensing model is of great practical significance. This paper proposes a novel CO/CO2 soft sensing model with incomplete output data based on relevance vector machine with hybrid kernel via data augmentation. First, a least absolute shrinkage and selection operator is employed for determining key input variables of the model, and an automatic fuzzy clustering framework is used to automatically identify multiple operating modes. Then, a relevance vector machine with hybrid kernel method is presented to model each operating mode separately. Meanwhile, considering the problem of incomplete input and output data, data augmentation is applied in modeling to enhance the model performance. Finally, the soft sensing model of CO/CO2 is formed. Experimental results and analyses using actual production data coming from the sintering production process demonstrate that the prediction performance and accuracy of the proposed model outperform some existing algorithms.

Data-based bipartite formation control for multi-agent systems with communication constraints.

This article investigates data-driven distributed bipartite formation issues for discrete-time multi-agent systems with communication constraints. We propose a quantized data-driven distributed bipartite formation control approach based on the plant's quantized and saturated information. Moreover, compared with existing results, we consider both the fixed and switching topologies of multi-agent systems with the cooperative-competitive interactions. We establish a time-varying linear data model for each agent by utilizing the dynamic linearization method. Then, using the incomplete input and output data of each agent and its neighbors, we construct the proposed quantized data-driven distributed bipartite formation control scheme without employing any dynamics information of multi-agent systems. We strictly prove the convergence of the proposed algorithm, where the proposed approach can ensure that the bipartite formation tracking errors converge to the origin, even though the communication topology of multi-agent systems is time-varying switching. Finally, simulation and hardware tests demonstrate the effectiveness of the proposed scheme.

Multi-Agent Safe Graph Reinforcement Learning for PV Inverters-Based Real-Time Decentralized Volt/Var Control in Zoned Distribution Networks

To realize real-time voltage/var control (VVC) in active distribution networks (ADNs), this paper proposes a new multi-agent safe graph reinforcement learning method to optimize reactive power output from PV inverters. The network is divided into several zones, and a decentralized framework is proposed for coordinated control of reactive power output in each zone to regulate voltage profiles and minimize network energy loss. The VVC problem is formulated as a multi-agent decentralized partially observable constrained Markov decision process. Each zone has a central control agent that embeds graph convolution networks (GCNs) in the policy network to improve the decision-making capability. The GCN extracts graph-structured features from the ADN topology, reflecting the relationship between VVC and grid topology, and can filter noise and impute missing data. The training process includes primal-dual policy optimization to rigorously satisfy voltage safety constraints. Simulations on a 141-bus distribution system demonstrate that the proposed method can effectively minimize network energy loss and reduce voltage deviations, even in the presence of noisy or incomplete input measurements.

Cross-domain point cloud completion for multi-class indoor incomplete objects via class-conditional GAN inversion

Indoor point clouds from real-world scans are often incomplete and sparse due to limited observation views and severe occlusion between objects. Point cloud completion can restore missing parts to ensure data integrity, so that subsequent tasks like point cloud classification, segmentation, object detection and three-dimensional (3D) reconstruction can obtain more accurate and complete geometric and semantic information. Existing point cloud completion methods focus more on synthetic data or require separate modeling on each class of objects, which struggle to meet the inherent requirements of the completion task for real-world indoor scenes containing multi-class objects with significant pose variations. In this research, we present an efficient class-conditional generative adversarial network (GAN) inversion framework for the task of recovering incomplete objects in real-world indoor scenes, in which we first train a class-conditional GAN to learn rich shape priors for multi-class objects from synthetic data domain and then find an optimal latent code in GAN’s latent space that gives a complete shape that best recovers the given incomplete input from real-world data domain. The performance of our proposed method is evaluated on both real-world and synthetic datasets. In comparison to ShapeInversion (Zhang et al., 2021), our method demonstrates an improvement in F-score by 3.7 and 2.0 percentage points while reducing the chamfer distance (CD) by 4.15 and 1.1 on ScanNet dataset and CRN dataset, respectively. The developed code will be publicly available at: https://github.com/CNU-DLandCV-lab/cgan-inversion.

Leveraging electronic device outputs for service quality assessment of public transportation systems: An initial look at the Philippines’ service contracting program through data analytic lens

Electronic devices such as CCTVs and GPS trackers are becoming more prevalent in public utility vehicles. These electronic devices produce data that can be aggregated and processed to generate insights regarding the service quality provided by public transportation systems. We describe a methodology for processing CCTV-produced videos and GPS logs, and utilize it to perform a first order, limited assessment of the Philippines’ Service Contracting Program, which ran during the COVID-19 pandemic. Our analyses uncovered vast discrepancies between data-supported figures and manually-generated estimates provided by the transport operators, especially with regards to vehicle occupancy and total number of commuters served. Our study also uncovered imperfections in the data generated by the electronic devices, which place caveats or limitations on the insights generated by the data analyses. We provide and suggest ways on how the analyses can be modified to take into account such missing data, thus enabling us to come up with educated, reasonable insights despite incomplete input datasets.

The spatiotemporal effect of factor price distortion on capacity utilization in China’s iron and steel industry

Under the situation of global climate change and green transition, China’s iron & steel industry (ISI) faced great pressure to energy conservation and cut backward overcapacity. To explore the characteristics of overcapacity and inherent impacting factor of capacity utilization, the study focused on the relationship of incomplete input factor market and overcapacity. Based on provincial data from 2003 to 2019, the study estimated capacity utilization rate in China’s ISI sector, and further evaluated the impact of factor price distortion on capacity utilization rate and its spatiotemporal effect through dynamic spatial model. Firstly, the capacity utilization rate China’s ISI sector showed regional heterogeneity with an average level of 68%, and the capacity utilization rate would be promoted by previous capacity utilization rate and capacity utilization rate of neighbor provinces; Secondly, capital and energy price distortion showed significant negative effect on capacity utilization, while labor price distortion positively promoted capacity utilization; Thirdly, from the perspective of spatial effect, capital price distortion showed significantly negative effect on capacity utilization. Therefore, deepening factor market reform and reducing the intervention of local government would be effective way to improve capacity utilization and promote cleaning production.

Plant disease prescription recommendation based on electronic medical records and sentence embedding retrieval

BackgroundIn the era of Agri 4.0 and the popularity of Plantwise systems, the availability of Plant Electronic Medical Records has provided opportunities to extract valuable disease information and treatment knowledge. However, developing an effective prescription recommendation method based on these records presents unique challenges, such as inadequate labeling data, lack of structural and linguistic specifications, incorporation of new prescriptions, and consideration of multiple factors in practical situations.ResultsThis study proposes a plant disease prescription recommendation method called PRSER, which is based on sentence embedding retrieval. The semantic matching model is created using a pre-trained language model and a sentence embedding method with contrast learning ideas, and the constructed prescription reference database is retrieved for optimal prescription recommendations. A multi-vegetable disease dataset and a multi-fruit disease dataset are constructed to compare three pre-trained language models, four pooling types, and two loss functions. The PRSER model achieves the best semantic matching performance by combining MacBERT, CoSENT, and CLS pooling, resulting in a Pearson coefficient of 86.34% and a Spearman coefficient of 77.67%. The prescription recommendation capability of the model is also verified. PRSER performs well in closed-set testing with Top-1/Top-3/Top-5 accuracy of 88.20%/96.07%/97.70%; and slightly worse in open-set testing with Top-1/Top-3/Top-5 accuracy of 82.04%/91.50%/94.90%. Finally, a plant disease prescription recommendation system for mobile terminals is constructed and its generalization ability with incomplete inputs is verified. When only symptom information is available without environment and plant information, our model shows slightly lower accuracy with Top-1/Top-3/Top-5 accuracy of 75.24%/88.35%/91.99% in closed-set testing and Top-1/Top-3/Top-5 accuracy of 75.08%/87.54%/89.84% in open-set testing.ConclusionsThe experiments validate the effectiveness and generalization ability of the proposed approach for recommending plant disease prescriptions. This research has significant potential to facilitate the implementation of artificial intelligence in plant disease treatment, addressing the needs of farmers and advancing scientific plant disease management.