Multi-source Model Research Articles

Prediction of therapy response of breast cancer patients with machine learning based on clinical data and imaging data derived from breast [18F]FDG-PET/MRI.

To evaluate if a machine learning prediction model based on clinical and easily assessable imaging features derived from baseline breast [18F]FDG-PET/MRI staging can predict pathologic complete response (pCR) in patients with newly diagnosed breast cancer prior to neoadjuvant system therapy (NAST). Altogether 143 women with newly diagnosed breast cancer (54 ± 12years) were retrospectively enrolled. All women underwent a breast [18F]FDG-PET/MRI, a histopathological workup of their breast cancer lesions and evaluation of clinical data. Fifty-six features derived from positron emission tomography (PET), magnetic resonance imaging (MRI), sociodemographic / anthropometric, histopathologic as well as clinical data were generated and used as input for an extreme Gradient Boosting model (XGBoost) to predict pCR. The model was evaluated in a five-fold nested-cross-validation incorporating independent hyper-parameter tuning within the inner loops to reduce the risk of overoptimistic estimations. Diagnostic model-performance was assessed by determining the area under the curve of the receiver operating characteristics curve (ROC-AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy. Furthermore, feature importances of the XGBoost model were evaluated to assess which features contributed most to distinguish between pCR and non-pCR. Nested-cross-validation yielded a mean ROC-AUC of 80.4 ± 6.0% for prediction of pCR. Mean sensitivity, specificity, PPV, and NPV of 54.5 ± 21.3%, 83.6 ± 4.2%, 63.6 ± 8.5%, and 77.6 ± 8.1% could be achieved. Histopathological data were the most important features for classification of the XGBoost model followed by PET, MRI, and sociodemographic/anthropometric features. The evaluated multi-source XGBoost model shows promising results for reliably predicting pathological complete response in breast cancer patients prior to NAST. However, yielded performance is yet insufficient to be implemented in the clinical decision-making process.

MsFfTsGP: Multi-source features-fused two-stage grade prediction of zinc tailings in lead-zinc flotation process via multi-stream 3D convolution with attention mechanism

The tailings grade is a crucial indicator for controlling and optimizing mineral flotation processes. However, detecting/predicting tailings grade online faces challenges due to intricate physicochemical reactions in three-phase slurries with the strongly-coupled characteristics of the technical process. This paper introduces a novel multi-source features-fused two-stage tailings grade prediction model called MsFfTsGP by integrating multiple-source froth image sequences (FISs) with available traditional technical variables (TTVs, e.g., the reagent dosage and measured grades) from multiple flotation cells, mimicking the on-site two-stage prediction mode of experienced operators. Namely, an initial evaluation is made based on the source-ore characteristics using the FISs of the first rougher, while the second stage provides a sophisticated prediction of the tailing grade by considering both the source-ore characteristics and the flotation behavior of the whole circuit process. Specifically, FISs captured from the first rougher are fed into a multi-stream 3D convolutional network (Ms3DCN) for making an initial prediction. Then, the initial prediction result is combined with the FIS information from the last scavenger (also processed by the Ms3DCN) and available TTVs based on the attention mechanism to make a sophisticated prediction. Extensive experiments on a real industrial lead-zinc flotation process demonstrated a significant improvement of the proposed model in prediction accuracies. Integrating multisource FISs with TTVs from multiple coupled flotation cells shows great potential for achieving accurate tailings grade prediction, making MsFfTsGP a promising tool for enhancing the control and optimization of flotation processes.

Improved O3 predictions in China by combining chemical transport model and multi-source data with machining learning techniques

Accurate ozone (O3) predictions is crucial for assessing its impact on public health and developing effective prevention and control measures. While ground-based observations are considered highly accurate, the limited and uneven spatial coverage poses challenges. Air quality models can provide complete spatiotemporal coverage but are subject to biases due to simplifications in physicochemical mechanisms and uncertainties associated with emission and meteorology inputs. In this study, we aimed to improve the accuracy of the predicted O3 concentrations from the Community Multiscale Air Quality (CMAQ) model by using the machine learning techniques. First, we compared three machine learning algorithms, namely Light Gradient Boosting Machine (LightGBM), Random Forest, and eXtreme Gradient Boosting. Results showed that LightGBM exhibited the highest correlation coefficient (R) of 0.84, making it the preferred algorithm for further analysis. Subsequently, two multi-source data prediction models based on LightGBM were constructed to improve the accuracy of predicted daily maximum 8-h O3 (O3-Max8h). The first model, referred to as LGBR, utilized the predicted air pollutant concentrations and meteorological fields as input variables, while the second model, named LGBR-CHAP, incorporated ChinaHighAirPollutants (CHAP) O3-Max8h as an additional input variable. Validation results demonstrated significant improvements in the LGBR and LGBR-CHAP models compared to the original CMAQ model. On the daily scale, the root mean square error and mean bias of the predicted O3-Max8h decreased by 3.15 and 2.07 μg/m3, respectively, for the LGBR model, while they decreased by 5.61 and 4.18 μg/m3, respectively, for the LGBR-CHAP model. On the monthly scale, the R of the original CMAQ model varied from 0.2 in the March to 0.91 in June, and LGBR improved all months (0.4–0.92) as did LGBR-CHAP (0.5–0.94). Spatially, the O3-Max8h simulated by the original CMAQ model is better in eastern China but less skillful in western China. After optimization by the LGBR and LGBR-CHAP models, the national average R improved from 0.77 to 0.83 and 0.88, respectively. The LGBR-CHAP model exhibited better predictive capacity than the LGBR model and was subsequently employed to generate high-resolution (10 km × 10 km) and full-coverage (100%) O3-Max8h data. This dataset will prove valuable for future air pollution and health studies.

Privacy-Preserving Cross-Area Traffic Forecasting in ITS: A Transferable Spatial-Temporal Graph Neural Network Approach

Traffic forecasting is essential in improving and maintaining safety and orderliness in intelligent transportation systems (ITS). As a deep learning approach, graph neural networks (GNN) based spatial-temporal association mining methods are promising in traffic forecasting. However, current GNN-based methods usually require a high number of training data, and when the sample volume is small, the performance of the model drops dramatically. The existing transfer methods can solve this problem by leveraging knowledge from other data-rich areas, but the domain adaption method with access to source data still faces the non-neglectable problem of private information leakage in the source area. A solution that can solve cross-area transfer without access to source data is still missing. In this paper, to fill the gap, we propose a Transferable Federated Inductive Spatial-Temporal Graph Neural Network (T-ISTGNN) framework to transfer spatial-temporal dependency information in cross-area data to accomplish traffic state forecasting. First, we introduce a multi-source model aggregation scheme based on federated learning to retain the traffic information of the source areas. Second, we propose a transfer method between source and target areas based on hypothesis transfer learning to achieve domain adaption under source domain data protection. Third, we propose a GNN-based method called Inductive Spatial-Temporal Graph Neural Network (ISTGNN) for traffic forecasting. Experiments on real-world datasets demonstrate that T-ISTGNN is capable of cross-area traffic state forecasting under the restriction of preserving the privacy of source areas.

Resilience-oriented comparative study of SOP-based service restoration in distribution systems

Soft open points (SOPs) can perform load shifting and voltage improvement, and are critical to improving distribution service restoration (DSR) during power outages. Although multiple DSR strategies utilizing local resources have been reported, their effectiveness has not been sufficiently studied. To solve this problem and utilize the flexibility of SOP, a comparative study of different strategies for SOP-based DSR is presented in this paper. First, various SOP-based DSR models, including single-source, single-island model (SSM), islanding partition model (IPM) and multi-source collaborative model (MCM), are proposed and formulated as mixed-integer nonlinear programming (MINLP) problems, incorporating static voltage characteristics of loads. Second, to enhance the solution efficiency, mixed-integer second order cone programming is introduced to convert the MINLP problems into easily solvable convex optimization problems. A restorative index is defined to evaluate the effectiveness of various SOP-based DSR strategies. Finally, a comparative simulation analysis of the proposed models on the IEEE 33-bus system illustrates that SOP-based MCM achieves an 94 % power output for distributed energy resources, while exhibiting the lowest RI for the total loads, standing at just 9.5 %. Furthermore, the RI for critical loads consistently remains at 0, demonstrating the superiority of SOP-based MCM in comparison to other strategies.

Multi-source and multimodal data fusion for improved management of a wastewater treatment plant

Wastewater treatment plants (WWTPs) are complex systems presenting stochastic, non-linear, and non-stationary behavior, which makes their operational management very challenging. In this context, data collected from distributed sources across the plant play a central role in the optimized operation and control of WWTPs. However, even when available, the use of collected data is far from trivial due to the coexistence of asynchronous measurements, data with different granularity, measurements of different quality (precision, accuracy), multimodal sources (sensors, spectra, images, hyphenated instrumentation), among other aspects related to the data life cycle. Such heterogeneity in process data characteristics hinders the application of most off-the-shelf data analytics methods. Flexible solutions able to cope with the complexity of systems and of the data they generate are therefore necessary to overcome these limitations and enable an effective analysis and operation of WWTPs. In this article, data-fusion approaches for handling multiple heterogeneous sources of process data are developed and comparatively tested. Priority is given to solutions that can flexibly be adapted to different specific operational contexts. The methodologies are tested on an industrial case study (WWTP), where the concentration of a toxin in the effluent stream is to be predicted from available heterogeneous data. Single- and multi-source modeling approaches are contemplated and a nested cross-validation method was developed to handle the time-series nature of the models. Bayesian fusion synergistically combines data from different sources considering their uncertainty, standing out among the methodologies tested as offering a good balance in terms of accuracy (RMSEP = 1.34), stability (prequential IQR = 0.034), and flexibility (to accommodate missing and new sources).

Water sources for street trees in mesic urban environments

Street trees support climate resiliency through a variety of pathways, such as offsetting urban heat and attenuating storm water runoff. While urban trees in arid and semiarid ecosystems have been shown to take up water from irrigation, it is unknown where street trees in mesic cities obtain their water. In this study, we use natural abundance stable isotopes to estimate the proportional sources of water taken up by Acer platanoides street trees in Boston, Massachusetts, United States, including precipitation, irrigation, groundwater, and wastewater. We use Bayesian multisource mixing models to estimate water sources by comparing the natural abundance isotopic ratios of hydrogen and oxygen across potential water sources with water extracted from tree stem samples. We find that during the summer of 2021, characterized by anomalously high rainfall, street trees predominantly utilized water from precipitation. Precipitation accounted for 72.3 % of water extracted from trees sampled in August and 65.6 % from trees sampled in September. Of the precipitation taken up by street trees, most water was traced back to large storm events in July, with July rainfall alone accounting for up to 84.0 % of water found within street trees. We find strong relationships between canopy cover fractions and the proportion of precipitation lost to evapotranspiration across the study domain, supporting the conclusion that tree planting initiatives result in storm water mitigation benefits due to utilization of water from precipitation by urban vegetation. However, while the mature trees studied here currently support their water demand from precipitation, the dependency of street trees on precipitation in mesic cities may lead to increased water stress in a changing climate characterized by a higher frequency and severity of drought.

A model of integrating college students’ consumerism education and ideological and political education by applying multivariate data fusion

Abstract This paper explores the relationship between college students’ consumption views and their consumption views under Civic Education in the context of multiple datasets. Secondly, a multi-agent, multi-source heterogeneous data collection model is used to collect data on consumption view and Civic Education. A CNN model is designed to study the correlation between consumption view and Civic Education. The CNN and LSTM multi-grain data classification models are used to fuse the features of consumer outlook and Civic Education, respectively. Then, build an online learning platform for college students’ consumerism. Finally, the platform’s fusion effect and learning results for college students were analyzed and studied. The average JS index of data fusion was 93.32%, and the fusion effect was good. The total score of students’ pre and post-test consumerism scores increased from 86 to 100, and the effect of college students’ consumerism education was positive.

Intelligent Perception of Distribution Network State Based on Multi-source Data Fusion

This article proposes an intelligent state perception method based on multi-source data fusion to address the issue of the difference between the state perception range and the actual range caused by the abnormal operation status of active distribution networks. In this paper, by combining Ant colony optimization algorithms and fusion rules, a multi-source data fusion model of the operation situation is constructed. The operation situation indicators, such as line loss rate and light load rate, are determined. Using multi-source data fusion information entropy to train samples, the range of abnormal operating states of the power grid is determined. By fusing multi-source voltage data and using the multi-source data fusion method to distinguish voltage data, four operating state perception results of voltage are obtained. Finally, the perception data is modified using Analytic Hierarchy Process to ensure that the state perception results meet the compatibility and consistency requirements with the actual results. According to the experimental results, the proposed method can accurately perceive the fluctuation of abnormal data within 40~50 ranges, which is consistent with the actual value fluctuation range, so as to obtain accurate perception results under different operating conditions.

Application and evaluation of CNN-LSTM classification regression based multi-source precipitation data fusion model in water resources research in Qinghai Province

Qinghai Province is located in the Qinghai-Tibet Plateau region, with complex and diverse topography and sparse precipitation stations, which makes it difficult to obtain reliable precipitation data. This study proposes a classification and regression model based on a deep learning algorithm, which combines a convolutional neural network (CNN) and a long short-term memory neural network (LSTM), with the CNN extracting the spatial features of multi-source data, the LSTM capturing their temporal dependencies. The regression results are used to determine whether rainfall is occurring and to further calibrate the non-rainfall component of the precipitation forecast results. ERA5, IMERG, CHIRPS and DEM were selected as feature data and rain gauge data as label data. The findings indicate that the proposed CNN-LSTM classification regression model (CLCR) is superior to other models (CNN, CNN-LSTM, LSTM). The Kling-Gupta efficiency (KGE) of the data fused using CLCR was 0.66, which was significantly better than that of the raw rainfall data (0.53, -0.36, 0.34) and other models (0.58, 0.65, 0.63). CLCR also showed more performance in daily precipitation detection than other models and raw precipitation data, with Critical Success Index (CSI), Probability of Detection (POD), and False Alarm Ratio (FAR) 0.61, 0.25 and 0.76 respectively. This study generated a high-precision daily rainfall dataset with a precision of 0.01° resolution for 2013-2017 in Qinghai Province, which provides reliable data support for hydrological studies in Qinghai Province.

Magnetic inversion approach for modeling data acquired across faults: various environmental cases studies

An effective extension to the particle swarm optimizer scheme has been developed to visualize and modelize robustly magnetic data acquired across vertical or dipping faults. This method can be applied to magnetic data sets that support various investigations, including mining, fault hazards assessment, and hydrocarbon exploration. The inversion algorithm is established depending on the second horizontal derivative technique and the particle swarm optimizer algorithm and was utilized for multi-source models. Herein, the inversion method is applied to three synthetic models (a dipping fault model contaminated without and with different Gaussian noises levels, a dipping fault model affected by regional anomaly, and a multi-source model) and three real datasets from India, Australia, and Egypt, respectively. The output models confirm the inversion approach’s accuracy, applicability, and efficacy. Also, the results obtained from the suggested approach have been correlated with those from other methods published in the literature.

Conversion and fusion method of multi-source and different populations maintainability prior data

Maintainability is an important universal quality characteristic that reflects the convenience, speed and economy of weapon and equipment maintenance. Making full use of multi-source data to accurately verify the degree to which the developed equipment meets the maintainability requirements is an important basis for equipment identification and acceptance. To solve the low reliability of equipment maintainability verification results caused by inaccurate comprehensive prior distribution obtained by fusing multi-source and different populations' prior data, a method of data conversion and fusion is proposed. A data conversion model based on the mean value ratio of failure mode maintenance data is constructed. The conversion factor is defined according to objective data to convert the different populations' prior data to the same populations. Next, a comparison of the prior distribution fitting performance of Bayes bootstrap, bootstrap, and two improved sample-resampling methods to are used obtain the closest fitting distribution to the true distribution. By constructing a multi-source data fusion model based on improved KL divergence, a symmetrical KL divergence is constructed to describe the similarity between each prior distribution and the field distribution for the weighted fusion of multi-source prior distribution in addition to determining and testing the normal comprehensive prior distribution. The results show that the conversion and fusion method effectively converts the multi-source and different populations’ maintainability prior data and obtains an accurate, comprehensive prior distribution by fusion, laying the foundation for applying the Bayes test method to verify the quantitative index of equipment maintainability.

Combining GEDI and sentinel data to estimate forest canopy mean height and aboveground biomass

Forest canopy mean height (CMH) and aboveground biomass (AGB) are key indicators for evaluating forest ecosystem productivity. In this study, we proposed a new approach to integrate field measurement data, GEDI LiDAR, sentinel, and terrain data to construct multi-source data-driven forest CMH and AGB models at a 30-m resolution. First, we employed the RFE-SVM (Recursive Feature Elimination- Support Vector Machine) method to determine the features sensitive to forest height and AGB. Second, we used three regression models to construct the CMH model to extend the GEDI point data to wall-to-wall CMH maps thereby providing sensitive features for AGB estimation. Third, we jointly selected the features and field measurement data to build a model to estimate AGB. The CMH and AGB models, evaluated within the study area, achieved R2 values of 0.64 and 0.89, respectively. Fourth, we performed transferability tests for the AGB model. The AGB model built based on data from study area was applied to three other test areas, resulting in R2 values of 0.66, 0.76, and 0.91, respectively. Overall, this study presented a method that utilizes extensive open data with great potential for mapping forest CMH and AGB over large areas.

Semi-supervised mixture multi-source exchangeability model for leveraging real-world data in clinical trials.

The traditional trial paradigm is often criticized as being slow, inefficient, and costly. Statistical approaches that leverage external trial data have emerged to make trials more efficient by augmenting the sample size. However, these approaches assume that external data are from previously conducted trials, leaving a rich source of untapped real-world data (RWD) that cannot yet be effectively leveraged. We propose a semi-supervised mixture (SS-MIX) multisource exchangeability model (MEM); a flexible, two-step Bayesian approach for incorporating RWD into randomized controlled trial analyses. The first step is a SS-MIX model on a modified propensity score and the second step is a MEM. The first step targets a representative subgroup of individuals from the trial population and the second step avoids borrowing when there are substantial differences in outcomes among the trial sample and the representative observational sample. When comparing the proposed approach to competing borrowing approaches in a simulation study, we find that our approach borrows efficiently when the trial and RWD are consistent, while mitigating bias when the trial and external data differ on either measured or unmeasured covariates. We illustrate the proposed approach with an application to a randomized controlled trial investigating intravenous hyperimmune immunoglobulin in hospitalized patients with influenza, while leveraging data from an external observational study to supplement a subgroup analysis by influenza subtype.

A study of robotic search strategy for multi-radiation sources in unknown environments

This paper focuses on the multiple radiation source search problems, where the mobile robot identifies the number and parameters of sources online while exploring an unknown environment. The radiation superposition and the limited observations improve the difficulty of estimation, and the exploration trajectory is also associated with estimation. A novel search strategy based on receding horizon planning is proposed, which includes the observation, estimation, and exploration modules. The observation module filters and records the radiation intensity for estimation. In the estimation module, an adaptive differential evolution algorithm is integrated into the peak suppression particle filter to avoid the local optimum. The multi-source radiation gain model is conceived to determine the observation position in the exploration module. The strategy trades off exploration of unknown areas and exploitation of known radiation fields. The results of simulations and experiments demonstrate that the proposed strategy can identify the parameters and quantities of all sources in multi-modal radiation fields. Furthermore, our strategy exhibits superior performance in searching for multi-radiation sources in unknown environments compared with the boustrophedon path and the Next-Best-View planner.

A multi-source data-driven model of lake water level based on variational modal decomposition and external factors with optimized bi-directional long short-term memory neural network

An accurate prediction of lake water levels is of great significance to water resource regulation, flood prevention and mitigation. However, water level fluctuations have been increasingly serious due to abnormal climate and extreme events. In view of this, a VMD-EF-OBILSTM model was constructed for lake water levels based on multiple sources of hydrological and meteorological variables. In this model, water level data are transformed into low-frequency internal and high-frequency external terms by variable modal decomposition (VMD), and they are combined with external factors (EF) for multivariate prediction. The optimized bi-directional long short-term memory (OBILSTM) invokes the attention mechanism and optimizes the model's hyperparameters by whale optimization algorithm (WOA). Ultimately, the predictions of each component are linearly combined to obtain the forecast values. The empirical results with water level data from Poyang Lake in China show that the multi-source deep learning model can achieve higher prediction accuracy and lower prediction uncertainty.

Joint Optimal Dispatch of Complex Urban Raw Water Supply: A Case Study of Lanxi City, Zhejiang Province, China

Water resources play a vital role in supporting urban economic and social development and ecological and environmental protection. Water shortage is a key factor that restricts the high-quality development of cities, while the coordinated and optimized allocation of urban raw water is an important measure to alleviate the water supply–demand imbalance. The current research on urban water supply issues and their underlying causes still needs to be further strengthened. Similarly, the formulation of rules for multi-source and cross-district water supply should pay more attention. This paper proposes a general analytical process consisting of four main stages: problem identification, system generalization, rule formulation, and model construction and solution for the joint optimal scheduling of raw water in a complex urban water supply system. This study investigates the significant water resource wastage and structural water shortage in the reservoirs of Lanxi City. The optimal scheduling plan is proposed by formulating rational rules for inter-district water supply and establishing a multi-source and multi-objective joint optimization scheduling model. Compared to the current independent scheduling scheme and multi-water source joint dispatching scheme based on the current dispatch diagram, the optimal scheduling plan effectively reduced the cumulative water shortage by 68.04 million m3 and 29.72 million m3, respectively, and increased the urban water supply guarantee rate in all districts of the city to over 90%. This study offers valuable insights to urban water resource managers, empowering them to develop optimal multiple water source supply rules that align to the specific characteristics of other case studies.

Multi-source heterogeneous data fusion model based on fuzzy mathematics

Sensors as the sensing end of intelligent control can be used to collect various data instead of human beings. In the context of technological development, the variety of sensors leads to multiple and structurally unequal data sources, and fusion of these data becomes a problem for consideration. The study constructs an intuitionistic fuzzy transformation method to handle data with various attributes with the help of fuzzy mathematical concepts, which characterizes the data based on the hesitancy and ideal solutions under Gaussian distribution. Simulations of classical classification data show that the intuitionistic fuzzy transformation method can effectively differentiate the affiliation of data points in the dataset, and the results of 800 simulations show that the qualitative accuracy of the algorithm can reach 89%, while the causes of abnormal data are explored and it is found that the attributes of the dataset based on Gaussian distribution are too close to each other as the cause of misclassification; the algorithm is also optimized from multi-dimensional considerations, and a An optimization operator based on the distance method of superior and inferior solutions was constructed and simulated for several optimization paths. The results show that the study uses an optimization scheme that is significantly better than the existing fuzzy operator, and 800 times can improve the accuracy rate up to 95.23%, which is 14.01% higher than that of a single attribute. This indicates that the intuitionistic fuzzy algorithm of this study has some rationality and is able to fuse the data of multiple attributes of the sensor for determination and provide the necessary basis for decision making.

Trajectory-Based Spatiotemporal Multi-Task Multi-Graph Network for Traffic State Prediction

As a critical application in intelligent transportation systems, traffic state prediction still faces various challenges, such as unsatisfactory capability of utilizing multi-source data and modeling spatiotemporal network relevancies. Therefore, we propose a trajectory-based multi-task multi-graph convolutional network (Tr-MTMGN), a novel spatiotemporal deep learning framework for traffic state prediction on a citywide scale. This method firstly mines the underlying information from vehicle trajectories and designs a multi-graph convolution block to investigate spatial correlations. Sequentially, the multi-head self-attention layer is integrated into the multi-task learning framework to capture the temporal dependencies of the traffic state. The proposed model was evaluated on field data collected in Zhangzhou, China, and demonstrated superior performance when compared with existing state-of-the-art baselines.

Multi-source information fusion-based dynamic model for online prediction of rate of penetration (ROP) in drilling process

Rate of penetration (ROP) accurate prediction can effectively reduce drilling cycle time and operation cost. However, the complexity and variability of drilling process is not conducive to accurate ROP modeling. The existing ROP models focus on static prediction considering drilling data, which cannot effectively solve the ROP mutation problem caused by complex formations. Some ROP online prediction studies have great real-time capture capability, but without considering the formation lithology adequately, and can be limited in practical application. In this paper, a multi-source information fusion-based dynamic model for online prediction of ROP in drilling process is proposed, which consists of three stages (drilling data pre-processing, optimized ROP modeling, and ROP model updating). In the first stage, three drilling parameters including weight on bit, rotational speed, and torque are filtered by using the outlier removing and wavelet filtering techniques. In the second stage, a novel ROP modeling method, named as hybrid bat algorithm optimized - restricted Boltzmann machine - back propagation neural network (HBAO-RBM-BPNN) is proposed to build the ROP prediction model. In the last stage, formation drillability variation and time interval are used as update conditions, and the established ROP model is updated via a moving window strategy to realize high accuracy online prediction of ROP. Multi-source (formation and drilling) information fusion is considered in this key stage. Finally, the proposed method and seven ROP prediction methods (two offline and five online) are compared through the use of industrial data from a drilling site in Dandong area, Northeast China. The comparison results verify the effectiveness of the proposed method, which lays a foundation for intelligent optimization control of complex geological drilling process.