Short-term Memory Neural Network Model Research Articles

A novel coupled rainfall prediction model based on stepwise decomposition technique

The traditional decomposed ensemble prediction model decomposes the entire rainfall sequence into several sub-sequences, dividing them into training and testing periods for modeling. During sample construction, future information is erroneously mixed into the training data, making it challenging to apply in practical rainfall forecasting. This paper proposes a novel stepwise decomposed ensemble coupling model, realized through variational mode decomposition (VMD) and bidirectional long short-term memory neural network (BiLSTM) models. Model parameters are optimized using an improved particle swarm optimization (IPSO). The performance of the model was evaluated using rainfall data from the Southern Four Lakes basin. The results indicate that: (1) Compared to the PSO algorithm, the IPSO algorithm-coupled model shows a minimum decrease of 2.70% in MAE and at least 2.62% in RMSE across the four cities in the Southern Four Lakes basin; the IPSO algorithm results in a minimum decrease of 25.58% in MAE and at least 28.19% in RMSE for the VMD-BiLSTM model. (2) When compared to IPSO-BiLSTM, the VMD-IPSO-BiLSTM based on the stepwise decomposition technique exhibits a minimum decrease of 26.54% in MAE and at least 34.16% in RMSE. (3) The NSE for the testing period of the VMD-IPSO-BiLSTM model in each city surpasses 0.88, indicating higher prediction accuracy and providing new insights for optimizing rainfall forecasting.

Interpretable baseflow segmentation and prediction based on numerical experiments and deep learning

Baseflow is a crucial water source in the inland river basins of high-cold mountainous region, playing a significant role in maintaining runoff stability. It is challenging to select the most suitable baseflow separation method in data-scarce high-cold mountainous region and to evaluate effects of climate factors and underlying surface changes on baseflow variability and seasonal distribution characteristics. Here we attempt to address how meteorological factors and underlying surface changes affect baseflow using the Grey Wolf Optimizer Digital Filter Method (GWO-DFM) for rapid baseflow separation and the Long Short-Term Memory (LSTM) neural network model for baseflow prediction, clarifying interpretability of the LSTM model in baseflow forecasting. The proposed method was successfully implemented using a 63-year time series (1958–2020) of flow data from the Tai Lan River (TLR) basin in the high-cold mountainous region, along with 21 years of ERA5-land meteorological data and MODIS data (2000–2020). The results indicate that: (1) GWO-DFM can rapidly identify the optimal filtering parameters. It employs the arithmetic average of three methods, namely Chapman, Chapman-Maxwell and Eckhardt filter, as the best baseflow separation approach for the TLR basin. Additionally, the baseflow significantly increases after the second mutation of the baseflow rate. (2) Baseflow sources are mainly influenced by precipitation infiltration, glacier frozen soil layers, and seasonal ponding. (3) Solar radiation, temperature, precipitation, and NDVI are the primary factors influencing baseflow changes, with Nash-Sutcliffe efficiency coefficients exceeding 0.78 in both the LSTM model training and prediction periods. (4) Changes in baseflow are most influenced by solar radiation, temperature, and NDVI. This study systematically analyzes the changes in baseflow and response mechanisms in high-cold mountainous region, contributing to the management of water resources in mountainous basins under changing environmental conditions.

Deep learning-based modeling of the cyclic behavior of replaceable fuse buckling-restrained braces (BRBs)

This paper presents a novel approach for predicting the behavior of buckling restrained braces (BRBs) using long short-term memory (LSTM) and residual neural network (ResNet) models. Nonlinear finite element analysis (NLFEA) usually requires 3D modeling of complex parts and assigning various material and simulation properties before conducting the analysis, which can be time-consuming and computationally costly. LSTM networks are particularly well-suited for modeling sequential data, such as time series, and ResNet is a discriminative deep learning-based model that can handle deeper architectures without the issue of vanishing gradient. Developing a general LSTM model requires a comprehensive database for training, cross-validation, and blind testing. Even if such a database were available, concatenating data from different BRB specimens would not be feasible when modeling a data series. Such a method would intertwine the features of the individual specimens, rendering the data useless for machine-learning purposes. On the other hand, a well-trained LSTM can predict cyclic behavior for specimens that resemble the training data. As such, a group of well-trained models can produce superior predictions when used on individually familiar data. Hence, there is a need for a prescreening classification stage. The framework presented herein utilizes an AI classifier (ResNet) that automatically recognizes the BRB specimen type of the input data and directs it to the appropriate LSTM model for prediction. The presented framework demonstrates excellent predictions with an accuracy greater than 99%. It is customizable, adaptable, and scalable to include different BRB specimens. The input data consisted of the displacement of the BRB specimens based on the AISC 341–10 qualification recommendations, while the output was the corresponding force (hysteretic response to cyclic loading). The proposed framework can be valuable in earthquake engineering applications as it demonstrates superior predictions.

Transfer learning in environmental data-driven models: A study of ozone forecast in the Alpine region

Many environmental variables, in particular, related to air or water quality, are measured in a limited number of points and often for a limited time span. This forbids the development of accurate models for interesting locations with missing or insufficient data and poses the question of whether a model developed for another measurement site can be reliably applied. Such a question is particularly critical when the model is entirely data-driven, such as a neural network. In this context, the paper proposes a procedure to evaluate the expected performance of an existing neural network model applied to a new unmonitored station. This transferability assessment is exemplified by the problem of forecasting ozone concentrations in different environmental settings around the Alpine Arc. Long Short-Term Memory (LSTM) neural network models are applied for predicting hourly concentrations in 20 stations of different types (urban, rural, and mountain). The analysis of the results allows us to determine the expected performance of such models in new cases and reduce the transferability uncertainty when the existing models can be partitioned into clusters. The LSTM models demonstrate the possibility of high accuracy in ozone forecasting at all sites. Given the significant impacts of this gas on human health and the environment, this can contribute to better decision-making and mitigation strategies for air pollution control.

Leveraging distributed AI for multi-occupancy prediction in Cognitive Buildings

Cognitive Buildings are autonomous smart environments capable of setting themselves according to some self-learned rules. Such rules are inferred according to, e.g., the inhabitants’ behaviors, users’ needs, and specific policies for optimizing security, energy, and comfort management. To do this, it is of foremost importance to gather information about users’ habits like room occupancy. Indeed, Cognitive Buildings can effectively exploit information about sensors in the different rooms, thus being able to detect, learn, and forecast the presence of users in the buildings and act in accordance with these predictions. In this direction, this paper proposes an innovative approach for multi-occupancy prediction in Cognitive Buildings, incorporating a multi-layer hierarchy for Federated Learning, the utilization of IoT devices at the Edge, the implementation of long short-term memory neural network models, and the exploitation of Edge Computing. The approach also introduces a versatile design template for developing real distributed systems for occupancy prediction. The proposed approach uses a distributed paradigm to safeguard data privacy so that the collected data is used to train separate local deep learning models, which are then merged in the Cloud. The paper validates the approach by providing a preliminary prototype realized at ICAR-CNR, Rende Italy, and presents a performance analysis, which shows that the occupancy is predicted with an 84.5% accuracy.

Evaluation of multimodal data-driven financial risk prediction methods for corporate green credit

With the rapid economic development and increasingly serious environmental problems, many regions have launched green credit policies. Green credit can reduce the loan interest rate of the environmental protection industry and lower the financing threshold. Traditional risk prediction methods cannot comprehensively evaluate the green credit risk of the enterprise based on the degree of green environmental protection and the industry environment in which the enterprise is located, resulting in the inconsistency between the credit financial risk prediction and the actual results, which increases the bank credit risk. In order to strengthen the management level of green credit and reduce the probability of non-performing loans, a scientific risk assessment method was constructed by using a combination of automatic encoding network and bidirectional long short-term memory neural network model to predict the financial risks of green credit, driven by multi-modal data. Through the study of multimodal data, this paper took green credit financial risk as the research object, aggregated the information of various enterprises to improve the bank’s capital utilization rate, and also promoted enterprises to take the initiative to transform into the direction of green environmental protection. Finally, the experiment proved that multimodal data fusion model was more superior than random forest in risk prediction, reducing the bank’s non-performing loan rate by 3.1% and improving the bank’s risk control level.

LSTM deep learning long-term traffic volume prediction model based on Markov state description

ABSTRACT When applying long short-term memory (LSTM) neural network model to traffic prediction, there are limitations in exploiting spatial-temporal traffic state features. The interpretability of models has not received enough attention. This study suggests an LSTM traffic flow prediction model that can anticipate traffic volume 24 h in advance. The model makes use of the traffic flow state information obtained from the fuzzy C-means clustering method by clustering the multi-day historical traffic flow data. Markov chain is used to capture the label feature of traffic flow using transition probability matrix information. To show the efficacy of the suggested technique, experiments were conducted using real traffic volume data from a city in China. The simulation results demonstrate that the proposed model can attain greater prediction accuracy, and the network training time may be significantly reduced.

Real-time mapping of crop canopy temperature using a wireless network of infrared thermometers on a central pivot

Efficient irrigation is essential for sustainable food production and regional water security. Temperature measurements provide information about crop water status, which allows the estimation of water needs and precise irrigation scheduling. This paper proposes a technology for agricultural irrigation management that combines cost-effective implementation with the ability to map crop canopy temperatures using a wireless network of infrared thermometers on a center pivot. The technology integrates Wireless Sensor Networks (WSN) with scalability options, IoT capabilities, and web data availability, enabling precise plant water stress quantification based on canopy temperature and soil water content. It facilitates real-time irrigation prescriptions while the center pivot is in motion, with canopy temperature sensors providing accurate readings of +/- 0.5 °C without additional calibration. Nonlinear Autoregressive Network with Exogenous Inputs (NARX) and Long Short-Term Memory (LSTM) neural network models were trained to predict irrigation prescriptions, achieving Root Mean Squared Error (RMSE < 4.18 mm) and correlation coefficient (r^2 > 0.82). This technology will allow the definition of real-time and site-specific prescriptions for variable-rate irrigation systems using infrared temperature sensors. Future work includes exploring LoRaWAN technologies, implementing a Differential Global Positioning System (DGPS), improving spatial resolution with data fusion algorithms, refining irrigation prescription models, reducing sensor costs and developing models considering soil and crop effects.

Deep learning: an study on financial crisis forewarning in small and medium-sized listed enterprises

Early warning of financial crisis will greatly promote the stable development of small and medium-sized listed enterprises (listed SMEs). In this article, 13 warning indicators were selected for financial crisis prediction from five aspects: profit level, debt service level, business level, cash level, and development level. Then, the parameters of the long short-term memory (LSTM) neural network model were optimised by the whale optimisation algorithm (WOA), resulting in the WOA-LSTM model. The WOA-LSTM model achieved an accuracy of 0.975 in predicting financial crises for listed SMEs. The performance of the WOA-LSTM model was significantly enhanced when using the filtered 13 indicators as inputs, compared to using the original 24 indicators. The findings prove the dependability of the WOA-LSTM model in warning financial crises of listed SMEs and the feasibility of its application in practice.

Real-time prediction of mooring tension for semi-submersible platforms

The harsh operating environment of deepwater semi-submersible production platforms poses severe challenges to the safe operation and maintenance of their production processes. Using artificial intelligence technology to improve the operation and maintenance level of large-scale offshore engineering equipment has become an inevitable trend. Accurate mooring tension prediction for semi-submersible production platforms can effectively identify potential damage and ensure the platform's safe and efficient production operation. Compared with traditional cyclic neural networks, transformer neural networks have faster calculation speeds, better generalization abilities and scalability, and better abilities to capture long-distance dependencies. In this study, the first deepwater semi-submersible production platform in the South China Sea, “DeepSea One,” was selected as the research object, and an improved novel transformer neural network architecture was proposed for the mooring tension prediction of semi-submersible production platforms using actual long-term mooring tension data obtained during the platform's production process. The proposed method was compared with the commonly used long short-term memory neural network model, and the comparison proved the effectiveness of the proposed method in predicting the dynamic tension of semi-submersible production platform mooring. In addition, the proposed method is also applicable for the prediction of other time series data in the field of shipbuilding and offshore engineering, which is of certain value for generalization.

Dynamic response prediction of high-speed train on cable-stayed bridge based on genetic algorithm and fused neural networks

To predict the dynamic response of high-speed trains (HSTs) passing through cable-stayed bridges (CSBs), this paper proposed a prediction framework based on the genetic algorithm (GA) and the neural network. Firstly, an HST-CSB coupled dynamic model was established using the rigid-flexible coupling method. Subsequently, a fused Temporal Convolutional Network and Long Short-Term Memory (TCN-LSTM) neural network model was developed, and its hyperparameters were optimized using the GA. The dataset was created based on the HST-CSB coupled dynamic model for training and testing the model. Additionally, the generalization capability of the model after adopting different track deformations was discussed. The main results were as follows: Comparative analysis with various neural network models revealed that the GA-TCN-LSTM model exhibited the most accurate prediction results across different train speeds. The significant temperature fluctuations of the bridge led to a drastic deterioration of the track smoothness, resulting in decreased predictive performance of the neural network models. The GA-TCN-LSTM model could optimize the hyperparameters, demonstrating good robustness and generalization capabilities. The predictive performance of the model improved when using the German low-interference spectrum, attributed to the greater irregularities in the German low-interference spectrum, resulting in larger train dynamic responses and stronger feature characteristics in the training data for the neural network. Different neural network models exhibited varying adaptability to track irregularities, with the GA-TCN-LSTM model demonstrating high accuracy in predicting dynamic responses under different train speeds.

Comparison of Long Short-Term Memory and Convolutional Neural Network Models for Emergency Department Patients’ Arrival Daily Forecasting

Background: One of the most critical challenges in the emergency department (ED) is overcrowding, which creates negative consequences for patients and staff. Therefore, predicting the rate of patients entering the ED can help manage resources in this department effectively. Objectives: According to the time of data collection, we intended to predict the volume of patient admissions to the ED in epidemic conditions, such as COVID-19 and non-epidemic. In addition, we planned to compare the performance of the LSTM and CNN models. Methods: The collected data consists of three main categories. The first category pertains to air pollutants, provided by the Tehran air quality control organization. The second type relates to data from the Iran Meteorological Organization, and the third category includes the number of patients admitted to the ED of a hospital in Tehran. We also incorporated binary indicators for epidemic and non-epidemic conditions in the dataset. The data collection period spans from February 2018 to March 2022. We employed the Dickey-Fuller test to assess the stationarity of the data. After preprocessing, we independently developed long short-term memory (LSTM) neural network and convolutional neural network (CNN) models, considering various time windows of previous days. Keras and Tensorflow libraries in Python, along with the Google Colab environment, were utilized to execute the models. Results: The LSTM model exhibited the lowest root mean square error (RMSE) and mean absolute error (MAE) with a time window of the last seven days, while the CNN model outperformed the LSTM model with a time window of the previous 13 days. Additionally, the CNN model required less execution time than the LSTM model. Conclusions: In conclusion, deep learning algorithms prove suitable for analyzing multivariate time series data. The CNN model demonstrated the lowest prediction error.

Forecasting Methane Data Using Multivariate Long Short-Term Memory Neural Networks

AbstractOver the past few decades, Alberta has witnessed a remarkable expansion in its oil and gas sector. Unfortunately, this growth has come at a cost, as Alberta has become the fastest-growing source of pollutant emissions in greenhouse gases (GHGs), sulphur emissions, and water pollution in Canada. Among these GHGs, methane stands out as the second most prevalent GHG, possessing a global warming potential ~ 28 times higher than carbon dioxide over a span of 100 years, and ~ 80 times higher over a period of 20 years. Since 1986, the Alberta Energy Regulator (AER) has been diligently gathering data on methane concentrations. Although this data is publicly available, its analysis has not been thoroughly explored. Our study aims to investigate the impact of temperature, wind speed, and wind direction on the predictions of methane concentration time series data, utilizing a long short-term memory (LSTM) neural network model. Our findings indicate that the inclusion of climate variables enhances the predictive capabilities of the LSTM model. However, the results show that it is not obvious which variable has the most impact on the improvement although temperature appears to have a better effect on improving predictive performance compared to wind speed and direction. The results also suggest that the variance of the input data does not affect forecasting performance.

An improved sparrow search algorithm and CNN-BiLSTM neural network for predicting sea level height

Accurate prediction of sea level height is critically important for the government in assessing sea level risk in coastal areas. However, due to the nonlinear, time-varying and highly uncertain characteristics of sea level change data, sea level prediction is challenging. To improve the accuracy of sea level prediction, this paper uses a new swarm intelligence algorithm named the sparrow search algorithm (SSA), which can imitate the foraging behavior and antipredation behavior of sparrows, to determine optimal solutions. To avoid the algorithm falling into a local optimal situation, this paper integrates the sine–cosine algorithm and the Cauchy variation strategy into the SSA to obtain an algorithm named the SCSSA. The SCSSA is used to optimize the parameter values of the CNN-BiLSTM (convolutional neural network combined with bidirectional long short-term memory neural network) model; finally, a combined neural network model (named SCSSA-CNN-BiLSTM) is proposed. In this paper, the time series data of seven tidal stations located in coastal China are used for experimental analysis. First, the SCSSA-CNN-BiLSTM model is compared with the CNN-BiLSTM model to predict the time series data of SHANWEI Station. With respect to the training and test sets of data, the SCSSA-CNN-BiLSTM model outperforms the other models on all the evaluation metrics. In addition, the remaining six tide station datasets and five neural network models, including the SCSSA-CNN-BiLSTM model, are used to further study the performance of the proposed prediction model. Four evaluation indices including the root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and coefficient of determination (R2) are adopted. For six stations, the RMSE, MAE, MAPE and R2 of SCSSA-CNN-BiLSTM model are ranged from 20.9217 ~ 27.8427 mm, 9.4770 ~ 17.8603 mm, 0.1322% ~ 0.2482% and 0.9119 ~ 0.9759, respectively. The experimental analysis results show that the SCSSA-CNN-BiLSTM model makes effective predictions at all stations, and the prediction performance is better than that of the other models. Even though the combination of SCSSA algorithm may increase the complexity of the model, indeed the proposed model is a new prediction method with good accuracy and robustness for predicting sea level change.

Integration of deep learning and improved multi-objective algorithm to optimize reservoir operation for balancing human and downstream ecological needs

Dam (reservoir)-induced alterations of flow and water temperature regimes can threaten downstream fish habitats and native aquatic ecosystems. Alleviating the negative environmental impacts of dam-reservoir and balancing the multiple purposes of reservoir operation have attracted wide attention. While previous studies have incorporated ecological flow requirements in reservoir operation strategies, a comprehensive analysis of trade-offs among hydropower benefits, ecological flow, and ecological water temperature demands is lacking. Hence, this study develops a multi-objective ecological scheduling model, considering total power generation, ecological flow guarantee index, and ecological water temperature guarantee index simultaneously. The model is based on an integrated multi-objective simulation-optimization (MOSO) framework which is applied to Three Gorges Reservoir. To that end, first, a hybrid long short-term memory and one-dimensional convolutional neural network (LSTM_1DCNN) model is utilized to simulate the dam discharge temperature. Then, an improved epsilon multi-objective ant colony optimization for continuous domain algorithm (ε-MOACOR) is proposed to investigate the trade-offs among the competing objectives. Results show that LSTM _1DCNN outperforms other competing models in predicting dam discharge temperature. The conflicts among economic and ecological objectives are often prominent. The proposed ε-MOACOR has potential in resolving such conflicts and has high efficiency in solving multi-objective benchmark tests as well as reservoir optimization problem. More realistic and pragmatic Pareto-optimal solutions for typical dry, normal and wet years can be generated by the MOSO framework. The ecological water temperature guarantee index objective, which should be considered in reservoir operation, can be improved as inflow discharge increases or the temporal distribution of dam discharge volume becomes more uneven.

High-Precision Microscale Particulate Matter Prediction in Diverse Environments Using a Long Short-Term Memory Neural Network and Street View Imagery.

In this study, we propose a novel long short-term memory (LSTM) neural network model that leverages color features (HSV: hue, saturation, value) extracted from street images to estimate air quality with particulate matter (PM) in four typical European environments: urban, suburban, villages, and the harbor. To evaluate its performance, we utilize concentration data for eight parameters of ambient PM (PM1.0, PM2.5, and PM10, particle number concentration, lung-deposited surface area, equivalent mass concentrations of ultraviolet PM, black carbon, and brown carbon) collected from a mobile monitoring platform during the nonheating season in downtown Augsburg, Germany, along with synchronized street view images. Experimental comparisons were conducted between the LSTM model and other deep learning models (recurrent neural network and gated recurrent unit). The results clearly demonstrate a better performance of the LSTM model compared with other statistically based models. The LSTM-HSV model achieved impressive interpretability rates above 80%, for the eight PM metrics mentioned above, indicating the expected performance of the proposed model. Moreover, the successful application of the LSTM-HSV model in other seasons of Augsburg city and various environments (suburbs, villages, and harbor cities) demonstrates its satisfactory generalization capabilities in both temporal and spatial dimensions. The successful application of the LSTM-HSV model underscores its potential as a versatile tool for the estimation of air pollution after presampling of the studied area, with broad implications for urban planning and public health initiatives.

Privacy-Preserving Federated Learning for Value-Added Service Model in Advanced Metering Infrastructure

Advanced metering infrastructure (AMI) is the backbone of the next generation smart city and smart grid; it not only provides near real-time two-way communication between the consumers and the energy systems, but also enables the third parties (TPs) to provide relevant value-added services to the consumers to improve user satisfaction. However, the existing services are implemented in a centralized manner, which has potential and associated security and privacy risks also increased with Internet-of-Things (IoT) devices. To better balance the quality of the services and ensure users’ privacy, a TP AMI service model based on differentially private federated learning (FL) is proposed in this article. Instead of sending the private energy data to the cloud server, the proposed service model trains the neural network models locally, and only model parameters are shared with the central server. Moreover, the identity of individuals is eliminated by adding random Gaussian noise during the secure aggregation. Furthermore, an attention-based bidirectional long short-term memory (ATT-BLSTM) neural network model is adopted to solve the long-range dependency problem of conventional neural networks. In the case study, a residential short-term load forecasting (STLF) task is implemented to evaluate the performance of the proposed model. Compared with other state-of-the-art energy service models, the proposed one can achieve similar accuracy as the typical centralized model and balances the trade-off between privacy loss and prediction accuracy flexibly.

Estimation of PM2.5 Concentration across China Based on Multi-Source Remote Sensing Data and Machine Learning Methods

Long-term exposure to high concentrations of fine particles can cause irreversible damage to people’s health. Therefore, it is of extreme significance to conduct large-scale continuous spatial fine particulate matter (PM2.5) concentration prediction for air pollution prevention and control in China. The distribution of PM2.5 ground monitoring stations in China is uneven with a larger number of stations in southeastern China, while the number of ground monitoring sites is also insufficient for air quality control. Remote sensing technology can obtain information quickly and macroscopically. Therefore, it is possible to predict PM2.5 concentration based on multi-source remote sensing data. Our study took China as the research area, using the Pearson correlation coefficient and GeoDetector to select auxiliary variables. In addition, a long short-term memory neural network and random forest regression model were established for PM2.5 concentration estimation. We finally selected the random forest regression model (R2 = 0.93, RMSE = 4.59 μg m−3) as our prediction model by the model evaluation index. The PM2.5 concentration distribution across China in 2021 was estimated, and then the influence factors of high-value regions were explored. It is clear that PM2.5 concentration is not only related to the local geographical and meteorological conditions, but also closely related to economic and social development.

Predicting state level suicide fatalities in the united states with realtime data and machine learning

Digital trace data and machine learning techniques are increasingly being adopted to predict suicide-related outcomes at the individual level; however, there is also considerable public health need for timely data about suicide trends at the population level. Although significant geographic variation in suicide rates exist by state within the United States, national systems for reporting state suicide trends typically lag by one or more years. We developed and validated a deep learning based approach to utilize real-time, state-level online (Mental Health America web-based depression screenings; Google and YouTube Search Trends), social media (Twitter), and health administrative data (National Syndromic Surveillance Program emergency department visits) to estimate weekly suicide counts in four participating states. Specifically, per state, we built a long short-term memory (LSTM) neural network model to combine signals from the real-time data sources and compared predicted values of suicide deaths from our model to observed values in the same state. Our LSTM model produced accurate estimates of state-specific suicide rates in all four states (percentage error in suicide rate of −2.768% for Utah, −2.823% for Louisiana, −3.449% for New York, and −5.323% for Colorado). Furthermore, our deep learning based approach outperformed current gold-standard baseline autoregressive models that use historical death data alone. We demonstrate an approach to incorporate signals from multiple proxy real-time data sources that can potentially provide more timely estimates of suicide trends at the state level. Timely suicide data at the state level has the potential to improve suicide prevention planning and response tailored to the needs of specific geographic communities.

Load Day-Ahead Automatic Generation Control Reserve Capacity Demand Prediction Based on the Attention-BiLSTM Network Model Optimized by Improved Whale Algorithm

Load forecasting is a research hotspot in academia; in the context of new power systems, the prediction and determination of load reserve capacity is also important. In order to adapt to new forms of power systems, a day-ahead automatic generation control (AGC) reserve capacity demand prediction method based on the Fourier transform and the attention mechanism combined with a bidirectional long and short-term memory neural network model (Attention-BiLSTM) optimized by an improved whale optimization algorithm (IWOA) is proposed. Firstly, based on the response time, Fourier transform is used to refine the distinction between various types of load reserve demand, and the power of the AGC reserve band is calculated using Parseval’s theorem to obtain the reserve capacity demand sequence. The maximum mutual information coefficient method is used to explore the relevant influencing factors of the AGC reserve sequence concerning the data characteristics of the AGC reserve sequence. Then, the historical daily AGC reserve demand sequences with relevant features are input into the Attention-BiLSTM prediction model, and the improved whale algorithm is used to automatically find the optimal hyperparameters to obtain better prediction results. Finally, the arithmetic simulation results show that the model proposed in this paper has the best prediction performance with the upper (0.8810) and lower (0.6651) bounds of the coefficient of determination (R2) higher than the other models, and it has the smallest mean absolute percentage error (MAPE) and root mean square error (RMSE).