Industrial circulating water system temperature prediction method based on BiLSTM-ARIMA-Attention

By adopting a two-way long and short-term memory-autoregressive integrated moving average neural network method based on attention mechanism in industrial circulating water systems, the problem of insufficient efficiency and accuracy of return water temperature prediction in the existing technology is solved, and a higher Prediction accuracy and stability.

CN118981750APending Publication Date: 2024-11-19XIANGTAN UNIV
View PDF 0 Cites 4 Cited by

Patent Information

Application Number
CN202411012096.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-26
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The existing technology has problems that prediction efficiency and accuracy are difficult to guarantee in the prediction of return water temperature prediction in industrial circulation water systems. Especially when dealing with complex nonlinear and long-term dependency data, deep learning models are prone to overfitting, resulting in accuracy. The rate is affected.

Method used

Using the two-way long and short-term memory-autoregressive comprehensive moving average neural network method based on attention mechanism, the two-way long and short-term memory neural network training, attention mechanism enhancement and autoregressive comprehensive moving average model correction are established. The temperature prediction model of the circulating water system improves the model's processing ability of long-term dependencies and nonlinear data.

Benefits of technology

It improves the accuracy and stability of return water temperature prediction, reduces interference from irrelevant information, corrects systematic errors and deviations, and improves the prediction accuracy and stability of the model.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981750A_ABST
    Figure CN118981750A_ABST
Patent Text Reader

Abstract

The invention provides an industrial circulating water system temperature prediction method which comprises the following steps: firstly, acquiring original data such as temperature, flow and pressure of a blast furnace system and a circulating water system, and preprocessing; a feature reduction method based on fuzzy inconsistency measurement is adopted to measure feature importance, redundant features are removed, and a key feature subset is obtained; dividing the data into a training set, a verification set and a test set, and performing training and evaluation by using a bidirectional long-short-term memory neural network model; predicting by using the verification set, and calculating a residual sequence; weighting the features in local and global dimensions by adopting an attention mechanism, and enhancing the difference of different time steps; then using an autoregressive comprehensive moving average model to correct a residual error, and integrating the residual error with a predicted value to obtain a final prediction result; a comprehensive model is established based on this, and prediction performance is evaluated by using a test set. The method can accurately predict the return water temperature and improve the system efficiency and the energy-saving effect.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of blast furnace cooling equipment, and in particular to a method for predicting return water temperature of an industrial circulating water system based on a bidirectional long short-term memory-autoregressive integrated moving average neural network (BiLSTM-ARIMA-Attention) with an attention mechanism. Background Art

[0002] Industrial circulating water systems are widely used in industries such as chemical, steel, and electric power, mainly for cooling equipment and heat in process flows. The system usually includes components such as cooling towers, heat exchangers, pumps, and pipes. Cooling water circulates in the system, absorbs the heat generated by the equipment, and then dissipates the heat through the cooling tower and returns to the system for re-circulation. Return water temperature is an important parameter to measure the operating efficiency of industrial circulating water systems. Accurately predicting return water temperature can help optimize cooling water flow and adjust system parameters, thereby improving cooling effects, reducing energy consumption, and extending equipment life.

[0003] At present, the methods involved in predicting the return water temperature of industrial circulating water systems mostly use traditional machine learning methods. These methods manually extract features such as ambient temperature, cooling tower efficiency, system flow, equipment load, etc., and use them to establish prediction models. However, this method usually requires rich expert experience and relevant field knowledge, and the prediction efficiency and accuracy are difficult to guarantee. In addition, since the return water temperature is affected by many complex factors, traditional methods have limitations in dealing with nonlinearity and long-term dependencies. Although deep learning methods perform well in dealing with complex nonlinear relationships and time series data, it is still difficult to fully and accurately capture all time series features when the data is highly nonlinear. Deep learning models are easily overfitted when trained under the influence of data noise and sample imbalance, which affects the accuracy of return water temperature prediction. Summary of the invention

[0004] The present invention proposes a bidirectional long short-term memory-autoregressive integrated moving average neural network method based on an attention mechanism for predicting the return water temperature of an industrial circulating water system to solve the problems raised in the above background technology.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] S1: Obtain the original operation data of the circulating water system and blast furnace system;

[0007] S2: Data cleaning and preprocessing to obtain temperature prediction time data set;

[0008] S3: Perform deep learning feature extraction and simplification on the model and output simplified feature set data;

[0009] S4: Bidirectional long short-term memory neural network is used to train and validate the time dataset;

[0010] S5: Use the attention mechanism method to enhance the model's ability to capture long-term dependencies;

[0011] S6: Use the autoregressive integrated moving average model to correct the residual sequence;

[0012] S7: Establish a temperature prediction model for the circulating water system.

[0013] Compared with the prior art, the present invention provides a temperature prediction method, which comprises: after data cleaning and preliminary preprocessing of the collected original operation data of the blast furnace system and the circulating water system, the original data are classified into a training set, a verification set and a test set through variance filtering and correlation analysis; considering that the industrial data has large noise and complex nonlinearity, a deep learning method based on fuzzy inconsistency measurement is used to extract and simplify features, and the simplified feature set data is output; a bidirectional long short-term memory neural network is used to capture the medium- and long-term dependencies in the time series data, and the accuracy and stability of the prediction are improved by considering the past and future information at the same time, and the prediction sequence samples are obtained, and compared with the real verification set to obtain the residual sequence; an attention mechanism is introduced to improve the model's ability to capture key information, reduce the interference of irrelevant information on the prediction results, and further optimize the model's expression ability and prediction effect; an autoregressive integrated moving average model is used to correct the residual sequence, correct the residual signal that has not been fully captured, correct the systematic errors and deviations that may exist in the original prediction, and further improve the prediction accuracy and stability of the model; a bidirectional long short-term memory-autoregressive integrated moving average neural network model based on the attention mechanism is established, and the prediction set is verified and the model prediction results are evaluated. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic flow chart of a temperature prediction method provided by the present invention.

[0015] Figure 2 A schematic flow chart of a temperature prediction method provided by the present invention.

[0016] Figure 3 A schematic flow chart of a temperature prediction method provided by the present invention.

[0017] Figure 4 A schematic flow chart of a temperature prediction method provided by the present invention.

[0018] Figure 5 A schematic flow chart of a temperature prediction method provided by the present invention.

[0019] Figure 6 A schematic flow chart of a temperature prediction method provided by the present invention.

[0020] Figure 7 A schematic flow chart of a temperature prediction method provided by the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the present invention clearer, a bidirectional long short-term memory-autoregressive integrated moving average neural network return water temperature prediction method based on attention mechanism of the present invention will be further described in detail in combination with the accompanying drawings and embodiments.

[0022] A bidirectional long short-term memory-autoregressive integrated moving average neural network return water temperature prediction method based on attention mechanism, the method comprising:

[0023] S101: Based on historical data, the original operation data of the circulating water system and the blast furnace system are obtained with a sampling frequency of 20 minutes to construct a time series;

[0024] S102: Cleaning the original data and obtaining a temperature prediction time data set through normalization preprocessing;

[0025] S103: A feature reduction method based on fuzzy inconsistency measurement is proposed to remove some redundant features and classify them into training set, validation set and test set;

[0026] S104: using a bidirectional long short-term memory neural network to train the model, determining the model hyperparameters according to the spatiotemporal characteristics between the variables, and obtaining a residual sequence based on the validation prediction value and the validation set;

[0027] S105: Use the attention mechanism method to enhance the model's ability to capture long-term dependencies;

[0028] S106: using an autoregressive integrated moving average model, determining model hyperparameters through autocorrelation calculation and partial autocorrelation calculation, and performing error correction on the residual sequence;

[0029] S107: Establish a temperature prediction model and use a test set to evaluate the model.

[0030] In one embodiment, the acquired characteristic signals include temperature and flow data, which are acquired through sensors installed in the circulating water system and the blast furnace system, and there are time response delays and spatial position differences between the sensors.

[0031] In one embodiment, the time series data may include at least one of the following: a temperature signal, a flow signal, and a pressure signal.

[0032] In one embodiment, the original operation data of the blast furnace system and the circulating water system are collected through industrial sites, and there are a lot of noise and abnormal values ​​in the data.

[0033] In one embodiment, due to improper manual operation and other reasons, some features lack original data, resulting in blank values ​​and showing complex nonlinear characteristics.

[0034] In one embodiment, data from each sensor at the same time is collected at a frequency of 20 minutes, and the data types covered include but are not limited to temperature, flow rate, and pressure.

[0035] In one embodiment, by deeply studying the operation process of the industrial circulating water system and the blast furnace system, a batch of data that is basically irrelevant to the total return water pipe return water temperature prediction process is manually screened out to simplify the data complexity.

[0036] In one embodiment, noise and outliers are effectively removed through preprocessing methods such as data cleaning and filtering. Then, the time series data is processed using a normalization method to ensure that data with different characteristics are on a uniform scale for subsequent analysis and model training.

[0037] In one embodiment, in order to reduce the redundancy of deep learning features, a feature reduction method based on fuzzy inconsistency measurement is proposed. By calculating the fuzzy inconsistency between features, some redundant features are eliminated and simplified feature set data is output.

[0038] In one embodiment, the processed time series data is divided into a training set, a validation set, and a test set, wherein the ratio of the three data sets is 7:2:1.

[0039] In one embodiment, a bidirectional long short-term memory neural network model is constructed based on the processed time series simplified data. According to the spatiotemporal characteristics of the data, a time step of 200 and an overlapping time point of 150 are selected, and the remaining model hyperparameters are set according to the original default values ​​to train the training set data.

[0040] In one embodiment, a bidirectional long short-term memory neural network is used to first capture the long-term dependencies in the time series data, and the accuracy and stability of the prediction are improved by considering both past and future information, and the predicted sequence samples are obtained and compared with the actual verification set to obtain the residual sequence.

[0041] In one embodiment, by introducing a local attention mechanism and a global attention mechanism to distinguish the differences between different time series features, the local attention mechanism weights the output features of each time step, and the global attention mechanism weights the fusion features of all time steps, thereby enhancing the differences between different time series features, so that the model can better capture the characteristics of temperature and flow signals.

[0042] In one embodiment, an autoregressive integrated moving average model is used to correct the prediction residual sequence of the bidirectional long short-term memory neural network model. The hyperparameters q and p of the autoregressive integrated moving average model are determined by autocorrelation and partial autocorrelation calculations, respectively, to correct the residual signal that has not been fully captured, to correct the possible systematic errors and deviations in the original prediction, and to fuse the corrected residual sequence with the prediction value of the validation set to obtain the final prediction value.

[0043] In one embodiment, an autoregressive integrated moving average-bidirectional long short-term memory neural network model is established, and the output is predicted by inputting the final test set, thereby enhancing the model's processing of long-term data dependencies and data nonlinearity.

[0044] In one embodiment, the operating data of the circulating water system and the blast furnace system are periodically acquired and preprocessed and predicted in real time to ensure that the system can respond to changes in a timely manner and improve the real-time and accuracy of the prediction.

[0045] like Figure 2 As shown, the method for processing the original data includes the following contents:

[0046] S201: Obtaining original operation data of the circulating water system and the blast furnace system;

[0047] S202: Manually screen out data that is basically irrelevant to the prediction object according to the process flow;

[0048] S203: pre-processing the original operation data, such as variance filtering method, correlation analysis, etc.;

[0049] S204: Obtain a temperature prediction time data set.

[0050] In one embodiment, various sensors of the blast furnace system are used to collect original operating data of the circulating water system and the blast furnace system; data that is basically irrelevant to the prediction object is manually screened out according to the process flow to ensure the relevance and validity of the data.

[0051] In one embodiment, the filtered raw operating data is preprocessed, such as using variance filtering, correlation analysis and other methods, to eliminate invalid features, retain key features that have a greater impact on temperature prediction, and obtain standardized operating data.

[0052] In one embodiment, the obtained standardized operation data is used to construct a prediction time data set according to the requirements of temperature prediction. The specific steps include normalizing the pre-processed data and converting it into a standard format suitable for input into a short-term memory neural network model of a deep learning algorithm.

[0053] In one embodiment, in order to improve the accuracy and robustness of the temperature prediction model, a variety of time series analysis methods can be used to process the prediction time data set, such as differential processing, sliding average, etc., to eliminate trends and noise in the data, and finally generate a high-quality temperature prediction time data set for model training and prediction.

[0054] like Figure 3 As shown in the figure, a feature reduction method based on fuzzy inconsistency measurement is used to reduce the features of the dataset to obtain a subset of key deep learning features.

[0055] In one embodiment, let U be the sample fault deep learning feature set, U=C∪D, where C is the sample condition feature set, and D is the sample decision feature set, that is, the class attribute of the sample set.

[0056] In one embodiment, for a sample x in a discrete system k ,x1∈U, if x k If the sample x has the same feature value as x1 on feature u∈Q but different decision value, then k Inconsistent with x1 on feature u∈Q, denoted as Conflict u (x k ,x1)=1; otherwise, sample x k The feature set u∈Q is consistent with the sample x1, which is denoted as Conflict u (x k ,x1)=0.

[0057] In one embodiment, the discrete sample x k The inconsistent set under feature u∈Q is defined as k The set of fuzzy inconsistent samples under feature u is denoted as Conflict u (x k ), defined as follows:

[0058] Conflict u (x k )={∪x1}

[0059] st.Conflict u (x k ,x1)∈(0,1)

[0060] In one embodiment, sample x k The fuzzy inconsistency measure of the sample set U under the feature u∈Q is denoted as CR u (x k )=μ u (x k ). The fuzzy inconsistency measure of sample set U under feature u is Where |U| is the number of samples.

[0061] In one embodiment, a feature reduction method is used to simplify the deep learning feature set of water supply pump faults to generate a simplified time data feature set.

[0062] like Figure 4 As shown, the training set in the temperature prediction time data set is used to train the bidirectional long short-term memory neural network, including:

[0063] S401: Determine the time step and overlapping time points of the bidirectional long short-term memory neural network.

[0064] S402: Using the training set in the temperature prediction time data set to train a bidirectional long short-term memory neural network, extracting temperature change features in the time dimension.

[0065] S403: Use the validation set bidirectional long short-term memory neural network in the temperature prediction time data set for verification to obtain a predicted value, and compare it with the validation set to obtain a residual sequence.

[0066] In one embodiment, the time characteristics of the temperature data are first analyzed to determine the periodicity and trend of the temperature change. According to the frequency of data collection and the requirements of the prediction task, an appropriate time step is selected to capture the periodic change characteristics of the data. Then, the overlapping time points, that is, the overlapping parts of each sliding window, are determined. The overlapping time points are selected without increasing the computational complexity of the model to ensure that the model can capture subtle changes in the time series.

[0067] In one embodiment, a training set in a temperature prediction time data set is used to train a bidirectional long short-term memory neural network. The bidirectional long short-term memory neural network model is divided into a forward long short-term memory neural network and a reverse long short-term memory neural network. The model divides the input temperature prediction time data set into several time step slices for traversal.

[0068] In one embodiment, the bidirectional long short-term memory neural network mines the characteristics of temperature change in the time dimension by continuously updating the shared weights of the forward long short-term memory neural network and the reverse long short-term memory neural network. The use of a bidirectional long short-term memory neural network can extract temperature change characteristics from both the forward and reverse directions, thereby making more accurate temperature predictions.

[0069] In one embodiment, the validation set in the temperature prediction time data set is used to validate the bidirectional long short-term memory neural network, the obtained predicted value is compared with the actual value in the validation set, and the residual sequence is calculated. The residual sequence is then analyzed to evaluate the prediction performance of the model.

[0070] like Figure 5 As shown, according to the extraction of temperature features by the deep neural network, the attention mechanism method is used to weight the temperature features of different time steps to enhance the differences between the temperature features at each time point in the time series, including:

[0071] S501: Get the final prediction value, use the local attention mechanism to weight the temperature features of each time step, differentiate the weights of temperature features at different time steps, and enhance the differences in temperature changes in the time series.

[0072] S502: The global attention mechanism is used to weight the fusion features of the entire time series, which effectively balances the importance of different time points in the time series and improves the accuracy and stability of temperature prediction.

[0073] In one embodiment, since the attention mechanism method can weight the temperature features in the time dimension and has the advantage of plug-and-play, we introduced an attention mechanism module into the temperature time series prediction model to enhance the model's ability to extract temperature change features.

[0074] In one embodiment, by fusing the local attention mechanism and the global attention mechanism, we weight the temperature features of each time step in the time series and the fused features of the entire time series. This method effectively increases the difference in temperature changes at different time points, further improves the accuracy and stability of temperature prediction, and obtains key deep features, thereby improving the efficiency and accuracy of the prediction model.

[0075] like Figure 6 As shown, the residual sequence is used to train the autoregressive integrated moving average model, and the corrected residual is added to the predicted value of the validation set to obtain the final predicted value, which includes:

[0076] S601: Determine the parameters q, p and d of the autoregressive integrated moving average model through autocorrelation, partial autocorrelation and stationarity judgment;

[0077] S602: using the residual sequence to modify the autoregressive integrated moving average model, and extracting residual features in the time dimension;

[0078] S603: Add the corrected residual to the predicted value of the validation set to obtain the final predicted value. In this way, the model can not only capture the basic trend and periodic changes in the time series, but also further correct the errors in the prediction and improve the accuracy of the prediction.

[0079] In one embodiment, a residual sequence is used to train an autoregressive integrated moving average model. First, the parameters q, p and d of the autoregressive integrated moving average model are determined by autocorrelation, partial autocorrelation and stationarity judgment; then, the autoregressive integrated moving average model is trained using the residual sequence, and residual features are extracted and corrected in the time dimension to improve the accuracy of the prediction.

[0080] In one embodiment, the corrected residual is added to the predicted value of the validation set to obtain the final predicted value. In this way, the model can not only capture the basic trend and periodic changes in the time series, but also further correct the errors in the prediction, thereby improving the accuracy and reliability of the prediction.

[0081] like Figure 7 As shown, a prediction model is established based on the temperature time series, including:

[0082] S701: training and verifying the model according to the temperature time series;

[0083] S702: Save the parameters obtained from the network model, establish a temperature prediction model, and use a test set to evaluate the model.

[0084] In one embodiment, model training and verification are performed based on the temperature time series characteristics, and a test set is used to establish the effectiveness of the temperature prediction model.

Claims

1. A method for predicting the temperature of an industrial circulating water system, characterized in that: The method comprises: S1: Obtain the original operating data of the circulating water system and blast furnace system, and collect various operating parameters closely related to temperature changes, including but not limited to flow rate, pressure, ambient temperature, etc., to ensure the comprehensiveness and accuracy of the data; S2: Clean and preprocess the acquired raw data, remove outliers and missing values, and perform standardization to obtain a clean and standardized temperature prediction time data set to ensure the effectiveness of model training; S3: Use deep learning feature reduction method based on fuzzy inconsistency measurement to reduce redundant variables in the model. Weigh the importance of deep learning features and remove some redundant features to obtain a subset of key deep learning features. S4: Input the processed temperature prediction time data set into the bidirectional long short-term memory neural network model for training and verification. The bidirectional long short-term memory neural network model can extract bidirectional long-term dependency features from time series data, thereby improving the prediction performance of the model; S5: Introducing the attention mechanism into the bidirectional long short-term memory neural network model, the local attention mechanism and the global attention mechanism are used to enhance the model's ability to capture long-term dependencies, so that the model can better focus on important features and improve prediction accuracy. S6: Compare the prediction results of the bidirectional long short-term memory neural network model with the actual data and calculate the prediction residual. Then use the autoregressive integrated moving average model to correct the residual sequence to further improve the accuracy and reliability of the prediction; S7: Based on the training and correction results, the final circulating water system temperature prediction model is established, and the model is predicted and evaluated. The model can accurately predict the temperature changes of the industrial circulating water system, provide a scientific decision-making basis for system operation and maintenance, and ensure the efficient and stable operation of the system.

2. The method for predicting temperature of an industrial circulating water system according to claim 1, characterized in that: The raw data is cleaned and normalized to obtain standardized time series data, including: firstly, a detailed analysis of the system operation mechanism is performed to manually select variables that are not related to the predicted value; secondly, according to the various operating parameters collected from the circulating water system and the blast furnace system, after data cleaning, filtering and other signal preprocessing methods, a data normalization method is used to convert the one-dimensional discrete time series data into a standardized numerical range. The normalization processing formula is: Where X is the original data, X norm is the normalized data, X max and X min are the maximum and minimum values ​​of the discrete points respectively.

3. The method for predicting the temperature of an industrial circulating water system according to claim 2, characterized in that: Performing data enhancement processing on the time series data includes: performing data enhancement on the preprocessed time series data, increasing the diversity and quantity of the data by re-screening and processing the data, so as to enhance the generalization ability of the model. Data enhancement methods include but are not limited to sliding window method, time slicing method and random sampling method. The deep learning feature reduction method based on fuzzy inconsistency measurement according to claim 1 is characterized in that, during the feature reduction process, the samples in the inconsistent set will always remain inconsistent with the original samples, and the number of sample inconsistent sets will not decrease. Define discrete samples x k The inconsistent set under feature u∈Q is denoted as Conflict u (x k ), defined as follows: Conflict u (x k )={∪x1} st.Conflict u (x k ,x1)∈(0,1) The smaller the fuzzy inconsistency between samples, the smaller the conflict between samples and the lower the redundancy in the sample set. In this process, the decision value of the sample is not changed. By continuously adding features for comparison, the inconsistency of the features is gradually reduced, and finally a feature subset without redundancy is obtained.

4. The method for predicting the temperature of an industrial circulating water system according to claim 1, characterized in that: The simplified data is input into the bidirectional long short-term memory neural network model for training and verification, which includes: using the bidirectional long short-term memory neural network model to capture the characteristics of long-term dependencies, extracting the characteristics of temperature changes from the time dimension, and training and verifying it to obtain preliminary prediction results. The basic structural formula of the bidirectional long short-term memory neural network model is: h t =σ(W h ·[h t-1 ,x t ]+b h ) y t =σ(W y ·h t +b y ) where h t is the hidden state, x t is the input, y t is the output, W h and W y is the weight matrix, b h and b y is the bias and σ is the activation function.

5. The method for predicting the temperature of an industrial circulating water system according to claim 1, characterized in that: The attention mechanism is used to enhance the prediction ability of the model, including: introducing two modules, local attention mechanism and global attention mechanism, into the bidirectional long short-term memory neural network model to enhance the model's ability to focus on important features and improve the accuracy of temperature prediction. The calculation formula of the attention mechanism is: Among them, Q is the query matrix, K is the key matrix, V is the value matrix, and d k The dimension of the key.

6. The method for predicting the temperature of an industrial circulating water system according to claim 1, characterized in that: The prediction residual sequence is corrected, which includes: using an autoregressive integrated moving average model to correct the prediction residual of the bidirectional long short-term memory neural network model to further improve the accuracy and reliability of the prediction. The expression of the autoregressive integrated moving average model is: Y t =φ1Y t-1 +φ2Y t-2 +...+φ p Y t-p +θ1ε t-1 +θ2ε t-2 +...+θ q e t-q +e t Among them, Y t is the time series, φ i is the autoregressive parameter, θ i is the moving average parameter, ε t is the white noise term.

7. The method for predicting the temperature of an industrial circulating water system according to claim 1, characterized in that: Based on the corrected results, a temperature prediction model is established, which includes: integrating the temperature prediction results corrected by the bidirectional long short-term memory neural network and the autoregressive integrated moving average model to establish the final circulating water system temperature prediction model for predicting the temperature changes of the industrial circulating water system. The comprehensive expression of the model is: in, is the final prediction result, y t (pred) is the prediction result of the bidirectional long short-term memory neural network model, Y t (residual) is the correction result of the residual by the autoregressive integrated moving average model.

8. The method for predicting the temperature of an industrial circulating water system according to claim 1, characterized in that: The prediction model is used for practical applications, including: according to the established temperature prediction model, the temperature change of the industrial circulating water system is predicted in real time, providing a scientific decision-making basis for system operation and maintenance, and ensuring the efficient and stable operation of the system. Through the prediction results of the model, the system operation parameters can be adjusted in time, the system performance can be optimized, potential failures can be prevented, and the safety and economic benefits of the system can be improved.

Citation Information

Cited By

  • Photovoltaic power generation power prediction model training method, prediction method, device, equipment and medium

    CN119853033A

  • Blast furnace molten iron temperature prediction method, device, equipment and medium

    CN122133425A

  • Industrial circulating water system multi-target dynamic collaborative optimization method based on Bi-LSTM-PINNs-ADMM

    CN122362906A

  • A Multi-Objective Dynamic Co-optimization Method for Industrial Circulating Water Systems Based on Bi-LSTM-PINNs-ADMM

    CN122362906B