Multi-element load prediction method and system for electric-thermal-hydrogen multi-energy coupling system

Abstract

The invention provides a multi-element load prediction method for an electric-thermal-hydrogen multi-energy coupling system, and the method comprises the following steps: S1, obtaining data, and constructing a multi-element load prediction feature data set; s2, according to the feature data set, a Spearman correlation analysis method is adopted to analyze correlation between data, abnormal data is detected and reconstructed by adopting a variational auto-encoder algorithm, and reconstruction of the whole feature data set is completed; s3, dividing the data set into a training set, a test set and a verification set; s4, constructing a multi-layer stacked BiLSTM multi-element load prediction model combined with a multi-head probability sparse self-attention mechanism; and S5, introducing the training set and the verification set into the combined model for training to obtain optimal model parameters, and then inputting the test set into the combined model to complete multi-element load prediction. According to the method, the coupling characteristic between data is considered, the key elements can be extracted from the characteristic data set of the electrothermal hydrogen multi-energy coupling system, and the prediction precision and the calculation efficiency can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of energy system engineering technology, and in particular to a multi-element load prediction method and system for an electric-thermal-hydrogen multi-energy coupling system. Background Art

[0002] With the continuous growth of global energy demand and the improvement of environmental protection awareness, the optimization and intelligent management of energy systems have become a hot topic in current research. Traditional energy systems usually operate independently and lack coordination and complementarity, resulting in low energy efficiency and waste of resources. In this context, multi-energy coupling systems have emerged. Multi-energy coupling systems organically combine and coordinate multiple energy forms such as electrical energy, thermal energy and hydrogen energy, give full play to the advantages of various energy sources, and improve the overall efficiency and sustainability of the energy system.

[0003] In a multi-energy coupling system, accurate load forecasting is crucial for the optimal operation and stable scheduling of the system. Load forecasting not only needs to consider the power load, but also the heat load and hydrogen load, which makes the forecasting task more complicated. Traditional load forecasting methods are mainly based on a single energy form and are difficult to meet the needs of multi-energy coupling systems. In addition, the load characteristics of multi-energy coupling systems are affected by many factors such as meteorological conditions. How to effectively analyze and utilize these coupling relationships to improve forecasting accuracy is a difficult problem that needs to be solved urgently.

[0004] In recent years, with the development of big data and artificial intelligence technology, deep learning methods have shown great potential in the field of load forecasting, especially technologies such as long short-term memory networks and self-attention mechanisms, which have significant advantages in processing time series data and extracting key features. However, a single model still has certain limitations when facing complex multi-element load forecasting tasks, such as being unable to fully capture the intrinsic coupling relationship and spatial characteristics between multiple energy loads.

[0005] Therefore, there is an urgent need for a multi-element load forecasting method and system for an electric-thermal-hydrogen multi-energy coupling system to solve the above problems. Summary of the invention

[0006] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a multi-element load forecasting method and system for an electric-thermal-hydrogen multi-energy coupling system to solve the problems raised in the background technology.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is: a multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system, comprising the following steps: S1. Construct characteristic data set: obtain multi-element load and meteorological historical data of the electric-thermal-hydrogen multi-energy coupling system, so as to construct a multi-element load prediction characteristic data set; S2. Reconstruct the data: Based on the feature data set, the Spearman correlation analysis method is used to analyze the correlation between the data, and the strongly correlated features are selected. For abnormal data, the variational autoencoder algorithm is used to detect and reconstruct it, thereby completing the reconstruction of the entire feature data set; S3, data division: divide the data set into training set, test set and validation set; S4. Construct a multivariate load forecasting model: Construct a multivariate load forecasting model that combines a multi-head probabilistic sparse self-attention mechanism with a multi-layer stacked BiLSTM to enhance the accuracy of multivariate load forecasting; S5. Multi-element load forecasting: The training set and validation set are introduced into the combined model for training to obtain the optimal model parameters, and then the test set is input into the combined model to complete the multi-element load forecasting of electricity, heat, and hydrogen. Preferably, the step S1 is specifically to collect historical data of system operation in order to explore the deep coupling and multi-energy coordination characteristics of multiple energy sources and comprehensively consider the influence of meteorological factors such as temperature and humidity on load characteristics; including key variables such as electric load, thermal load, hydrogen load, etc., and also to collect meteorological data such as temperature, humidity, wind speed, and sunshine intensity, so as to construct a multi-element load prediction feature data set.

[0008] Preferably, the step S2 is specifically to select strongly correlated features using the Spearman analysis method for the historical data collected in S1. The Spearman correlation does not assume that the two sets of data are sampled from a normal distribution, and takes a value of [-1, 1]. The larger the value, the higher the correlation. The calculation formula is:

[0009] in, is the correlation coefficient, n is the sample size, , , They are the ranks of the two variables sorted by size.

[0010] Preferably, in step S2, abnormal data detection and reconstruction: a variational autoencoder algorithm is used to complete abnormal data detection and reconstruction, which mainly includes data preprocessing, building a variational autoencoder model, training a variational autoencoder model, abnormal data detection and data reconstruction.

[0011] Preferably, in step S3, the training set contains known inputs and corresponding outputs. The model adjusts internal parameters by learning the patterns and relationships in these data to minimize the loss function. The model learns features and rules from the training set and continuously adjusts model parameters through an optimization algorithm. The training process includes multiple iterations. The model uses the training set data for forward propagation and back propagation in each iteration to update weights and biases.

[0012] Preferably, the validation set in step S3 does not participate in the training of the model, but is used to evaluate the performance of the model during the training process to prevent the model from overfitting the training data. During the training process, the model is evaluated on the validation set after each training cycle or several cycles, and the loss and other performance indicators are calculated. The hyperparameters of the model, including the learning rate, regularization parameters, network structure, etc., can be adjusted through the evaluation results of the validation set.

[0013] Preferably, the test set in step S3 is used to finally evaluate the performance of the model, and is started after the model training and tuning are completed to provide the performance of the model on unseen data and evaluate the generalization ability of the model.

[0014] Preferably, the specific steps of constructing the prediction model in step S4 are: S4.1. Define the input layer of the model and set the input shape according to the feature dimension of the data; S4.2, introduce a multi-head self-attention mechanism to enable the model to focus on different parts of the load data, thereby capturing more temporal dependencies; S4.3. Based on multi-head self-attention, we increase sparsity and use the probabilistic sparsity mechanism to focus only on important attention heads, thus reducing computational complexity. S4.4. Use BiLSTM layers to enable the model to simultaneously learn data dependencies from the past to the future and from the future to the past. S4.5. Two more layers of BiLSTM are stacked on the basis of the first layer to capture deeper temporal features.

[0015] Preferably, the specific steps of step S5 are: S5.1. Select the optimizer and loss function. Use the Adam optimizer and mean square error loss function. S5.2. Use the training set data to train the model. During the process, monitor the loss changes of the validation set to prevent overfitting. Use the early stopping method to automatically stop training. When the loss of the validation set no longer decreases, restore to the best model. S5.3. Evaluate model performance on the test set and calculate mean square error or other evaluation metrics to measure the prediction accuracy of the model. S5.4. Analyze the performance of the model on the test set, check the difference between the predicted results and the actual values, and evaluate the generalization ability of the model.

[0016] In addition, the present invention also discloses a multi-element load forecasting system for an electric-thermal-hydrogen multi-energy coupling system, including a data acquisition module, a data reconstruction module, a data processing module, a model building and training module, and a forecasting and result analysis module; The data acquisition module collects the electricity, heat, hydrogen load data and meteorological historical data of the electric-heat-hydrogen multi-energy coupling system; the data reconstruction module preprocesses the acquired raw data, including cleaning, standardization and data reconstruction, to ensure data quality and provide high-quality feature input for model training; the data processing module divides the data set into training set, validation set and test set; the model construction and training module constructs and trains a multi-layer stacked long short-term memory neural network model, and introduces a multi-head probabilistic sparse self-attention mechanism to enhance the model's prediction ability; the prediction and result analysis module uses the constructed prediction model to predict future electricity, heat and hydrogen loads, visualizes the prediction results, generates reports, and provides support for decision-making.

[0017] The present invention has the following beneficial effects: 1. The present invention takes the electric-heat-hydrogen multi-energy coupling system as the research object, predicts the electric-heat-hydrogen multi-energy load of the system, and constructs a multi-energy load prediction system to realize the visualization of the results and provide support for system decision-making; 2. The present invention adopts the Spearman correlation analysis method to analyze the intrinsic coupling relationship between the electric thermal hydrogen multi-load and the meteorological factors, which can screen the strong correlation features and ensure the high quality of the feature set; 3. The present invention uses a variational autoencoder to detect and reconstruct outliers in the feature set to complete the construction of feature engineering; 4. The present invention constructs a multi-layer stacked BiLSTM prediction model, which mines the characteristics of multi-dimensional data periodicity and correlation of the electric-thermal-hydrogen multi-energy coupling system, and realizes the capture of data time and space characteristics; 5. The present invention introduces a multi-head probabilistic sparse self-attention mechanism into the multi-layer stacked BiLSTM prediction model to further enhance the network's spatial feature extraction capability and robustness; 6. The prediction system of the present invention can automatically detect and reconstruct data based on the input characteristic information, and input it into the prediction module to complete multivariate load prediction and visualization of the results, with high flexibility and practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is an overall block diagram of a multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system; Figure 2 It is a detailed flow chart of multi-element load forecasting of the present invention; Figure 3 is a block diagram of a multi-element load forecasting system of the present invention; Figure 4 Schematic diagram of the VAE structure of the present invention Figure 5 It is the LSTM structure diagram of the present invention; Figure 6 It is the BiLSTM structure diagram of the present invention; Figure 7 This is the attention weight calculation diagram of the present invention. DETAILED DESCRIPTION

[0019] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: See also Figures 1 to 7 A multivariate load forecasting method for an electric-thermal-hydrogen multi-energy coupling system includes the following steps: S1. Construct characteristic data set: obtain multi-element load and meteorological historical data of the electric-thermal-hydrogen multi-energy coupling system, so as to construct a multi-element load prediction characteristic data set; S2. Reconstruct the data: Based on the feature data set, the Spearman correlation analysis method is used to analyze the correlation between the data, and the strongly correlated features are selected. For abnormal data, the variational autoencoder algorithm is used to detect and reconstruct it, thereby completing the reconstruction of the entire feature data set; S3, data division: divide the data set into training set, test set and validation set; S4. Construct a multivariate load forecasting model: Construct a multivariate load forecasting model that combines a multi-head probabilistic sparse self-attention mechanism with a multi-layer stacked BiLSTM to enhance the accuracy of multivariate load forecasting; S5. Multi-element load forecasting: The training set and validation set are introduced into the combined model for training to obtain the optimal model parameters, and then the test set is input into the combined model to complete the multi-element load forecasting of electricity, heat, and hydrogen. Preferably, the step S1 is specifically to collect historical data of system operation in order to explore the deep coupling and multi-energy coordination characteristics of multiple energy sources and comprehensively consider the influence of meteorological factors such as temperature and humidity on load characteristics; including key variables such as electric load, thermal load, hydrogen load, etc., and also to collect meteorological data such as temperature, humidity, wind speed, and sunshine intensity, so as to construct a multi-element load prediction feature data set.

[0020] Preferably, the step S2 is specifically to select strongly correlated features using the Spearman analysis method for the historical data collected in S1. The Spearman correlation does not assume that the two sets of data are sampled from a normal distribution, and takes a value of [-1, 1]. The larger the value, the higher the correlation. The calculation formula is:

[0021] in, is the correlation coefficient, n is the sample size, , , They are the ranks of the two variables sorted by size.

[0022] Preferably, in step S2, abnormal data detection and reconstruction: a variational autoencoder algorithm is used to complete abnormal data detection and reconstruction, which mainly includes data preprocessing, building a variational autoencoder model, training a variational autoencoder model, abnormal data detection and data reconstruction.

[0023] Preferably, in step S3, the training set contains known inputs and corresponding outputs. The model adjusts internal parameters by learning patterns and relationships in these data to minimize the loss function. The model learns features and rules from the training set, and continuously adjusts model parameters through optimization algorithms. The training process includes multiple iterations, and the model uses the training set data for forward propagation and back propagation in each iteration to update weights and biases. The scale and quality of the training set directly affect the learning effect of the model. The more data, the richer the features that the model can learn, and the stronger the generalization ability of the model. The data in the training set should be representative and cover various situations in the problem space so that the model can learn comprehensive features.

[0024] Preferably, the validation set in step S3 does not participate in the training of the model, but is used to evaluate the performance of the model during the training process to prevent the model from overfitting the training data. During the training process, the model is evaluated on the validation set after each training cycle or several cycles, and the loss and other performance indicators are calculated. The hyperparameters of the model can be adjusted through the evaluation results of the validation set, including the learning rate, regularization parameters, network structure, etc. The validation set can help detect whether the model is overfitting or underfitting. If the model performs well on the training set but performs poorly on the validation set, an overfitting problem may occur. The evaluation results of the validation set are used to select the final model architecture and hyperparameter settings to ensure the generalization ability of the model.

[0025] Preferably, the test set in step S3 is used to finally evaluate the performance of the model, and is started after the model training and tuning are completed to provide the performance of the model on unseen data and evaluate the generalization ability of the model.

[0026] Preferably, the specific steps of constructing the prediction model in step S4 are: S4.1. Define the input layer of the model and set the input shape according to the feature dimension of the data; S4.2, introduce a multi-head self-attention mechanism to enable the model to focus on different parts of the load data, thereby capturing more temporal dependencies; S4.3. Based on multi-head self-attention, we increase sparsity and use the probabilistic sparsity mechanism to focus only on important attention heads, thus reducing computational complexity. S4.4. Use BiLSTM layers to enable the model to simultaneously learn data dependencies from the past to the future and from the future to the past. S4.5. Two more layers of BiLSTM are stacked on the basis of the first layer to capture deeper temporal features. The prediction model consists of three parts: three layers of stacked BiLSTM, a multi-head probabilistic sparse self-attention mechanism layer, and a fully connected layer; The first layer of BiLSTM processes the input features and captures the bidirectional dependencies of the time series. The second layer of BiLSTM further extracts the time series features. The third layer of BiLSTM completes the final feature extraction. Its expression is:

[0027] The multi-head probabilistic sparse self-attention mechanism transforms the output of BiLSTM into query (Q), key (K), and value (V) matrices through linear transformation, then calculates the attention weights, introduces sparsity, controls the concentration of attention through probability distribution, and finally uses multiple attention heads to calculate in parallel, then concatenates the results and transforms them linearly. The expression is:

[0028]

[0029]

[0030]

[0031] in, is the learnable weight matrix, is the dimension of the key vector, is the weight matrix of the linear transformation; Finally, the features output by the attention mechanism are passed to the fully connected layer, which is used to map the features to load prediction values.

[0032] Preferably, the specific steps of step S5 are: S5.1. Select the optimizer and loss function. Use the Adam optimizer and mean square error loss function. S5.2. Use the training set data to train the model. During the process, monitor the loss changes of the validation set to prevent overfitting. Use the early stopping method to automatically stop training. When the loss of the validation set no longer decreases, restore to the best model. S5.3. Evaluate the model performance on the test set and calculate the mean square error or other evaluation indicators to measure the prediction accuracy of the model; S5.4, analyze the performance of the model on the test set, check the difference between the predicted results and the actual values, and evaluate the generalization ability of the model. S5 is load forecasting, and different evaluation indicators are used for analysis. The evaluation indicators mainly include: MSE, RMSE, and MAE; MSE measures the square error between the predicted value and the true value, RMSE is the square root of MSE, reflecting the standard deviation of the prediction error, and MAE measures the average of the absolute error between the predicted value and the true value.

[0033] In addition, the present invention also discloses a multi-element load forecasting system for an electric-thermal-hydrogen multi-energy coupling system, including a data acquisition module, a data reconstruction module, a data processing module, a model building and training module, and a forecasting and result analysis module; The data acquisition module collects the electricity, heat, hydrogen load data and meteorological historical data of the electric-heat-hydrogen multi-energy coupling system; the data reconstruction module preprocesses the acquired raw data, including cleaning, standardization and data reconstruction, to ensure data quality and provide high-quality feature input for model training; the data processing module divides the data set into training set, validation set and test set; the model construction and training module constructs and trains a multi-layer stacked long short-term memory neural network model, and introduces a multi-head probabilistic sparse self-attention mechanism to enhance the model's prediction ability; the prediction and result analysis module uses the constructed prediction model to predict future electricity, heat and hydrogen loads, visualizes the prediction results, generates reports, and provides support for decision-making.

[0034] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system, characterized by: The following steps are involved: S1. Construct characteristic data set: obtain multi-element load and meteorological historical data of the electric-thermal-hydrogen multi-energy coupling system, so as to construct a multi-element load prediction characteristic data set; S2. Reconstruct the data: Based on the feature data set, the Spearman correlation analysis method is used to analyze the correlation between the data, and the strongly correlated features are selected. For abnormal data, the variational autoencoder algorithm is used to detect and reconstruct it, thereby completing the reconstruction of the entire feature data set; S3, data division: divide the data set into training set, test set and validation set; S4. Construct a multivariate load forecasting model: Construct a multivariate load forecasting model that combines a multi-head probabilistic sparse self-attention mechanism with multi-layer stacked BiLSTM to enhance the accuracy of multivariate load forecasting; S5. Multi-element load forecasting: The training set and validation set are introduced into the combined model for training to obtain the optimal model parameters, and then the test set is input into the combined model to complete the multi-element load forecasting of electricity, heat, and hydrogen.

2. The multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system according to claim 1 is characterized in that: Specifically, the step S1 is to collect historical data of system operation in order to explore the deep coupling and multi-energy coordination characteristics of multiple energy sources and comprehensively consider the impact of meteorological factors such as temperature and humidity on load characteristics; including key variables such as electric load, thermal load, hydrogen load, etc., and also collect meteorological data such as temperature, humidity, wind speed, and sunshine intensity, so as to construct a multi-element load prediction feature data set.

3. The multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system according to claim 1 is characterized in that: The step S2 is specifically to select strongly correlated features using the Spearman analysis method for the historical data collected in S1. The Spearman correlation does not assume that the two sets of data are sampled from a normal distribution, and takes a value of [-1,1]. The larger the value, the higher the correlation. The calculation formula is:

4. Among them, is the correlation coefficient, n is the sample size, , , They are the ranks of the two variables sorted by size.

5. The multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system according to claim 1 is characterized in that: Abnormal data detection and reconstruction in step S2: Abnormal data detection and reconstruction are completed using a variational autoencoder algorithm, which mainly includes data preprocessing, building a variational autoencoder model, training a variational autoencoder model, abnormal data detection and data reconstruction.

6. The multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system according to claim 1 is characterized in that: In step S3, the training set contains known inputs and corresponding outputs. The model adjusts internal parameters by learning the patterns and relationships in these data to minimize the loss function. The model learns features and rules from the training set and continuously adjusts model parameters through optimization algorithms. The training process includes multiple iterations. The model uses the training set data for forward propagation and back propagation in each iteration to update weights and biases.

7. The multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system according to claim 1 is characterized in that: The validation set in step S3 does not participate in the training of the model, but is used to evaluate the performance of the model during the training process to prevent the model from overfitting the training data. During the training process, the model is evaluated on the validation set after each training cycle or several cycles to calculate the loss and other performance indicators. The hyperparameters of the model, including the learning rate, regularization parameters, network structure, etc., can be adjusted through the evaluation results of the validation set.

8. The multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system according to claim 1 is characterized in that: The test set in step S3 is used to finally evaluate the performance of the model. It is started after the model training and tuning are completed, and provides the performance of the model on unseen data to evaluate the generalization ability of the model.

9. The multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system according to claim 1 is characterized in that: The specific steps of constructing the prediction model in step S4 are: S4.

1. Define the input layer of the model and set the input shape according to the feature dimension of the data; S4.2, introduce a multi-head self-attention mechanism to enable the model to focus on different parts of the load data, thereby capturing more temporal dependencies; S4.

3. Based on multi-head self-attention, we increase sparsity and use the probabilistic sparsity mechanism to focus only on important attention heads, thus reducing computational complexity. S4.

4. Use BiLSTM layers to enable the model to simultaneously learn data dependencies from the past to the future and from the future to the past. S4.

5. Two more layers of BiLSTM are stacked on the basis of the first layer to capture deeper temporal features.

10. The multi-element load forecasting method for an electric-thermal-hydrogen multi-energy coupling system according to claim 1, characterized in that: The specific steps of step S5 are: S5.

1. Select the optimizer and loss function. Use the Adam optimizer and mean square error loss function. S5.

2. Use the training set data to train the model. During the process, monitor the loss changes of the validation set to prevent overfitting. Use the early stopping method to automatically stop training. When the loss of the validation set no longer decreases, restore to the best model. S5.

3. Evaluate model performance on the test set and calculate mean square error or other evaluation metrics to measure the prediction accuracy of the model. S5.

4. Analyze the performance of the model on the test set, check the difference between the predicted results and the actual values, and evaluate the generalization ability of the model.

11. A multi-element load forecasting system for an electric-thermal-hydrogen multi-energy coupling system, characterized in that: It includes data acquisition module, data reconstruction module, data processing module, model building and training module and prediction and result analysis module; The data acquisition module collects the electricity, heat, hydrogen load data and meteorological historical data of the electric-heat-hydrogen multi-energy coupling system; the data reconstruction module preprocesses the acquired raw data, including cleaning, standardization and data reconstruction, to ensure data quality and provide high-quality feature input for model training; the data processing module divides the data set into training set, validation set and test set; the model construction and training module constructs and trains a multi-layer stacked long short-term memory neural network model, and introduces a multi-head probabilistic sparse self-attention mechanism to enhance the model's prediction ability; the prediction and result analysis module uses the constructed prediction model to predict future electricity, heat and hydrogen loads, visualizes the prediction results, generates reports, and provides support for decision-making.

Images