Comprehensive energy system multi-element load prediction method based on graph attention mechanism and generative adversarial network

Abstract

The invention relates to an integrated energy system multi-element load prediction method based on a graph attention mechanism and a generative adversarial network, and belongs to the technical field of integrated energy system multi-element load prediction. According to the method, a new prediction framework based on GAN-GAT-TCN is provided, for strong interaction and strong coupling characteristics of multi-element load prediction of the integrated energy system, correlation analysis is carried out on preprocessed power, heat and cold load data and influence factors thereof by adopting cross correlation coefficients to obtain an adjacent matrix, and power, heat and cold loads are predicted by adopting GAN to obtain a prediction result. Wherein the GAT is used as a generator and can extract spatial dimension information among power, heat and cold load data and influence factors thereof, the TCN is used as a judgment device and can extract time dimension information in the power, heat and cold load data, a final prediction result is obtained through training in a mutual confrontation mode, the accuracy and robustness of multi-element load prediction are improved, and the prediction efficiency is improved. And a direction is provided for a multi-element load prediction technology of an integrated energy system.

Description

Technical Field

[0001] The present invention relates to a multi-element load forecasting method for an integrated energy system based on a graph attention mechanism and a generative adversarial network, and belongs to the technical field of multi-element load forecasting for an integrated energy system. Background Art

[0002] With the transformation of energy structure, building a comprehensive energy system with multiple energy complementarities has become a powerful measure to reduce system carbon emissions and improve energy efficiency. The comprehensive energy system integrates multiple energy sources and integrated energy storage technologies to achieve efficient energy planning and scheduling to meet users' multiple load demands such as cooling, heating, and electricity. However, the coupling of multiple energy sources and the access to new energy systems bring randomness and uncertainty to the load of the comprehensive energy system. Therefore, the accuracy of load forecasting is the basic condition for achieving optimal scheduling of the comprehensive energy system, which puts forward higher requirements for achieving accurate multiple load forecasting.

[0003] Scholars have done a lot of work on load forecasting. Traditional load forecasting refers to the prediction of a single type of load, and mostly uses mathematical statistics and machine learning methods. Yan TY uses time decomposition technology to apply linear regression to achieve accurate prediction of power load. Yang GH proposed a prediction method based on Holt-Winters exponential smoothing to improve the accuracy of load forecasting. Wu J developed a new prediction method based on support vector regression. Although statistical methods take into account the characteristics of time data, their effectiveness may be affected when facing nonlinear or non-stationary data due to the limitations of assumptions. Although machine learning can effectively solve the limitations of statistical methods in nonlinear mining, it cannot automatically learn the defects in sequence data and requires manual selection of input features, which may destroy the temporal integrity of the data.

[0004] In recent years, with the improvement of computer performance and the diversification of data types, deep learning has become the mainstream research method. It can extract high-dimensional abstract information from massive data layer by layer through multi-layer nonlinear mapping without building an accurate mathematical model. It also has strong generalization and nonlinear mapping capabilities. Memarzadeh G established a short-term load forecasting model based on the Long Short-Term Memory (LSTM) network by analyzing the dynamic characteristics and internal laws of the load. Kong W proposed a forecasting framework based on the LSTM recurrent neural network to overcome the high volatility and uncertainty of the load. However, a single model often has limitations. In order to improve the prediction accuracy, many scholars have tried to use a combination model to complement each other to improve the prediction effect. In order to improve the load prediction accuracy, Rafi SH proposed an integrated prediction method based on convolutional neural network and long short-term memory network. Hu W proposed a short-term load forecasting model based on optimized VMD-mRMR-LSTM, which effectively solves the shortcomings of traditional methods in considering the correlation of time series data and the comprehensiveness of eigenvalues. Yang XL constructed an LSTM neural network prediction model integrating the self-attention mechanism, deeply mined the characteristics of the load sequence, and significantly improved the load prediction accuracy.

[0005] The above literatures mostly focus on single load prediction, while the prediction of multiple loads in an integrated energy system is more complex and needs to consider the external factors of the integrated energy system as well as the transmission and mutual coupling between multiple internal energy flows. Therefore, Li C proposed a multi-load prediction method that integrates neural networks and transfer learning. The Pearson correlation coefficient is used to screen the main influencing factors, and the prediction model is established using convolutional neural networks and gated recurrent units. The maximum mean difference strategy is introduced to dynamically adjust the structure to adapt to complex environments. The results of the example show that this method significantly improves the prediction accuracy and model robustness. Liu H quantitatively analyzes the coupling relationship of multiple energy loads in an integrated energy system through a multivariate phase space reconstruction and support vector regression combined model, and achieves accurate prediction. Wang C proposed a multi-task learning model based on ResNet-LSTM network and attention mechanism to deeply explore the coupling relationship between multiple loads and improve the prediction accuracy. By introducing the attention mechanism, the shared features are differentiated for different subtasks, thereby realizing the joint prediction of multiple loads. The above literatures all explore the coupling relationship of multiple loads in an integrated energy system on a time scale, without considering the coupling relationship of multiple loads in space. At the same time, combining the temporal and spatial characteristics of multiple loads can improve the prediction accuracy. Summary of the invention

[0006] The purpose of the present invention is to provide a multi-element load forecasting method for an integrated energy system based on a graph attention mechanism and a generative adversarial network, aiming to solve the technical problem that it is difficult to ensure the accuracy and robustness of multi-element load forecasting under the spatial and temporal characteristics of multi-element load data.

[0007] In order to achieve the above objectives, the technical solution adopted by the present invention is: a method for multi-element load forecasting of an integrated energy system based on a graph attention mechanism and a generative adversarial network. This method targets the strong interaction and strong coupling characteristics of multi-element load data of an integrated energy system, and utilizes multi-energy interaction coupling characteristics and time series characteristics to establish a new forecasting framework GAN-GAT-TCN, which combines a graph attention convolutional network (GAT) for learning spatial feature representation and a time series convolutional network (TCN) for learning temporal feature representation, and generates adversarial learning between the GAT and TCN networks through a generative adversarial neural network (GAN) to generate data closer to the true value. This invention fully considers the spatial and temporal characteristics of multi-element load data, significantly improves the accuracy and robustness of multi-element load forecasting, and has high forecasting accuracy and good adaptability when dealing with multi-element load forecasting problems in integrated energy systems. The specific steps are:

[0008] Step 1: Preprocess the cooling, heating and power operation data of the integrated energy system, including data cleaning and data standardization, fill in the missing data in the original integrated energy load data, remove abnormal data, and perform normalization processing;

[0009] Step 2: Use the maximum mutual information coefficient (MIC) algorithm to perform correlation analysis on the variables of the cooling, heating and electricity loads of the integrated energy system and the factors affecting the load of the integrated energy system, calculate the correlation between the data, and obtain the MIC value;

[0010] Step 3: Select the influencing factor variables of the integrated energy system with high correlation as the factor screening set, and use the MIC value and the cooling, heating and electricity loads of the integrated energy system to form an adjacency matrix;

[0011] Step 4: Combine the cooling, heating and electricity load data of the integrated energy system and the screening set of integrated energy system factors, and divide them into training set, validation set and test set according to the proportion;

[0012] Step 5: Input the divided data into the GAN model, where GAT is used as the generator and TCN is used as the judge. The electricity, heat and cooling loads are trained and learned on the training set. The accuracy and reliability of the training model are verified on the validation set. Finally, the prediction results of electricity, heat and cooling loads are output on the test set.

[0013] Step 6: Use the evaluation index to calculate the error between the model prediction value and the actual value.

[0014] The Step 2 is specifically as follows:

[0015] Solar radiation, wind speed, dry bulb temperature, dew point temperature, humidity, and holiday data are selected as influencing factors of electric load, cooling load, and heating load data;

[0016] Divide the data of influencing factors of the integrated energy system and the cooling, heating and electricity operation data of the integrated energy system into grids;

[0017] The divided data is used as input to perform correlation analysis using the maximum mutual information coefficient MIC algorithm to obtain the mutual information value of the two groups of variables;

[0018] The mutual information value is used to measure the correlation between the influencing factor data of the integrated energy system and the cooling, heating and electricity operation data of the integrated energy system.

[0019] The Step 3 is specifically as follows:

[0020] Select the load of the integrated energy system and the variables affecting the load of the integrated energy system as nodes for constructing the graph;

[0021] The maximum mutual information coefficient value is used to represent the correlation between two nodes;

[0022] The first k variables are selected from large to small according to the MIC value to establish the adjacency relationship, forming an adjacency matrix with the maximum mutual information coefficient as the value between nodes.

[0023] The Step 5 is specifically as follows:

[0024] Using GAT as the generator and TCN as the discriminator, we simultaneously mine the spatiotemporal characteristics of multivariate data of integrated energy systems;

[0025] In the prediction process, the data set is input into the generator with GAT as the model to extract the spatial information of the data, and the prediction result of the generator is input into the discriminator and the true value for identification;

[0026] Analyze the true probability of the data, use the temporal convolution model used by the discriminator to extract the data time information, return to the generator training if it is identified as false, and output the prediction result if it is identified as true.

[0027] The Step 6 is specifically as follows:

[0028] The root mean square error RMSE, mean absolute error MAE and Nash efficiency coefficient NSE are used as evaluation indicators to analyze the model. The calculation formula is as follows:

[0029]

[0030] In the formula, is the predicted value of cooling, heating and electricity load of the comprehensive energy system;i It is the real value of cooling, heating and electricity load of the integrated energy system; It is the sample mean of cooling, heating and electricity load data of the integrated energy system.

[0031] The beneficial effects of the present invention are:

[0032] For spatial dimension information, the MIC algorithm is used to construct an adjacency matrix to connect the cooling, heating and electricity loads of the integrated energy system and the variables of related influencing factors of the integrated energy system. The graph attention network is used to learn the spatial relationship between multiple loads and their influencing factors. By introducing the attention mechanism, shared features are differentiated for different subtasks, considering the external factors of the integrated energy system and the transmission and mutual coupling between internal multiple energy flows, thereby realizing the joint prediction of multiple loads.

[0033] For time dimension information, a temporal convolutional network is used for extraction. The temporal convolutional network is mainly based on causal dilated convolution, which can extract the temporal characteristics of time series with a relatively small number of convolution layers. The present invention introduces GAN to realize the adversarial learning between GAT and TCN for spatiotemporal combined prediction. The integration of spatiotemporal information improves the prediction accuracy of the model to a certain extent, significantly improves the fitting effect of the actual value, and makes the fitting line closer to the ideal state. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic flow chart of the multi-element load forecasting method for an integrated energy system of the present invention. DETAILED DESCRIPTION

[0035] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] Example 1: Figure 1 As shown in the figure, a multi-element load forecasting method for an integrated energy system based on a graph attention mechanism and a generative adversarial network is provided, which specifically includes the following steps:

[0037] Step 1: Preprocess the cooling, heating and power operation data of the integrated energy system, including data cleaning and data standardization, fill in the missing data in the original integrated energy load data, remove abnormal data, and perform normalization processing;

[0038] Specifically, the linear interpolation method is used to fill the missing values. For the yth missing data that appears, the method of filling the missing data using the linear interpolation method is as follows:

[0039]

[0040] In the formula: m<y<m+l, m is the relative position of the missing data, l is the time interval of the missing data; T(y) is the filled integrated energy system load data; T(m+l) is the integrated energy system load data with the time sequence of m+l; T(m) is the integrated energy system load data with the time sequence of m.

[0041] Specifically, the data standardization is a normalization process, which normalizes the load data of the integrated energy system to eliminate the influence of the dimension. The formula is:

[0042]

[0043] In the formula, x is the original value of the data; is the normalized x; mean(X train ) represents the training set X train The average value of std(X train ) represents the training set X train The standard deviation of .

[0044] Step 2: Use the MIC algorithm to perform correlation analysis on the variables of the cooling, heating and electricity loads of the integrated energy system and the factors affecting the load of the integrated energy system, calculate the correlation between the data, and obtain the MIC value;

[0045] Step 2.1: Divide the data into grids;

[0046] Step 2.2: Calculate the mutual information value at different grid resolutions. The calculation steps are as follows:

[0047] Step 2.2.1: The integrated energy system load variable X is a discrete sequence of time series, and its value is X1, X2, …, X according to the time step interval. n , information entropy h(X) is used to measure the uncertainty of the load variable X of the integrated energy system. The information entropy h(X) is shown as follows:

[0048]

[0049] P(X1),P(X2),…,P(X n ) is the probability distribution of the load variable X of the integrated energy system.

[0050] Step 2.2.2: The joint probability of the integrated energy system load variable X and the integrated energy system influencing factor variable Y is recorded as P(X, Y), that is, the probability that the two variables are associated. The joint entropy is used to describe the degree of association between the integrated energy system load variable X and the integrated energy system influencing factor variable Y. The larger the joint entropy, the lower the degree of association between X and Y, and the higher the uncertainty. The joint entropy is shown as follows:

[0051]

[0052] Step 2.2.3: For the discrete integrated energy system load variable X and the integrated energy system influencing factor variable Y, the conditional entropy is used to describe the uncertainty of the other random variable when one of the random variables is fixed. The conditional entropy is shown as follows:

[0053]

[0054] Where: H(Y|X) is the unknown information of the comprehensive energy system influencing factor variable Y remaining when the comprehensive energy system load variable X is known.

[0055] The relationship between joint entropy and conditional entropy is as follows:

[0056] H(X,Y)=H(X)+H(Y|X)=H(Y)+H(X|Y)

[0057] The mutual information between the integrated energy system load variable X and the integrated energy system influencing factor variable Y can measure the amount of information reduced by a random variable due to the knowledge of another random variable, which is expressed by the following formula:

[0058]

[0059] Step 2.3: Standardize the mutual information value. The standardization formula is as follows:

[0060]

[0061] Where: I(X;Y) st is the standardized mutual information value, |X| and |Y| are the grid numbers occupied by the integrated energy system load variable X and the integrated energy system influencing factor variable Y, respectively.

[0062] Step 2.4: Select the largest element from the feature matrix as MIC.

[0063] Step 3: Select the influencing factor variables of the integrated energy system with high correlation as the factor screening set, and use the MIC value and the cooling, heating and electricity loads of the integrated energy system to form an adjacency matrix;

[0064] Specifically, an adjacency matrix is ​​constructed, and the variables affecting the integrated energy system with high correlation are selected as the factor screening set. The connection relationship between each variable is determined by MIC, and the size of the mutual information value between nodes determines the construction of the edge between nodes. First, the load and the variables affecting the load are selected as the construction graph G. var (V,E var )'s node V; E varIndicates the correlation between two nodes; select k variables with larger MIC values ​​to establish adjacency relationships and construct the adjacency matrix G var as follows:

[0065]

[0066] Where: It is the mutual information value between the integrated energy system load variable X and the integrated energy system influencing factor variable Y.

[0067] Step 4: Combine the cooling, heating and electricity load data of the integrated energy system and the screening set of integrated energy system factors, and divide them into training set, validation set and test set according to the proportion;

[0068] Specifically, the cooling, heating and electricity load data of the integrated energy system and the screening set of integrated energy system factors can be divided into a training set, a validation set and a test set in a ratio of 7:2:1;

[0069] Step 5: Input the divided data into the GAN model, where GAT is used as the generator and TCN is used as the judge. The electricity, heat and cooling loads are trained and learned on the training set. The accuracy and reliability of the training model are verified on the validation set. Finally, the prediction results of electricity, heat and cooling loads are output on the test set.

[0070] Step 5.1: Generative adversarial network uses the same discriminator loss function J D , the difference between different networks lies in the generator loss function J G , the discriminator loss function J D :

[0071]

[0072] Where G is the definition function of the generator; z is the input of the generator, including the load data of the integrated energy system, the variables of the influencing factors of the integrated energy system and their MIC values; β G is the parameter of the generator; D is the definition function of the discriminator; x is the input of the discriminator, which contains the heating, cooling and electricity load data of the integrated energy system trained by the generator; β D is the parameter of the generator; xP data Represents the statistical distribution P of the real sample data X of the cooling, heating and electricity load of the integrated energy system data , that is, x belongs to the real sample data; D(x) represents the discriminator input function; G(z) represents the generator input function.

[0073] Step 5.2: The generator continuously inputs data, and the discriminator continuously detects data, and the two ends repeatedly confront each other. The sum of the loss function of the generator and the discriminator is 0, that is:

[0074] J G=-J D

[0075] Generator function value J G The smaller the better. According to the above formula, we can know that the function value J G The smaller the discriminator function value J D The value function of the zero-sum game is:

[0076] V F (β D ,β G )=-J D (β D ,β G )

[0077] Then the optimal solution of the generator is:

[0078] β G* =argminmaxV F (β D ,β G )

[0079] The optimization objective function of the generated GAN network is obtained as:

[0080]

[0081] In the formula, zP z (z) means z conforms to the statistical distribution P of the code z .

[0082] Step 5.3: Input the integrated energy system load data, the integrated energy system influencing factor variables and their MIC values ​​into the generative adversarial neural network to obtain the generated data G(z).

[0083] Step 5.4: Input the real data x of the cooling, heating and electricity load of the integrated energy system and the generated data G(z) into the discriminator model for identification, analyze the true probability of these data, first take 50% of the data as real data, and the other 50% of the data as pseudo data. The purpose is to make the final output data close to the real data, and obtain an output D(x) close to 1 and an output D(G(z)) close to 0.

[0084] Step 5.5: Update the generator model parameters through the objective function and optimize the generator model.

[0085] Step 5.6: Update the discriminator model parameters through the objective function and optimize the discriminator model.

[0086] Step5.7: Repeat steps Step5.4-Step5.6 until the discriminator outputs D(x) and D(G(z)) converge to 0.5, then the training is completed.

[0087] Step 6: Use the evaluation index to calculate the error between the model prediction value and the actual value;

[0088] The root mean square error RMSE, mean absolute error MAE and Nash efficiency coefficient NSE are used as evaluation indicators to calculate the error between the model prediction value and the actual value, as shown below:

[0089]

[0090] In the formula, is the predicted value of cooling, heating and electricity load of the comprehensive energy system; i It is the real value of cooling, heating and electricity load of the integrated energy system; It is the sample mean of the cooling, heating and electricity load data of the integrated energy system. RMSE most directly represents the prediction accuracy of the model, MAE considers the allowable error in practical application, and NSE is the controllability of the model prediction output under the input variables.

[0091] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A multi-element load forecasting method for an integrated energy system based on graph attention mechanism and generative adversarial network, characterized in that: The following steps are involved: Step 1: Preprocess the cooling, heating and power operation data of the integrated energy system, including data cleaning and data standardization, fill in the missing data in the original integrated energy load data, remove abnormal data, and perform normalization processing; Step 2: Use the MIC algorithm to perform correlation analysis on the variables of the cooling, heating and electricity loads of the integrated energy system and the factors affecting the load of the integrated energy system, calculate the correlation between the data, and obtain the MIC value; Step 3: Select the influencing factor variables of the integrated energy system with high correlation as the factor screening set, and use the MIC value and the cooling, heating and electricity loads of the integrated energy system to form an adjacency matrix; Step 4: Combine the cooling, heating and electricity load data of the integrated energy system and the screening set of integrated energy system factors, and divide them into training set, validation set and test set according to the proportion; Step 5: Input the divided data into the GAN model, where GAT is used as the generator and TCN is used as the judge. The electricity, heat and cooling loads are trained and learned on the training set. The accuracy and reliability of the training model are verified on the validation set. Finally, the prediction results of electricity, heat and cooling loads are output on the test set. Step 6: Use the evaluation index to calculate the error between the model prediction value and the actual value.

2. According to claim 1, a multi-element load forecasting method for an integrated energy system based on a graph attention mechanism and a generative adversarial network is characterized in that: The Step 2 is specifically as follows: Solar radiation, wind speed, dry bulb temperature, dew point temperature, humidity, and holiday data are selected as influencing factors of electric load, cooling load, and heating load data; Divide the data of influencing factors of the integrated energy system and the cooling, heating and electricity operation data of the integrated energy system into grids; The divided data is used as input to perform correlation analysis using the maximum mutual information coefficient MIC algorithm to obtain the mutual information value of the two groups of variables; The mutual information value is used to measure the correlation between the influencing factor data of the integrated energy system and the cooling, heating and electricity operation data of the integrated energy system.

3. According to claim 1, a multi-element load forecasting method for an integrated energy system based on a graph attention mechanism and a generative adversarial network is characterized in that: The Step 3 is specifically as follows: Select the load of the integrated energy system and the variables affecting the load of the integrated energy system as nodes for constructing the graph; The maximum mutual information coefficient value is used to represent the correlation between two nodes; The first k variables are selected from large to small according to the MIC value to establish the adjacency relationship, forming an adjacency matrix with the maximum mutual information coefficient as the value between nodes.

4. According to claim 1, a multi-element load forecasting method for an integrated energy system based on a graph attention mechanism and a generative adversarial network is characterized in that: The Step 5 is specifically as follows: Using GAT as the generator and TCN as the discriminator, we simultaneously mine the spatiotemporal characteristics of multivariate data of integrated energy systems; In the prediction process, the data set is input into the generator with GAT as the model to extract the spatial information of the data, and the prediction result of the generator is input into the discriminator and the true value for identification; Analyze the true probability of the data, use the temporal convolution model used by the discriminator to extract the data time information, return to the generator training if it is identified as false, and output the prediction result if it is identified as true.

5. According to claim 1, a multi-element load forecasting method for an integrated energy system based on a graph attention mechanism and a generative adversarial network is characterized in that: The Step 6 is specifically as follows: The root mean square error RMSE, mean absolute error MAE and Nash efficiency coefficient NSE are used as evaluation indicators to analyze the model. The calculation formula is as follows: In the formula, is the predicted value of cooling, heating and electricity load of the comprehensive energy system; i It is the real value of cooling, heating and electricity load of the integrated energy system; It is the sample mean of cooling, heating and electricity load data of the integrated energy system.

Images