Multi-task collaborative prediction method and system for energy load price in integrated energy systems

Abstract

The present invention provides a multi-task collaborative prediction method and system for energy load prices in an integrated energy system. This method includes the steps of analyzing the spatio-temporal coupling characteristics between various uncertainty factors affecting the integrated energy system from the perspectives of both cross-correlation and auto-correlation, performing feature extraction on the results of the two correlation analyses, performing secondary extraction on the extracted features by means of a channel attention and sequential attention mechanism, and performing feature sharing on the data after secondary extraction by means of a two-layer collaborative prediction model. In the two-layer collaborative prediction model, the electricity price prediction is the main task, and the energy and load predictions are the secondary tasks. A hard sharing mechanism is used for the coupling between the corresponding loads of each task, and a soft sharing mechanism is used between different tasks to share information from the secondary task to the main task to obtain a collaborative prediction result. The present invention can perform correlation analysis and collaborative prediction on the two uncertainties of the energy aspect and the price aspect in a situation with many uncertainties in the integrated energy system.

Description

Technical Field

[0001] (Cross - reference to related applications) This invention claims the priority of a Chinese patent application with the application number 202310519941.5, titled "Multi - task Joint Prediction Method and System for Energy Load Price in an Integrated Energy System", which was filed with the China National Intellectual Property Administration on May 6, 2023. All of its content is incorporated into this invention by reference for all purposes and constitutes a part of this invention.

[0002] This invention belongs to the technical field of integrated energy system planning and relates to a multi - task joint prediction method and system for energy load price in an integrated energy system.

Background Art

[0003] The description of this part merely provides background technical information related to this invention and does not necessarily constitute prior art.

[0004] In an integrated energy system, the penetration rate of new energies such as wind power generation and solar power generation on the energy supply side has been gradually increasing. On the energy utilization side, a large number of flexible loads such as electric vehicles and energy storage batteries have been connected. Also, due to the recommendation of participation in the competitive trading of the energy market, the private capital participating has increased. Thus, the uncertainty factors in the system have significantly increased across several parts such as energy generation, trading, and consumption. Accurate prediction of uncertainty is a prerequisite for ensuring the efficient operation of the integrated energy system.

[0005] In conventional prediction methods, the three uncertainties of new energy power on the energy supply side, the load on the energy utilization side, and the energy price on the market side are often predicted independently. The model is simple and the method is relatively mature, and results such as point prediction, interval prediction, and probability prediction can be obtained through time series analysis or machine learning methods. However, in the integrated energy system, there are complex coupling relationships between multiple probability variables. In this situation, in order to achieve accurate prediction in the case of multiple variables, it is extremely important to comprehensively consider and analyze the correlations between probability variables and their influencing factors, and fully extract their spatial and temporal coupling characteristics. Compared with the single-variable independent prediction method, the joint prediction of multivariate probability variables can further explore the internal relationship of uncertain factors in the integrated energy system and improve the prediction accuracy and efficiency.

[0006] Simply put, regarding the complex coupling characteristics between various uncertainties on the energy-price plane in the integrated energy system, the conventional independent prediction method faces the problems of low efficiency and difficulty in accurate prediction. Further examining the mutual influence between the behavior habits of users on the energy utilization side, the response to load demand, and the energy price in the market, and constructing an appropriate prediction model to achieve accurate joint prediction of energy load prices in the integrated energy system is still a difficult problem.

Summary of the Invention

[0007] In order to solve the above problems, the present invention provides a multi-task joint prediction method and system for energy load prices in an integrated energy system. The present invention can perform correlation analysis and joint prediction on the two uncertainties of the energy aspect and the price aspect in a situation with many uncertainties in the integrated energy system.

[0008] According to some embodiments, the present invention adopts the following technical solutions.

[0009] Analyzing the spatio-temporal coupling characteristics between various uncertainty factors affecting the integrated energy system from both the perspectives of cross-correlation and auto-correlation; Constructing a training dataset based on the results of two correlation analyses and performing feature extraction; Performing secondary extraction on the extracted features by means of a channel attention and sequential attention mechanism; Performing feature sharing on the data after secondary extraction by means of a parameter sharing learning mechanism, in a multi-task learning method, taking price prediction as the main task, energy and load prediction as sub-tasks, using a hard sharing mechanism for the coupling between the corresponding loads of each type of task, using a soft sharing mechanism between different types of tasks, sharing information from the sub-tasks to the main task, and obtaining a joint prediction result; A multi-task joint prediction method for energy load price in an integrated energy system including the above.

[0010] As an optional embodiment, perform cross-correlation analysis between multiple loads, new energy power generation, energy price, time, and meteorological factors respectively by means of Pearson product-moment correlation coefficient, Spearman rank correlation coefficient, and Kendall rank correlation coefficient.

[0011] As an optional embodiment, analyze the sequential features of each factor by means of the autocorrelation coefficient to determine the data time length used for prediction.

[0012] As an optional embodiment, the step of performing feature extraction on the results of the two correlation analyses specifically includes performing feature extraction on the input data by means of a convolutional layer, the convolutional layer traversing the input data regularly, performing multiplication and addition of matrix elements on the input data, and superimposing deviation amounts; Transmitting the feature map output from the convolutional layer to the pooling layer by means of the activation of the ReLU function, the window of the pooling layer sliding over the entire area of the input according to the step size, traversing each position by the window, and calculating the output; Repeating the above steps to obtain the final output features.

[0013] As an alternative embodiment, in the step of performing secondary extraction on the extracted features by means of channel attention and a sequential attention mechanism, specifically, a channel attention module and a sequential attention module are connected in series, the input features are passed through the channel attention module to obtain a channel attention matrix, the features obtained by multiplying the matrix by the original map are used as the input to the sequential attention module, passed through the sequential attention module to obtain a temporal attention matrix, and the output features are obtained by multiplying the temporal attention matrix by the original map. This step includes the following operations.

[0014] As an alternative embodiment, the processing of the channel attention module includes passing the input feature sequence through two parallel MaxPool layers and AvgPool layers to compress the dimension of the feature sequence and the number of channels, then expanding the number of channels back to the original number of channels, passing through the ReLU activation function to obtain two activated results, adding these two output results element by element, and passing through the sigmoid activation function to obtain a channel attention matrix.

[0015] The processing of the sequential attention module includes concatenating the feature sequence obtained by the processing of the channel attention module with data of different channels according to the time dimension to obtain a feature sequence, performing feature extraction by means of a one-dimensional convolutional layer to convert the dimension of the feature sequence back to the original dimension, and passing through the sigmoid activation function to obtain a sequential attention matrix.

[0016] As an alternative embodiment, feature sharing is performed by means of an LSTM network. For the thermoelectric load, hard sharing is directly performed through the LSTM network. Among different LSTM network levels, for wind power generation, load characteristics, and electricity price characteristics, soft sharing is performed as the sum of weights. For the sub-task, information is shared with the main task during output.

[0017] A correlation analysis module configured to analyze the spatio-temporal coupling characteristics between various uncertainty factors affecting the integrated energy system from the perspectives of both cross-correlation and auto-correlation, A feature extraction module configured to perform feature extraction on the results of two correlation analyses, A secondary extraction module configured to perform secondary extraction on the extracted features by means of channel attention and sequential attention mechanisms, A feature sharing module configured to construct a two-layer feature sharing model by means of hard sharing and soft sharing mechanisms and perform feature sharing on the data after secondary extraction, wherein the two-layer feature sharing model utilizes the hard sharing mechanism for the coupling between the corresponding loads of each type of task according to the classification of the type of prediction task, utilizes the soft sharing mechanism between different types of tasks, shares information from subtasks to the main task, and obtains a joint prediction result. A multi-task joint prediction system for energy load prices in an integrated energy system comprising the above.

[0018] A computer-readable storage medium storing a plurality of commands, wherein the commands are loaded by a processor of a terminal device and the plurality of commands for executing the steps in the method are stored.

[0019] A terminal device comprising a processor for realizing each command and a computer-readable storage medium storing a plurality of commands loaded by the processor for executing the steps in the method.

[0020] Compared with the existing technology, the present invention has the following beneficial effects.

[0021] In a comprehensive energy system with a highly uncertain situation, the present invention performs correlation analysis on the uncertainties of three main categories: energy-load-price, and conducts joint prediction. The proposed multi-task learning model classifies the prediction targets according to the correlation degree and attribute type. In the two-layer feature sharing module, different sharing mechanisms are utilized according to the differences and commonalities of the task types, distinguishing the outer layer and the inner layer, thereby promoting the reasonable sharing of important features, enhancing the generalization ability of the model, and realizing efficient and accurate joint prediction of multiple uncertainties. The proposed deep learning network can realize the classification and extraction of spatial and temporal coupling features in the data, emphasize important features, and further improve the prediction accuracy of the model.

[0022] In order to make the above objects, features and advantages of the present invention clearer and more understandable, the following preferred embodiments are given and will be described in detail together with the accompanying drawings as follows.

[0023] The drawings in the specification constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments and their descriptions of the present invention are used to interpret the present invention and are not an improper limitation to the present invention.

Brief Description of the Drawings

[0024]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Embodiments for Carrying Out the Invention

[0025] Hereinafter, the present invention will be further described with reference to the drawings and examples.

[0026] It should be noted that the following detailed description is exemplary and is intended to provide a further description of the present invention. Unless otherwise specified, all technical terms and scientific terms used in this specification have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0027] Furthermore, the terms used herein are for the purpose of merely describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. The singular forms used herein are intended to include the plural forms as well, unless otherwise specified in the context. Also, in this specification, when the terms "comprising" and / or "including" are used, the presence of features, steps, operations, devices, components and / or combinations thereof is indicated.

[0028] As shown in Fig. 1, for a plurality of prediction targets and their complex coupling characteristics, in the vertical direction, for the two-layer multi-task learning model, based on the commonalities and differences between different tasks, each task is classified into three main categories: new energy prediction task, multi-load prediction task, and price prediction task. In the construction of the network model, according to the type of task, the feature sharing model is classified into an inner layer and an outer layer. Then, an appropriate sharing mechanism is utilized between different levels to achieve feature sharing between tasks with complex coupling characteristics, thereby enhancing the generalization ability of the model. In the horizontal direction, the model utilizes the CNN-SCAM-LSTM network. First, the CNN realizes feature extraction of the entire input data. Then, the proposed SCAM is utilized as an application of the attention mechanism to classify features according to the type of task, mine the spatio-temporal features of the data, and realize the extraction of important features. Finally, the LSTM constructs a feature sharing model, further obtains important auxiliary coupling information based on the distinction between the inner layer and the outer layer, and conducts hierarchical and targeted shared learning to improve the model prediction accuracy.

[0029] Note that the data in this case is derived from the wind power generation, electricity-thermal load, electricity price, and meteorological data of Denmark from 2010 to 2013 in the European power system open source dataset. Here, considering the influence of the wind power generation equipment capacity on the wind power generation prediction data, the wind power generation data is adjusted at different capacity ratios. The data of wind power generation, electricity-thermal load, and electricity price from 2010 to 2011 are shown in Fig. 2.

[0030] Perform a correlation analysis of the data.

[0031] Mutual correlation analysis In a comprehensive energy system, multiple loads, new energy power generation, energy prices, time, and meteorological factors interact with each other and have complex correlations due to differences in coupling mechanisms and sequential uncertainties. To analyze the influence mechanism between each influencing factor and the prediction target, fully mine the correlation information in multivariate data, effectively select the input variables of the joint prediction model, and construct the corresponding prediction network structure, it is necessary to quantitatively analyze the dataset.

[0032] The present invention examines the correlation between each influencing factor and the prediction target using the Pearson product-moment correlation coefficient, Spearman rank correlation coefficient, and Kendall rank correlation coefficient.

[0033] 1. Pearson product-moment correlation coefficient The Pearson correlation coefficient is used to measure the linear correlation between two variables X and Y. Its value ranges from +1 to -1, where 1 represents a perfect positive linear correlation, 0 represents a non-linear correlation, and -1 represents a perfect negative linear correlation. One of the important mathematical properties of the Pearson correlation coefficient is that it does not change when the positions and scales of the two variables change independently.

[0034] The following formula is the calculation formula corresponding to the Pearson product-moment correlation coefficient.

[0035]

Equation

[0036] 2. Spearman rank correlation coefficient The Spearman correlation coefficient is used to perform linear correlation analysis using the magnitudes of the ranks of two variables and is a non-parametric statistical method that does not depend on the distribution of the original variables. Therefore, its scope of application is much wider than that of the Pearson correlation coefficient. The Spearman correlation coefficient can be calculated even if the original data is rank data. The Spearman correlation coefficient can also be calculated for data that follows the Pearson correlation coefficient, but its statistical power is slightly lower than that of the Pearson correlation coefficient. When there are no duplicate values in the data and the two variables have a perfect monotonic relationship, the Spearman correlation coefficient is +1 or -1.

[0037] The following formula is the calculation formula corresponding to the Spearman rank correlation coefficient for sample data with a sample size of n.

[0038]

Number

[0039] In the formula, d i is the rank difference between data X i and Y i . The rank of a number is the position of this number when the numbers in the column where this number is located are sorted in ascending order. When there are the same numerical values, it is the arithmetic mean of their positions.

[0040] 3. Kendall Rank Correlation Coefficient The Kendall correlation coefficient is also a rank correlation coefficient and is used to reflect the correlation index of categorical variables. It is applicable when both variables are ordinal categories, and its value is represented by the Greek letter τ.

[0041] The following formula is the calculation formula corresponding to the Kendall rank correlation coefficient.

[0042]

Number

[0043] In the formula, C represents the number of pairs of elements having consistency in the sample data X and Y (two elements form a pair). D represents the number of pairs of elements having inconsistency in the sample data X and Y.

[0044] Generally, the Pearson correlation coefficient is a statistic for interval variables, the Spearman correlation coefficient is a statistic for ordinal variables, and the Kendall correlation coefficient is a statistic for nominal variables. The analysis results of the cases are shown in FIGS. 3(a) to 3(c). Based on the results of the cross-correlation analysis, appropriate influencing factors can be selected as the input dataset of the prediction network. It can be seen that the correlation relationships between each prediction target and the influencing factors are consistent even when measured by different correlation coefficients. Regarding the correlation relationship and degree of correlation, the electrical load has a moderate correlation with the heat load and electricity price, and a weak correlation with the temperature and irradiance. The heat load has a strong negative correlation with the temperature and a moderate correlation with the electrical load due to the influence of the thermoelectric coupling. The electricity price has a moderate correlation with the electrical load, which is consistent with the supply-demand relationship between the load and the price, and has a weak correlation with the wind power generation on the power supply side. The wind power generation mainly has a strong correlation with the wind speed due to the influence of meteorological factors.

[0045] Autocorrelation analysis Since the cross-correlation analysis cannot intuitively reflect the change pattern of each prediction target in the time domain, there is a lack of theoretical basis for determining the length of the sequence of the input data. As a result, when the time sequence is too long, feature redundancy occurs, and the model faces the problem of learning too many unnecessary parameters. In this way, not only does the computational load of the model increase, but also the prediction accuracy of the model in the test set decreases, and the overfitting phenomenon occurs in the model. On the other hand, if it is too short, it is impossible to perform high-precision prediction for highly non-linear time sequences. In contrast, the autocorrelation coefficient (Autocorrelation Function, ACF) can be used to analyze the sequential characteristics of the data. The following formula is the calculation formula of the autocorrelation coefficient.

[0046]

Equation

[0047] In the formula, k represents the delay time,

Number

[0048] Figure 4 shows the correlation of data with a two-week delay in electrical load, heat load, electricity price, and wind power generation. It can be seen that the electrical load has periodicity daily and weekly, the heat load has periodicity daily, the electrical load shows daily correlation and slightly weak weekly correlation, and the wind power generation has no obvious time characteristics. From the above analysis results, it can be seen that it is appropriate to select 24h or 24*7h as the data length.

[0049] Feature extraction is performed by a CNN network based on the results of the correlation analysis.

[0050] The structure of the feature extraction layer used in the present invention is shown in Figure 5. First, feature extraction is performed on the input data by a convolutional layer, and the key of its function is the convolutional kernel. The convolutional kernel traverses the input data regularly, performs multiplication and addition of matrix elements on the input data within its receptive field, and superimposes the deviation amounts. Next, through the activation of the ReLU function, the feature map output from the convolutional layer is transmitted to the pooling layer. The pooling layer is a downsampling process and is mainly used for data dimension reduction to avoid overfitting. Similar to the convolutional layer, the operator of the pooling layer is composed of a window with a fixed size, and the window also slides over the entire area of the input according to the step size, traverses each position by the window, and calculates the output. Finally, the final output features are obtained by repeating according to the above structure.

[0051] Further extraction of important features is performed based on the attention module.

[0052] The attention mechanism model used in the present invention is mainly an improvement with reference to the Convolutional Block Attention Module (CBAM), and the Sequential Convolution Attention Module (SCAM) is under consideration. As shown in FIG. 6, the SCAM model considered in the present invention has the same overall process structure as CBAM, and the Channel Attention Module (CAM) and the Sequential Attention Module (SAM) are connected in series.

[0053] The formula of the model is shown as follows.

[0054]

Number

[0055] In the formula, the feature

Number

Number

Number

Number

Number

[0056] The channel attention module in the provided SCAM model has an overall structure that remains unchanged compared to that of CBAM. Mainly, the data dimension and the level dimension are adjusted, and its model structure is shown in Figure 7. First, the input feature sequence is passed through two parallel MaxPool layers and AvgPool layers to compress the dimension of the feature sequence from C*T to C*1. Next, it is passed through the Share MLP module. In this module, the number of channels is compressed to 1 / r times (Reduction, reduction rate) of the original, and then expanded to the original number of channels. Through the ReLU activation function, two activated results are obtained. Finally, these two output results are added element-wise, and through the sigmoid activation function, a channel attention matrix is obtained.

[0057] The following formula is the expression of the above model.

[0058]

Number

[0059] The sequential attention module in the provided SCAM model has its model structure shown in Figure 8. First, using the Concat operation, the feature sequence obtained by the processing of the channel attention module is concatenated with the data of different channels according to the time dimension to obtain a feature sequence with a dimension of 1*(C*T). Next, feature extraction is performed by a 1D convolutional layer with a convolutional kernel size of 3, and the padding is set to 3 so that the dimension of the feature sequence does not change. Finally, using the view operation, the dimension of the feature sequence is converted to C*T, and through the activation function Sigmoid, a sequential attention matrix is obtained.

[0060] The expression of the above model is as follows.

[0061]

Number

[0062] Perform feature sharing based on the LSTM network.

[0063] After extracting important features by SCAM, the data is transmitted to the feature sharing layer. This part mainly considers the correlation between prediction targets, performs information sharing on the features extracted by the previous network, and the selection of appropriate network structure and feature sharing mechanism is the key point.

[0064] The energy-load-price joint prediction model constructed in the present invention has the characteristics of many parameters and a complex structure, is difficult to have the problem of overfitting, and the generalization ability of the model is also stronger. In the feature sharing part of multi-task learning, according to the type of task, the feature sharing model is classified into an inner layer and an outer layer, and different sharing mechanisms are used. The feature sharing between the prediction tasks of the three main categories of energy, load, and price is the outer layer, and a soft sharing mechanism is used considering the complex correlation and different influence mechanisms between them. The feature sharing between each subtask within each type of task is the inner layer. For example, in load prediction, the coupling between multiple loads is strong, and the feature sharing between different loads is the inner layer, and a hard sharing mechanism is used. Also, considering the influence of energy and load on price, a two-layer feature sharing model is constructed with price prediction as the main task and energy and load prediction as the sub-tasks. The specific model of this case is shown in Figure 9. The thermal power load performs hard sharing directly through the LSTM network, and between different LSTM network levels, the wind power generation, load characteristics, and electricity price characteristics perform soft sharing as the sum of weights, and the sub-tasks share information to the main task when the output of the fully connected layer.

[0065] Analysis of case results The construction and training of the two-layer joint prediction model of energy-load-price in the present invention are carried out under the framework of PyTorch deep learning. The hardware platform uses an Intel Core i 7 CPU, and the data of the calculation example is derived from the data of Denmark in Europe from 2010 to 2012. The training set and the validation set use the data from 2010 to 2011, and the test set uses the data from 2012. The electric-thermal load, wind power generation, and electricity price are predicted with a 24h step size.

[0066] In this case, the effectiveness of the designed two-layer multi-task joint prediction model is verified by comparing the prediction results of the proposed CNN-SCAM-LSTM-MTL model with those of the single-task CNN-LSTM model, the multi-task CNN-LSTM-MTL model, and the CNN-CBAM-LSTM-MTL model.

[0067] As shown in Figure 10, in order to compare with the prediction results, one-week data from January 2 to January 8 in the test set was selected. In the comparison between single-task learning and multi-task learning, the prediction effect of using only the CNN-LSTM-MTL model is inferior to that of single-task learning. However, the prediction effects of the CNN-CBAM-LSTM-MTL and CNN-SCAM-LSTM-MTL models using the attention mechanism are similar to that of the single-task learning CNN-LSTM model and closer to the actual value. It can be seen that the correlation between the prediction targets of the dataset used in the case is medium, and due to the mutual influence of the input features when only using the multi-task learning framework, the requirement for extracting important features becomes higher. Therefore, the necessity of using the attention mechanism is proved by this result.

[0068] In this case, the data for the entire year in the test set was predicted, and the past prediction accuracy was calculated using the mean absolute percentage error (MAPE) and the root mean square error (RMSE) as the evaluation indicators of the model. The comparison of the results is shown in Tables 1 and 2. In the CNN-LSTM-MTL network, when only the multi-task learning framework was used for joint prediction, it was difficult to extract the corresponding important features according to the differences in the prediction targets, and its prediction effect was rather inferior to that of the CNN-LSTM model. On the other hand, when the attention mechanism was introduced, it was found that the prediction effects of the CNN-CBAM-LSTM-MTL and CNN-SCAM-LSTM-MTL models were significantly improved. From this, it can be seen that the attention mechanism used can extract and classify important features.

[0069] Compared with the CNN-LSTM model, it can be seen that the CNN-CBAM-LSTM-MTL model still has a lower prediction accuracy for the electric-thermal load than single-task learning, but has good prediction effects for wind power generation and electricity prices. The electric-thermal load has obvious periodicity, while wind power generation and electricity prices have stronger randomness and volatility, and the sequential features of the data are different in the time dimension. CBAM can extract important features from the spatial dimension (i.e., the channels in the feature data), but for the feature data of different channels in the time dimension, it performs superposition processing to obtain a common attention weight matrix. As a result, CBAM cannot extract the sequential feature differences in the case data. Regarding this, the present invention considered SCAM.

[0070] The CNN-SCAM-LSTM-MTL model studied in the present invention has better overall prediction accuracy than the CNN-LSTM model and the CNN-CBAM-LSTM-MTL model. The SCAM studied in the present invention solves the drawbacks of the extraction of sequential features in CBAM, better balances the spatial and temporal characteristics of feature information, can decouple input features according to the differences of the prediction targets from two dimensions of space and time, and improves the prediction accuracy. Regarding this, the effectiveness of the two-layer joint prediction model of energy-load-price based on multi-task learning and SCAM studied in the present invention has been proven.

[0071]

Table 1

[0072]

Table 2

[0073] In this embodiment, the probability distribution information of the prediction accuracy in the MAPE and RMSE indicators in 2012 is statistically analyzed on a daily basis. The probability density functions are shown in FIGS. 11 and 12, and the confidence intervals at a 90% confidence level are shown in Tables 3 and 4. Overall, the CNN-SCAM-LSTM-MTL network has better probability peak values of prediction accuracy and confidence interval sizes, reflects the stability of the prediction results, further proves that the multi-task learning model studied in the present invention has better prediction accuracy, and can improve the generalization of the model by considering the training information in multi-tasks.

[0074]

Table 3

Table 4

[0075] As will be understood by those skilled in the art, the embodiments of the present invention may be provided as a method, a system, or a computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Further, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, magnetic disk memory, CD-ROM, optical memory, etc.) having computer-usable program code embodied therein.

[0076] The present invention has been described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be noted that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, may be realized by computer program commands. By providing these computer program commands to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device, a machine can be configured, whereby an apparatus for realizing the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram can be constructed by commands executed by the processor of the computer or other programmable data processing device.

[0077] These computer program commands may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, whereby a manufactured product including the command device is constructed by the commands stored in the computer-readable memory, and the command device realizes the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram.

[0078] These computer program commands may be loaded onto a computer or other programmable data processing apparatus, thereby constituting a process implemented by a computer by executing a series of operation steps on the computer or other programmable apparatus, and the commands executed on the computer or other programmable apparatus provide steps for realizing the functions specified in one or more flows of the flowchart and / or one or more blocks of the block diagram.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various changes and modifications are possible to the present invention. Any changes, equivalent substitutions, improvements, etc. made within the spirit and principle scope of the present invention should be included in the protection scope of the present invention.

[0080] As described above, the specific embodiments of the present invention have been described with reference to the drawings, but the protection scope of the present invention is not limited thereby. As can be understood by those skilled in the art, various changes or modifications that can be made by those skilled in the art without creative labor based on the technical solution means of the present invention are also within the protection scope of the present invention.

Claims

1. Analyzing the spatiotemporal coupling characteristics between each uncertainty factor affecting the total energy system in terms of both cross-correlation and auto-correlation; performing feature extraction on the results of the two correlation analyses; performing secondary extraction on the extracted features using channel attention and sequential attention mechanisms; A step of performing feature sharing on the data after secondary extraction through a parameter sharing learning mechanism, in which in the multi-task learning method, price prediction is the main task, energy,load prediction is the sub-task, a hard sharing mechanism is used for coupling between the corresponding loads of each type of task, and a soft sharing mechanism is used between different types of tasks, so as to share information from the sub-task to the main task, and a joint prediction result is obtained; A multi-task collaborative forecasting method for energy load prices in an integrated energy system, comprising:

2. The method for multi-task collaborative forecasting of energy load price in an integrated energy system according to claim 1, characterized in that: the cross-correlation analysis among multiple loads, new energy generation, energy price, time and weather factors is respectively performed by Pearson product-moment correlation coefficient, Spearman rank correlation coefficient and Kendall rank correlation coefficient.

3. The multi-task collaborative forecasting method for energy load prices in a comprehensive energy system according to claim 1, characterized in that the sequential features of each factor are analyzed by autocorrelation coefficient to determine the data time length used for forecasting.

4. The step of extracting features from the results of the two correlation analyses specifically includes the steps of: extracting features from the input data by a convolution layer; the convolution layer regularly traverses the input data, multiplying and adding matrix elements on the input data, and superimposing the deviation amounts; By activating the ReLU function, the feature map output from the convolution layer is transmitted to a pooling layer, and the window of the pooling layer slides over the entire region of the input according to a step size, traverses each position with the window, and calculates the output; The method for multi-task collaborative forecasting of energy load prices in a total energy system according to claim 1 , further comprising: a step of repeating the steps to obtain a final output characteristic.

5. The method for multi-task collaborative prediction of energy load prices in an integrated energy system as described in claim 1, characterized in that the step of performing secondary extraction on the extracted features through channel attention and sequential attention mechanisms specifically includes the steps of connecting a channel attention module and a sequential attention module in series, passing the input features through the channel attention module to obtain a channel attention matrix, multiplying the original map by the matrix to obtain the features as the input of the sequential attention module, passing the features through the sequential attention module to obtain a temporal attention matrix, and multiplying the original map by the temporal attention matrix to obtain output features.

6. The process of the channel attention module includes the steps of: passing the input feature sequence through two parallel MaxPool layers and AvgPool layers to compress the dimension of the feature sequence, compressing the number of channels and then expanding it to the original number of channels, and obtaining two activated results through a ReLU activation function; and adding these two output results element by element, and obtaining a channel attention matrix through a sigmoid activation function; Or, The multi-task collaborative prediction method for energy load prices in an integrated energy system as described in claim 1 or 5, characterized in that the processing of the sequential attention module includes the steps of: concatenating the feature sequence obtained by the processing of the channel attention module with the data of different channels according to the time dimension to obtain a feature sequence; and performing feature extraction by a one-dimensional convolution layer, converting the dimension of the feature sequence into a primitive dimension, and obtaining a sequential attention matrix through the activation function Sigmoid.

7. The multi-task collaborative forecasting method for energy load prices in an integrated energy system as described in claim 1, characterized in that feature sharing is performed by the LSTM network, thermoelectric loads perform hard sharing directly through the LSTM network, and between different LSTM network levels, wind power generation, load features and electricity price features perform soft sharing as the sum of weights, and secondary tasks share information to the primary task at the time of output.

8. a correlation analysis module configured to analyze the spatiotemporal coupling characteristics between each uncertainty factor affecting the total energy system in terms of both cross-correlation and auto-correlation; a feature extraction module configured to perform feature extraction on the results of the two correlation analyses; a secondary extraction module configured to perform secondary extraction on the extracted features by channel attention and sequential attention mechanisms; A feature sharing module configured to construct a two-layer feature sharing model by using hard sharing and soft sharing mechanisms, and perform feature sharing on the data after secondary extraction, in which the two-layer feature sharing model uses a hard sharing mechanism to connect the corresponding loads of each type of task according to the classification of the type of prediction task, and uses a soft sharing mechanism between tasks of different types, so as to share information from the secondary task to the primary task, and obtain a joint prediction result; A multi-task collaborative forecasting system for energy load prices in a comprehensive energy system, comprising:

9. A computer-readable storage medium having stored thereon a plurality of commands for executing the steps of the method according to any one of claims 1 to 7, when the computer-readable storage medium is loaded by a processor of a terminal device.

10. A terminal device, comprising: a processor for implementing each command; and a computer-readable storage medium on which are stored a plurality of commands for executing the steps of the method according to any one of claims 1 to 7, the computer-readable storage medium being loaded by the processor.

Images