Method and system for multitask collaborative forecasting of energy load prices in an integrated energy system

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 priority to a Chinese patent application bearing application number 202310519941.5 and entitled "Method and system for multi-task collaborative forecasting of energy load prices in an integrated energy system," filed with the State Intellectual Property Office of the People's Republic of China on May 6, 2023, the entire contents of which are incorporated by reference into and constitute a part of the present invention for all purposes.

[0002] The present invention belongs to the technical field of total energy system planning, and relates to a method and system for multi-task collaborative forecasting of energy load prices in a total energy system. [Background technology]

[0003] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0004] In the integrated energy system, the penetration rate of new energy sources such as wind power and solar power has gradually increased on the energy supply side, and flexible loads such as electric vehicles and energy storage batteries have been connected in large quantities on the energy usage side. Furthermore, the participation of private capital in competitive energy trading has been encouraged. Thus, the uncertainty factors in the system have significantly increased across multiple parts, including energy generation, trading, and consumption. Accurate prediction of uncertainty is a prerequisite for ensuring the efficient operation of the integrated energy system.

[0005] Traditional forecasting methods often independently forecast three uncertainties: renewable energy power on the energy supply side, load on the energy user side, and energy prices on the market side. The models are simple and the methods are relatively mature, using time sequence analysis or machine learning to obtain point, interval, and probabilistic forecasts. However, in integrated energy systems, complex coupling exists between multiple random variables. In this situation, to achieve accurate multivariate forecasts, it is crucial to comprehensively consider and analyze the correlations between random variables and their influencing factors and fully extract their spatial and temporal coupling characteristics. Compared to independent single-variable forecasting methods, joint forecasting of multivariate random variables can further explore the inherent relationships between uncertainties in integrated energy systems and improve forecast accuracy and efficiency.

[0006] In short, due to the complex coupling characteristics between uncertainties in the energy and price aspects of the integrated energy system, traditional independent forecasting methods face the problems of low efficiency and difficulty in achieving accurate forecasts. Further consideration of the mutual influence between the behavioral habits of energy users, response to load demand, and energy prices on the market side, and building an appropriate forecasting model to realize accurate joint forecasting of energy load prices in the integrated energy system remains a challenging task. Summary of the Invention

[0007] In order to solve the above problems, the present invention provides a method and system for multi-task collaborative forecasting of energy load prices in an integrated energy system, and the present invention can perform correlation analysis on two uncertainties, energy and price, and perform collaborative forecasting in a situation where there is a lot of uncertainty in the integrated energy system.

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

[0009] A step of analyzing the spatiotemporal coupling characteristics between each uncertainty factor affecting the overall energy system from the perspective of both cross-correlation and auto-correlation; constructing a training dataset based on the results of the two correlation analyses and performing feature extraction; 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 and load prediction are the sub-tasks, a hard sharing mechanism is used to connect the corresponding loads of each type of task, and a soft sharing mechanism is used between different types of tasks, information is shared from the sub-task to the main task, and a joint prediction result is obtained; Multi-task collaborative forecasting method for energy load prices in an integrated energy system including:

[0010] As an alternative embodiment, the Pearson product-moment correlation coefficient, the Spearman rank correlation coefficient and the Kendall rank correlation coefficient are used to perform the cross-correlation analysis among the multiple loads, the new energy generation, the energy price, the time and the weather factors, respectively.

[0011] As an alternative embodiment, the sequential features of each factor are analyzed by autocorrelation coefficients to determine the data time length to be used for prediction.

[0012] In an optional embodiment, the step of extracting features from the results of the two correlation analyses specifically includes the steps of: extracting features from input data using a convolution layer; the convolution layer regularly traverses the input data; multiplying and adding matrix elements on the input data; and superimposing deviation amounts; By activating the ReLU function, the feature map output from the convolution layer is transmitted to the pooling layer, and the window of the pooling layer slides over the entire input region according to the step size, traverses each position with the window, and calculates the output; Repeating the steps to obtain a final output feature.

[0013] In an optional embodiment, the step of performing secondary extraction on the extracted features using the 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 the output features.

[0014] In an optional embodiment, the processing of the channel attention module includes the steps of: passing the input feature sequence through two parallel MaxPool 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, obtaining two activation 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.

[0015] 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 data from different channels according to the time dimension to obtain a feature sequence; extracting features using a one-dimensional convolution layer, converting the dimension of the feature sequence into a primitive dimension, and obtaining a sequential attention matrix through an activation function Sigmoid.

[0016] In an alternative embodiment, feature sharing is performed by an LSTM network, with the thermoelectric load performing hard sharing directly through the LSTM network, and between different LSTM network levels, the wind power generation, load features and electricity price features performing soft sharing as the sum of weights, and the secondary tasks share information to the primary task at the output.

[0017] a correlation analysis module configured to analyze the spatiotemporal coupling characteristics between each uncertainty factor affecting the overall 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 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 using hard sharing and soft sharing mechanisms, and perform feature sharing on data after secondary extraction, wherein the two-layer feature sharing model uses hard sharing mechanisms to connect corresponding loads of each type of task according to the classification of the type of prediction task, and soft sharing mechanisms to connect tasks of different types, so as to share information from the secondary task to the primary task, thereby obtaining a joint prediction result; A multi-task collaborative forecasting system for energy load prices in an integrated energy system.

[0018] A computer-readable storage medium having stored thereon a plurality of commands, the commands being loaded by a processor of a terminal device to execute steps in the method.

[0019] A terminal device comprising: a processor for implementing each command; and a computer-readable storage medium that is loaded by the processor and stores a plurality of commands for performing the steps of the method.

[0020] Compared with existing technologies, the present invention has the following beneficial effects:

[0021] This invention performs correlation analysis and joint forecasting of three major categories of uncertainty: energy, load, and price, in situations where there is a lot of uncertainty in the integrated energy system. The proposed multi-task learning model classifies the prediction target according to the degree of correlation and attribute type. The two-layer feature sharing module uses different sharing mechanisms depending on the differences and similarities between task types, distinguishing between the outer and inner layers, thereby promoting rational sharing of important features, improving the generalization ability of the model, and achieving efficient and accurate joint forecasting of multiple uncertainties. The proposed deep learning network realizes classification and extraction of spatial and temporal joint features in the data, highlighting important features and further improving the model's prediction accuracy.

[0022] In order to make the above objects, features and advantages of the present invention clearer and more comprehensible, preferred embodiments will be described in detail below in conjunction with the accompanying drawings.

[0023] The drawings in the specification that form a part of the present invention are used to provide a further understanding of the present invention, and the schematic examples of the present invention and their descriptions are used to interpret the present invention and are not intended to be unduly limiting to the present invention. [Brief explanation of the drawings]

[0024] [Figure 1] 1 is a technical roadmap of a joint prediction method that takes multiple uncertainties into account in the present invention. [Figure 2] FIG. 1 is a diagram showing case data in the present invention. [Figure 3] FIG. 10 is a diagram showing the results of a cross-correlation analysis according to the present invention. [Figure 4] FIG. 1 is a diagram showing the results of autocorrelation analysis in the present invention. [Figure 5] FIG. 2 is a schematic diagram of a feature extraction layer in the present invention. [Figure 6] FIG. 1 is a structural diagram of a sequential convolutional attention module in the present invention. [Figure 7]FIG. 1 is a schematic diagram of a channel attention module of a sequential convolutional attention module in the present invention; [Figure 8] FIG. 1 is a schematic diagram of a sequential convolutional attention module in the present invention; [Figure 9] FIG. 2 is a schematic diagram of a feature sharing layer in the present invention. [Figure 10] FIG. 10 is a comparison diagram of model prediction results in the present invention. [Figure 11] FIG. 1 is a comparison diagram of the MAPE probability distribution of model prediction results in the present invention. [Figure 12] FIG. 10 is a comparison diagram of RMSE probability distributions of model prediction results in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0025] The present invention will now be further described with reference to the following figures and examples.

[0026] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present invention. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0027] Furthermore, the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit example embodiments according to the present invention. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly dictates otherwise. Also, when used herein, the terms "comprises" and / or "comprises" specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0028] As shown in Figure 1, for multiple prediction targets and their complex connection characteristics, the two-layer multi-task learning model vertically classifies tasks into three main categories: new energy forecasting tasks, multiple load forecasting tasks, and price forecasting tasks, based on the similarities and differences between different tasks. In constructing the network model, feature sharing models are divided into inner and outer layers according to the task type. Then, appropriate sharing mechanisms are used between different levels to achieve feature sharing between tasks with complex connection characteristics and improve the generalization ability of the model. Horizontally, the model uses a CNN-SCAM-LSTM network. First, CNN is used to extract features from the entire input data. Then, the proposed SCAM is used as an application of the attention mechanism to classify features according to the task type and mine the spatiotemporal features of the data to extract important features. Finally, a feature sharing model is constructed using LSTM. Important auxiliary connection information is further obtained based on the distinction between the inner and outer layers, and shared learning is performed hierarchically and targetedly, improving the model's prediction accuracy.

[0029] The data for this case comes from Denmark's wind power, heat load, electricity price, and weather data for 2010-2013 in the European Power System Open Source Dataset. The wind power data was adjusted for different capacity ratios to account for the impact of wind power capacity on wind power forecast data. The wind power, heat load, and electricity price data for 2010-2011 are shown in Figure 2.

[0030] Conduct correlation analysis of the data.

[0031] Cross-correlation analysis In an integrated energy system, multiple loads, new energy generation, energy prices, time, and weather factors interact with each other, and have complex correlations due to different coupling mechanisms and sequential uncertainties. A quantitative analysis of the data set is necessary to analyze the influence mechanisms between each influencing factor and the forecast target, fully mine the correlation information in the multivariate data, effectively select input variables for the joint forecasting model, and build the corresponding forecasting network structure.

[0032] The present invention examines the correlation between each influential factor and the predicted object using 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 measures the linear correlation between two variables X and Y. Its value lies between +1 and -1, where 1 is a perfect positive linear correlation, 0 is a nonlinear correlation, and -1 is a perfect negative linear correlation. One important mathematical property of the Pearson correlation coefficient is that it does not change when the location and scale of the two variables change independently.

[0034] The following formula is the calculation formula for the Pearson product-moment correlation coefficient:

[0035]

number

[0036] 2. Spearman rank correlation coefficient The Spearman correlation coefficient is a non-parametric statistical method that performs linear correlation analysis using the magnitude of the ranks of two variables and is not dependent on the distribution of the original variables. Therefore, its scope of application is much broader 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 conforms to the Pearson correlation coefficient, but its statistical power is slightly lower than that of the Pearson correlation coefficient. If there are no duplicate values ​​in the data and the two variables are completely monotonic, the Spearman correlation coefficient will be +1 or -1.

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

[0038]

number

[0039] In the formula, d i is Data X i and Y i The rank of a number is the position of that number when the numbers in the column in which it appears are sorted in ascending order, and if there are identical numbers, it is the arithmetic mean of their positions.

[0040] 3. Kendall rank correlation coefficient The Kendall correlation coefficient is also a rank correlation coefficient, which is intended to reflect a measure of correlation between categorical variables when the two variables are both ordered categories, and its value is represented by the Greek letter τ.

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

[0042]

number

[0043] In the formula, C denotes the number of pairs of consistent elements (two elements paired) in sample data X and Y. D denotes the number of pairs of inconsistent elements in 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 for the case are shown in Figures 3(a)-(c). Based on the results of the cross-correlation analysis, appropriate influencing factors can be selected as input data sets for the forecasting network. It can be seen that the correlation between each forecast target and influencing factors is consistent even when measured with different correlation coefficients. Regarding correlation and degree of correlation, the electrical load has a moderate correlation with the thermal load and electricity price, and a weak correlation with temperature and irradiance. The thermal load has a strong negative correlation with temperature and a moderate correlation with the electrical load due to the influence of thermoelectric coupling. The electricity price has a moderate correlation with the electrical load, consistent with the supply-demand relationship between load and price, and a weak correlation with wind power generation on the power source side. Wind power generation has a strong correlation with wind speed, mainly due to the influence of meteorological factors.

[0045] Autocorrelation analysis Cross-correlation analysis cannot intuitively reflect the time-domain change patterns of each prediction target, so determining the length of the input data sequence lacks a theoretical basis. As a result, if the time sequence is too long, feature redundancy occurs, and the model faces the problem of learning many unnecessary parameters. This not only increases the computational load of the model, but also reduces the model's prediction accuracy on the test set, resulting in overfitting. On the other hand, if the sequence is too short, it will not be possible to make high-accuracy predictions for highly nonlinear time sequences. To address this issue, the autocorrelation function (ACF) can be used to analyze the sequential features of data. The formula for calculating the autocorrelation coefficient is as follows:

[0046]

number

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

number

[0048] Figure 4 shows the correlations between data on electric load, thermal load, electricity price, and wind power generation with a two-week lag. It can be seen that electric load exhibits daily and weekly periodicity, thermal load exhibits daily periodicity, electric load exhibits daily correlation and a somewhat weak weekly correlation, and wind power generation has no obvious time characteristics. Based on the above analysis results, it can be seen that selecting 24 hours or 24*7 hours as the data length is appropriate.

[0049] Based on the results of the correlation analysis, feature extraction is performed using a CNN network.

[0050] The structure of the feature extraction layer used in this invention is shown in Figure 5. First, the convolutional layer performs feature extraction on the input data, and its key function is the convolutional kernel. The convolutional kernel traverses the input data regularly, multiplying and adding matrix elements to the input data within its receptive field, and then superimposing the deviations. Next, the feature map output from the convolutional layer is transmitted to the pooling layer by activating the ReLU function. The pooling layer is a downsampling process, primarily used to reduce data dimensionality to avoid overfitting. Similar to the convolutional layer, the pooling layer's operator consists of a fixed-size window. The window also slides over the entire input region according to a step size, traverses each position with the window, and calculates the output. Finally, the final output feature is obtained by repeating the above structure.

[0051] Important features are further extracted based on the attention module.

[0052] The attention mechanism model used in this invention is mainly improved by referring to the Convolutional Block Attention Module (CBAM), and a Sequential Convolution Attention Module (SCAM) is considered. As shown in Figure 6, the SCAM model considered in this invention has the same overall process structure as the CBAM, and is a serial connection of a Channel Attention Module (CAM) and a Sequential Attention Module (SAM).

[0053] The model equation is shown as follows:

[0054]

number

[0055] In the formula, the characteristic

number

number

number

number

number

[0056] The overall structure of the channel attention module in the provided SCAM model remains unchanged compared to that of CBAM, with only adjustments made to the data dimension and level dimension. The model structure is shown in Figure 7. First, the input feature sequence is passed through two parallel MaxPool and AvgPool layers to reduce the dimension of the feature sequence from C*T to C*1. Next, it is passed through the Share MLP module, which reduces the number of channels by 1 / r (reduction rate) and then expands it to the original number of channels. Two activation results are obtained through a ReLU activation function. Finally, these two output results are added element-by-element and then used through a sigmoid activation function to obtain the channel attention matrix.

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

[0058]

number

[0059] The model structure of the sequential attention module in the provided SCAM model is shown in Figure 8. First, the feature sequence obtained by the channel attention module is concatenated with data from different channels according to the time dimension using the concat operation to obtain a feature sequence with a dimension of 1*(C*T). Next, feature extraction is performed using a 1D convolution layer with a convolution kernel size of 3, and padding is set to 3 to prevent the dimension of the feature sequence from changing. Finally, the dimension of the feature sequence is converted to C*T using the view operation, and the sequential attention matrix is ​​obtained through the activation function Sigmoid.

[0060] The expression for the above model is as follows:

[0061]

number

[0062] Feature sharing is based on LSTM networks.

[0063] After the key features are extracted by SCAM, the data is transmitted to the feature sharing layer, which mainly considers the correlation between the prediction targets and shares information on the features extracted by the previous network. The key focus is on selecting an appropriate network structure and feature sharing mechanism.

[0064] The energy-load-price joint forecasting model constructed in this invention is characterized by a large number of parameters and a complex structure, making it less susceptible to overfitting and providing stronger generalization capabilities. In the feature sharing portion of multi-task learning, the feature sharing model is divided into inner and outer layers according to the task type, and different sharing mechanisms are used. Feature sharing between the three main categories of forecasting tasks—energy, load, and price—is the outer layer, and soft sharing mechanisms are used to account for the complex correlations and different influence mechanisms between them. Feature sharing between subtasks within each type of task is the inner layer. For example, in load forecasting, the connectivity between multiple loads is strong, and feature sharing between different loads is the inner layer, and hard sharing mechanisms are used. Furthermore, considering the impact of energy and load on price, a two-layer feature sharing model is constructed, with price forecasting as the primary task and energy and load forecasting as secondary tasks. A specific model for this case is shown in Figure 9. The thermal and power loads are hard-shared directly through the LSTM network, and between different LSTM network levels, the wind power generation, load features and electricity price features are soft-shared as the sum of weights. The secondary tasks share information with the primary task at the output of the fully connected layer.

[0065] Case outcome analysis The construction and training of the two-layer joint energy-load-price forecasting model in this invention was carried out under the PyTorch deep learning framework. The hardware platform used was an Intel Core i7 CPU. The data used in the calculation example was derived from data from Denmark in Europe from 2010 to 2012. The training set and validation set used data from 2010 to 2011, and the test set used data from 2012. Electricity-heat load, wind power generation, and electricity price were forecast with a step size of 24 hours.

[0066] In this case, we verify the effectiveness of the designed two-layer multi-task joint prediction model by comparing the prediction results of the proposed CNN-SCAM-LSTM-MTL model with 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, we selected data from the test set for one week, from January 2nd to January 8th, to compare with the prediction results. Comparing single-task learning and multi-task learning, we found that the prediction effect using only the CNN-LSTM-MTL model was inferior to single-task learning. However, the prediction effect of the CNN-CBAM-LSTM-MTL and CNN-SCAM-LSTM-MTL models using the attention mechanism was similar to that of the single-task learning CNN-LSTM model, and was closer to the actual value. This is because the correlation between prediction targets in the dataset used in this case was moderate. The mutual influence of input features when using only the multi-task learning framework increased the demand for extracting important features. This result demonstrates the necessity of using the attention mechanism.

[0068] In this case, we predicted data from the entire one-year test set, and used the mean absolute percentage error (MAPE) and root mean square error (RMSE) as model evaluation metrics to calculate past prediction accuracy. The results are compared in Tables 1 and 2. When using only the multi-task learning framework for joint prediction, the CNN-LSTM-MTL network found it difficult to extract key features corresponding to different prediction targets, resulting in a lower prediction performance than the CNN-LSTM model. However, when an attention mechanism was introduced, the prediction performance of the CNN-CBAM-LSTM-MTL and CNN-SCAM-LSTM-MTL models improved significantly. This demonstrates that the attention mechanism used is capable of extracting and classifying key features.

[0069] Compared with the CNN-LSTM model, the CNN-CBAM-LSTM-MTL model still has lower prediction accuracy for electricity-heat load than single-task learning, but it has better prediction effectiveness for wind power generation and electricity price. This is because electricity-heat load exhibits obvious periodicity, while wind power generation and electricity price exhibit stronger randomness and fluctuations, resulting in different sequential features in the time dimension. CBAM can extract important features from the spatial dimension (i.e., channels in the feature data), but from the time dimension, it performs overlap processing on feature data from different channels to obtain a common attention weight matrix. As a result, CBAM cannot extract sequential feature differences in the case data. In response, the present invention considers SCAM.

[0070] The CNN-SCAM-LSTM-MTL model proposed in this study has better overall prediction accuracy than the CNN-LSTM model and the CNN-CBAM-LSTM-MTL model. This is because the SCAM proposed in this study overcomes the shortcomings of sequential feature extraction in CBAM, better balances the spatial and temporal characteristics of feature information, and can decouple input features according to differences in the prediction target from both spatial and temporal dimensions, thereby improving prediction accuracy. In this regard, the effectiveness of the multi-task learning-based two-layer joint energy-load-price prediction model and SCAM proposed in this study has been proven.

[0071] [Table 1]

[0072] [Table 2]

[0073] In this example, the probability distribution information of the prediction accuracy of the MAPE and RMSE indices in 2012 was statistically collected on a daily basis, and the probability density functions are shown in Figures 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 had better probability peak values ​​of prediction accuracy and better confidence interval sizes, reflecting the stability of the prediction results. This further proves that the multi-task learning model considered in this invention has better prediction accuracy and can improve the generalization of the model by taking into account the training information in multi-tasks.

[0074] [Table 3] [Table 4]

[0075] As will be appreciated by those skilled in the art, 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. The present invention may also take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, magnetic disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0076] The present invention has been described with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. Each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, may be implemented by computer program commands. These computer program commands can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to configure a machine, whereby the commands executed by the processor of the computer or other programmable data processing device configure an apparatus for implementing the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.

[0077] These computer program commands may 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 the commands stored in the computer readable memory configure an article of manufacture that includes a command device, which implements the functions specified in one or more flows of the flowcharts and / or one or more blocks of the block diagrams.

[0078] These computer program commands may be loaded into a computer or other programmable data processing device, whereby a series of operational steps on the computer or other programmable device constitute a computer-implemented process, the commands executed on the computer or other programmable device providing steps for implementing the functions specified in one or more flows of the flowcharts and / or one or more blocks of the block diagrams.

[0079] The above description is merely a preferred embodiment of the present invention, and is not intended to limit the present invention. Those skilled in the art can make various modifications and variations to the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0080] Although the specific embodiments of the present invention have been described above with reference to the drawings, they are not intended to limit the scope of protection of the present invention. Those skilled in the art can understand that various modifications or variations that can be made by those skilled in the art based on the technical solutions of the present invention without any creative work are also within the scope of protection of the present invention.

Claims

1. A step of analyzing the spatiotemporal coupling characteristics between each uncertainty factor affecting the overall energy system from the perspective of both cross-correlation and auto-correlation; constructing an input data set for use in prediction based on the results of the analysis; performing feature extraction on the constructed input dataset using a convolutional layer; 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 and load prediction are the sub-tasks, a hard sharing mechanism is used to connect the corresponding loads of each type of task, and a soft sharing mechanism is used between different types of tasks, information is shared from the sub-task to the main task, and a joint prediction result is obtained; A method for multitask collaborative forecasting of energy, load, and price in an integrated energy system, comprising:

2. 2. The method for multi-task collaborative forecasting of energy, load, and 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 performed by Pearson product-moment correlation coefficient, Spearman rank correlation coefficient, and Kendall rank correlation coefficient, respectively.

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

4. The step of extracting features includes a step in which the convolutional layer regularly traverses the input data set, multiplies and adds matrix elements to the input data set, and superimposes deviation amounts; Activating a ReLU function transmits the feature map output from the convolution layer to a pooling layer, and a window of the pooling layer slides over the entire input region according to a step size, traversing each position with the window, and calculating an output; and repeating the steps of superimposing the deviation amounts and calculating the output to obtain final output characteristics.

5. 2. The method for multi-task collaborative forecasting of energy, load, and price in an integrated energy system according to claim 1, wherein the step of performing secondary extraction on the extracted features using channel attention and sequential attention mechanisms includes the steps of: connecting a channel attention module and a sequential attention module in series; passing 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 input to the sequential attention module; passing 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. A multi-task collaborative forecasting method for energy, load, and price in an integrated energy system according to claim 5, comprising: The processing of the channel attention module includes the steps of: passing the input feature sequence through two parallel MaxPool 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 the channel attention matrix through a sigmoid activation function.

7. A multi-task collaborative forecasting method for energy, load, and price in an integrated energy system according to claim 5, comprising: The method, wherein 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 data of different channels according to a time dimension to obtain a feature sequence; and performing feature extraction using a one-dimensional convolution layer, converting the dimension of the feature sequence into a primitive dimension, and obtaining a sequential attention matrix through an activation function Sigmoid.

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

9. a correlation analysis module configured to analyze the spatiotemporal coupling characteristics between each uncertainty factor affecting the overall energy system from the perspectives of both cross-correlation and auto-correlation; a dataset construction module configured to construct an input dataset for prediction based on the analysis result of the correlation analysis module; a feature extraction module configured to perform feature extraction on the constructed input dataset; 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 using hard sharing and soft sharing mechanisms, and perform feature sharing on data after secondary extraction, wherein the two-layer feature sharing model uses hard sharing mechanisms to connect corresponding loads of each type of task according to the classification of the type of prediction task, and soft sharing mechanisms to connect tasks of different types, so as to share information from the secondary task to the primary task, thereby obtaining a joint prediction result; A multi-task collaborative forecasting system for energy, load, and price in an integrated energy system, comprising:

10. 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 8, when the computer-readable storage medium is loaded by a processor of a terminal device.

11. A terminal device comprising: a processor for implementing each command; and a computer-readable storage medium on which a plurality of commands for executing the steps of the method according to any one of claims 1 to 8 are stored, the computer-readable storage medium being loaded by the processor.

Images