Short-term Prediction Method for Comprehensive Energy Diversified Load Based on Quadratic Decoupling

Abstract

The present invention provides a short-term prediction method for integrated energy multi-load based on secondary decoupling, which includes collecting a multi-load data set, reconstructing the winter data set and the summer data set in the multi-load data set for primary decoupling; decomposing the reconstructed winter data set and summer data set; dividing the decomposed data into a periodic term and a trend term for secondary decoupling; screening the trend term and predicting the screened trend term to obtain a trend term prediction result; performing multi-load prediction on the periodic term to obtain a periodic term prediction result, and adding the trend term prediction result and the periodic term prediction result to obtain a final prediction result. The present invention can effectively capture the coupling between multi-loads, and can achieve accurate decomposition of non-linear and non-stationary signals, improve the anti-noise ability and time-frequency resolution, and effectively improve the prediction accuracy while taking into account the balance.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated energy prediction, and particularly relates to a short-term prediction method for integrated energy multi-load based on quadratic decoupling. Background Art

[0002] When predicting multi-loads traditionally, it is easy to produce a tendentious prediction result. For example, the prediction accuracy of a certain load is greatly improved, but the accuracy of other load predictions is sacrificed, resulting in an imbalance in the final prediction result. To solve the problem of prediction imbalance in multi-load prediction, researchers have proposed a multi-task learning (MTL) model. MTL trains multiple related tasks simultaneously by sharing the underlying network structure, aiming to improve the performance of the model on all tasks. However, MTL also has a limitation, that is, it incorporates all input influencing factors into the model and cannot distinguish whether these factors have a positive or negative impact on the multi-load prediction result. Therefore, it does not fundamentally solve the problem of prediction imbalance. With the in-depth research, the Mixture of experts (MMoE) model has been proposed. MMoE assigns a dedicated expert network to each task and uses a gating mechanism to flexibly control the output combination of different experts to achieve the best adaptation to each task. This innovation effectively alleviates the problem of negative transfer between tasks, that is, the performance improvement of one task no longer comes at the expense of the performance of another task, making MMoE show significant advantages in multi-load prediction. However, despite the many improvements brought by the MMoE model, its network construction and parameter interaction mechanism are new challenges. Researchers have begun to explore improved models. For example, by optimizing the expert selection strategy, improving the gating mechanism, or designing a more efficient network structure, researchers are committed to breaking through the limitations of MMoE in aspects such as network construction and parameter interaction mechanism. These improvements not only help to improve the model performance but also reduce the consumption of computing resources, making multi-task learning more efficient and feasible in practical applications.

[0003] There are two common challenges in the field of multi-task learning. The phenomenon of performance imbalance between tasks. In multi-task learning, when the correlations between tasks are complex, it may occur that improving the performance of one task is accompanied by a decline in the performance of other tasks. In the IES multi-load forecasting task, an improvement in the forecasting effect of one load is often accompanied by a decline in the forecasting effects of the remaining loads. For example, when the accuracy of electric load forecasting increases, the forecasting accuracies of cooling load and heating load decrease. The negative transfer phenomenon. Negative transfer refers to the situation in multi-task learning where the correlations between tasks are weak or even conflicting, resulting in a decline in learning performance. The coupling characteristics between the various loads in IES are constantly changing. For example, the coupling degree between the various loads is stronger in winter and autumn. When it comes to summer, the heating load has large fluctuations and its correlations with the cooling load and electric load are weak, thus leading to the negative transfer phenomenon. The difference between the two lies in: First, the phenomenon of performance imbalance between tasks means that when improving the performance of one task, the performance of other tasks may be sacrificed. It is impossible to achieve ideal effects for all tasks simultaneously. Second, the negative transfer phenomenon means that the effect of multi-task learning is worse than that of training each task separately. It is not as good as independently forecasting each load. Jointly training the model will instead lead to a decline in the task effects. That is to say, when the correlations between loads or between two loads are weak, sharing information between loads through multi-task learning will affect the performance of the network, resulting in a decline in the forecasting accuracy of each load. This phenomenon is more obvious when the correlations between tasks are weak or even conflicting. MMOE uses a gating network combined with an expert network to handle task correlations. Although it alleviates the negative transfer phenomenon to a certain extent, it ignores the differences and interaction information between expert networks, and its potential needs to be further explored. To solve the above problems, existing technologies propose methods using prior knowledge to capture complex task correlations, using shared experts and task-specific experts to reduce the interference of harmful parameters in information, in the form of Progressive Layered Extraction (PLE), using multiple layers of experts and gating networks to extract deeper information from the underlying expert network and separate task parameters from the high-level expert network. Since there are a large number of coupling features and seasonal features between the cooling load, heating load, and electric load in the integrated energy system, through PatchTST data segmentation, the time series of IES multi-loads can be segmented into many data blocks, and through mask capture, each data block can be reconstructed. Since the reconstructed data has received the seasonal and coupling correlation structures, the trend term and seasonality are more prominent. Existing research has shown that the decomposition and aggregation of time series help to extract the deep trends and seasonal changes in the sequence data. Provide a new and more effective PatchTST-PLE hybrid forecasting model for the integrated energy system.

[0004] In summary, in the existing technologies, it is difficult to effectively capture the coupling between multiple loads, and the anti-noise ability, time-frequency resolution, prediction accuracy, as well as the accuracy and balance in the short-term multi-load prediction technology of integrated energy are insufficient. Summary of the Invention

[0005] Based on this, the objective of the present invention is to provide a short-term multi-load prediction method for integrated energy based on secondary decoupling to solve the deficiencies in the above-mentioned existing technologies.

[0006] The present invention provides a short-term multi-load prediction method for integrated energy based on secondary decoupling, and the method includes:

[0007] Collect a multi-load data set, and divide the multi-load data set into a first-half data set and a second-half data set according to seasonal characteristics and data coupling relationships, and reconstruct the winter data set and the summer data set in the multi-load data set based on the first-half data set and the second-half data set for primary decoupling;

[0008] Decompose the reconstructed winter data set and the reconstructed summer data set based on TTAO-VMD to obtain the decomposed data;

[0009] Divide the decomposed data into a periodic term and a trend term through refined composite multi-scale sample entropy for secondary decoupling;

[0010] Use MIV feature screening to screen the trend term, and use the Lasso linear regression model to predict the screened trend term to obtain a trend term prediction result;

[0011] Use Progressive Layered Extraction multi-task learning to perform multi-load prediction on the periodic term to obtain a periodic term prediction result, and add the trend term prediction result and the periodic term prediction result to obtain a final prediction result.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: By reconstructing the winter dataset and the summer dataset, the first coupling is performed. By dividing the decomposed data into a periodic term and a trend term, the secondary decoupling is carried out, so as to effectively capture the coupling of the multi-source load components. By decomposing the reconstructed data through TTAO-VMD, the accurate decomposition of non-linear and non-stationary signals can be realized, the anti-noise ability and the time-frequency resolution are improved. By dividing through the refined composite multi-scale sample entropy, the feature redundancy and data noise in the feature extraction process can be reduced, and the prediction accuracy is improved. And by performing multi-source load prediction on the periodic term through Progressive Layered Extraction multi-task learning, the prediction accuracy can be effectively improved while taking into account the balance.

[0013] Further, the step of dividing the multi-source load dataset into a first-half-year dataset and a second-half-year dataset according to the seasonal characteristics and the data coupling relationship includes:

[0014] Dividing the load sequences in the multi-source load dataset into a number of subsequences;

[0015] Using a random mask to cover part of the data in a number of the subsequences to capture the coupling relationship between the multi-source loads for feature extraction and reconstruction.

[0016] Further, the step of reconstructing the winter dataset and the summer dataset in the multi-source load dataset based on the first-half-year dataset and the second-half-year dataset includes:

[0017] The multi-source load dataset includes spring data, summer data, autumn data and winter data. Arrange the spring data, the summer data, the autumn data and the winter data, and divide the spring data, the summer data, the autumn data and the winter data into a first-half-year data set and a second-half-year dataset according to the seasonal characteristics and the data coupling relationship;

[0018] Based on PtachTST, extract the local information in the multi-source load sequence in the multi-source load dataset, and mine the time dependence relationship of the multi-source load sequence through the multi-head attention mechanism to obtain a feature vector;

[0019] Based on a number of independent attention mechanisms, obtain the attention distribution of different subspaces of the multi-source load sequence;

[0020] Based on BiLSTM, perform bidirectional modeling on the feature vector;

[0021] Train PatchTST-BiLSTM based on the training set to obtain the trained PatchTST-BiLSTM, and adjust the hyperparameters on the training set to obtain the load reconstruction result under the optimal hyperparameter configuration;

[0022] Reconstruct the trained PatchTST-BiLSTM based on the test set to obtain the final IES cloudy load reconstruction result.

[0023] Further, the step of decomposing the reconstructed winter dataset and the reconstructed summer dataset based on TTAO-VMD includes:

[0024] Introduce a dynamic tuning mechanism into the TTAO algorithm;

[0025] Decompose the reconstructed winter dataset and the reconstructed summer dataset according to the TTAO algorithm after introducing the dynamic tuning mechanism.

[0026] Further, the adjustment formula for the central frequency of the TTAO algorithm is:

[0027] ;

[0028] In the formula, represents the central frequency, represents the th frequency component, represents at time step when, represents at time step when the th central frequency of the frequency component, represents the partial derivative, represents the objective function in the decomposition process, represents the step size of the central frequency;

[0029] The adjustment formula for the bandwidth parameter of the TTAO algorithm is:

[0030] ;

[0031] In the formula, represents the th bandwidth parameter, represents the value of the th bandwidth parameter at time step , represents the step size of the bandwidth parameter.

[0032] Further, the expression of the objective function in the decomposition process is:

[0033] ;

[0034] In the formula, represents the expression of the objective function in the decomposition process, represents the th intrinsic mode function, is The derivative of represents the frequency bandwidth of the signal, represents at time step the th value of the bandwidth parameter.

[0035] Furthermore, the step of dividing the decomposed data into periodic terms and trend terms by refined composite multi-scale sample entropy includes:

[0036] Based on refined composite multi-scale sample entropy and using a fine-grained scale, select the similarity threshold for the decomposed data, and evaluate it in combination with several entropy metrics;

[0037] Introduce refined scale decomposition to capture the complex dynamic features in the decomposed data, so as to divide the decomposed data into periodic terms and trend terms.

[0038] Furthermore, the step of screening the trend terms by MIV feature screening and predicting the screened trend terms by Lasso linear regression model includes:

[0039] Through the MIV model and based on the electrical load, thermal load, and cooling load, perturb the numerical magnitudes of the meteorological data and calendar rules in the trend terms to obtain the MIV values;

[0040] Use the Lasso linear regression model and minimize the loss function with L1 norm penalty to eliminate the preset features in the MIV values.

[0041] Furthermore, the expression of the loss function for minimizing the L1 norm penalty is:

[0042] ;

[0043] In the formula, represents the loss function for minimizing the L1 norm penalty, represents the actual value of the th sample, represents the predicted value of the th sample, represents the regression coefficient of the MIV model, represents the th regression coefficient corresponding to the feature, represents the number of samples, represents the number of features represents the regularization parameter

[0044] Furthermore, the step of using Progressive Layered Extraction multi-task learning to perform multivariate load prediction on the periodic term to obtain the periodic term prediction result includes:

[0045] Based on Progressive Layered Extraction, use a multi-layer expert network and a gated network to extract the deep information in the periodic term;

[0046] Use a fully connected layer as a learnable weight matrix, and assign weights to each layer of the expert network through the Softmax function;

[0047] Integrate the deep information extracted by each layer of the expert network after weight assignment to output the periodic prediction result. Description of the Drawings

[0048] Figure 1 is a flowchart of the short-term prediction method for the integrated energy multi-load based on quadratic decoupling in the embodiment of the present invention;

[0049] Figure 2 is a schematic structural diagram of the graph network model in the embodiment of the present invention;

[0050] Figure 3 is a schematic diagram of predicting using the Lasso linear regression model and Progressive Layered Extraction multi-task learning in the embodiment of the present invention;

[0051] Figure 4 is a schematic diagram of the Progressive Layered Extraction prediction model in the embodiment of the present invention.

[0052] The following specific embodiments will further illustrate the present invention in conjunction with the above-mentioned drawings. Specific Embodiments

[0053] To facilitate the understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0054] It should be noted that when an element is referred to as being "fixedly provided on" another element, it can be directly on the other element or there can also be an intermediate element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there may be an intermediate element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are only for illustrative purposes.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this invention belongs. The terms used herein in the description of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0056] Please refer to Figure 1 and Figure 2 , which shows the short-term prediction method for multiple loads of integrated energy based on secondary decoupling in the first embodiment of the present invention. The method includes steps S1 to S5:

[0057] S1. Collect a multi-load data set, and divide the multi-load data set into a first-half data set and a second-half data set according to seasonal characteristics and data coupling relationships, and reconstruct the winter data set and the summer data set in the multi-load data set based on the first-half data set and the second-half data set for primary decoupling;

[0058] Specifically, step S1 includes steps S11 to S18:

[0059] S11. Divide the load sequence in the multi-load data set into several subsequences;

[0060] S12. Use random masking to cover part of the data in several of the subsequences to capture the coupling relationship between multiple loads for feature extraction and reconstruction;

[0061] It can be understood that dividing the load sequence in the multi-load data set into several subsequences (Patches), that is, regarding a time block as a token. When performing Patch segmentation on the load data, two consecutive Patch regions can be overlapping or non-overlapping, which is determined by the length P of the Patch and the step size S (the non-overlapping region between two consecutive Patches). The load sequence is segmented into subsequences as tokens, represents the length of the time series, and random masking is used to cover part of the data in the Patch to capture the coupling relationship between multiple loads for feature extraction and reconstruction;

[0062] S13. The multi-source load dataset includes spring data, summer data, autumn data, and winter data. Arrange the spring data, summer data, autumn data, and winter data, and divide the spring data, summer data, autumn data, and winter data into the first-half dataset and the second-half dataset according to seasonal characteristics and data coupling relationships.

[0063] S14. Based on PatchTST, extract the local information within the multi-source load sequence in the multi-source load dataset, and mine the temporal dependence relationship of the multi-source load sequence through the multi-head attention mechanism to obtain a feature vector.

[0064] S15. Based on a number of multi-head attention mechanisms, obtain the attention distributions of different subspaces of the multi-source load sequence.

[0065] S16. Based on BiLSTM, perform bidirectional modeling on the feature vector.

[0066] S17. Based on the training set, train PatchTST-BiLSTM to obtain the trained PatchTST-BiLSTM, and adjust the hyperparameters on the training set to obtain the load reconstruction result under the optimal hyperparameter configuration.

[0067] S18. Based on the test set, reconstruct the trained PatchTST-BiLSTM to obtain the final IES cloudy load reconstruction result.

[0068] It can be understood that PatchTST is used as an encoder to extract the local information within the multi-source load sequence, where the multi-head attention mechanism further mines the temporal dependence relationship of the multi-source load, and obtains the attention distributions of different subspaces of the input sequence by running multiple independent attention mechanisms in parallel, improving the model's ability to understand and capture complex dependence relationships.

[0069] It should be noted that BiLSTM is used as a decoder to perform bidirectional modeling on the feature vector obtained by the encoder, train PatchTST-BiLSTM based on the training set, and adjust the hyperparameters on the training set, so as to obtain the load reconstruction result under the optimal hyperparameter configuration. Reconstruct the IES multi-source load prediction model PatchTST-BiLSTM with the optimal hyperparameter configuration on the test set to obtain the final IES multi-source load reconstruction result.

[0070] S2. Based on TTAO-VMD, decompose the reconstructed winter dataset and the reconstructed summer dataset to obtain the decomposed data.

[0071] Specifically, step S2 includes steps S21 to S22:

[0072] S21. Introduce a dynamic tuning mechanism into the TTAO algorithm;

[0073] S22. Decompose the reconstructed winter dataset and the reconstructed summer dataset according to the TTAO algorithm after introducing the dynamic tuning mechanism;

[0074] It should be noted that the TTAO algorithm is based on similar triangles. As the iteration progresses, new vertices are continuously generated in the search space and are used to form similar triangles of different sizes. Each triangle is regarded as a basic evolutionary unit with four entities, namely the three vertices of the triangle and an internal random vertex. By dynamically adjusting the parameters, the decomposition process can better adapt to the time-varying characteristics and complexity of the signal. By introducing a dynamic tuning mechanism, the TTAO algorithm enables the tuning parameters to be adaptively adjusted with the change of the signal, thereby improving the accuracy and stability of the VMD decomposition result;

[0075] It should be noted that in this embodiment, the adjustment formula for the center frequency of the TTAO algorithm is:

[0076] ;

[0077] In the formula, represents the center frequency, represents the th frequency component, represents at time step , represents at time step the center frequency of the th frequency component, represents the partial derivative, represents the objective function in the decomposition process,

[0078] The adjustment formula for the bandwidth parameter of the TTAO algorithm is:

[0079] ;

[0080] In the formula, represents the th bandwidth parameter, represents the value of the th bandwidth parameter at time step , represents the step size of the bandwidth parameter.

[0081] The expression of the objective function in the decomposition process is:

[0082] ;

[0083] In the formula, represents the expression of the objective function in the decomposition process, represents the th intrinsic mode function, is The derivative of represents the frequency bandwidth of the signal, represents at the time step the value of the th bandwidth parameter.

[0084] It should be noted that during the optimization process, the TTAO optimization algorithm dynamically adjusts the step size according to the current signal characteristics and :

[0085] ;

[0086] ;

[0087] In the formula, , respectively represent the parameter for controlling the adjustment amplitude of the center frequency and the parameter for controlling the adjustment amplitude of the bandwidth parameter.

[0088] S3. Divide the decomposed data into periodic terms and trend terms through refined composite multi-scale sample entropy for secondary decoupling;

[0089] Specifically, the step S3 includes steps S31 to S32:

[0090] S31. Based on the refined composite multi-scale sample entropy, select the similarity threshold for the decomposed data using a fine-grained scale and evaluate it in combination with several entropy metrics;

[0091] S32. Introduce a refined scale decomposition to capture the complex dynamic features in the decomposed data, so as to divide the decomposed data into periodic terms and trend terms;

[0092] It can be understood that the refined composite multi-scale sample entropy optimizes the selection of the similarity threshold r through a finer-grained scale decomposition, combines multiple entropy metrics to enhance the stability of the result, can more accurately evaluate the complexity of time series data, introduces a refined scale decomposition to obtain a more accurate entropy value, thereby better capturing the complex dynamic features in the data, and divides the load data into periodic term and trend term components by optimizing the scale decomposition and sample entropy calculation process.

[0093] It should be noted that the refined composite multi-scale sample entropy divides the decomposed data into periodic terms and trend terms. The data shows relatively obvious patterns. The trend term component is relatively stable, and the periodic patterns and change modes of each load in the periodic terms are complex. Therefore, it is the second decoupling.

[0094] S4. Use MIV feature screening to screen the trend terms, and use the Lasso linear regression model to predict the screened trend terms to obtain the trend term prediction result;

[0095] Specifically, the step S4 includes steps S41 to S42:

[0096] S41. Through the MIV model and based on the electrical load, thermal load, and cooling load, perturb the numerical magnitudes of the meteorological data and calendar rules in the trend terms to obtain the MIV values;

[0097] S42. Use the Lasso linear regression model and minimize the loss function with L1 norm penalty to eliminate the preset features in the MIV values;

[0098] It can be understood that there are too many influencing factors composed of weather factors and calendar rules in the trend terms, which may have a counterproductive effect on the final prediction result. The MIV model is added for feature screening. The MIV model is based on the electrical load, thermal load, and cooling load. By perturbing the numerical magnitudes of the input meteorological data and calendar rules, the MIV values of each input feature are obtained. The absolute value of the MIV feature represents the importance of the input feature to the output variables of electrical load, cooling load, and thermal load. The Lasso regression model effectively compresses the model coefficients by minimizing the loss function containing the L1 norm penalty, realizes the automatic elimination of unimportant features, thereby constructing a more concise and easy-to-interpret model, and reduces the risk of overfitting;

[0099] In this embodiment, the expression of the loss function for minimizing the L1 norm penalty is:

[0100] ;

[0101] In the formula, represents the loss function for minimizing the L1 norm penalty, represents the actual value of the th sample, represents the predicted value of the th sample, represents the regression coefficient of the MIV model, represents the regression coefficient corresponding to the th feature, represents the number of samples, represents the number of features, denotes the regularization parameter, which is used to control the strength of the L1 regularization term.

[0102] It should be noted that after quadratic decoupling, for the relatively stable trend term, it reflects the general change trend of the load, has strong regularity, and can obtain relatively accurate prediction results without considering the coupling between loads. Moreover, the MIV feature screening is adopted to further enhance the prediction accuracy. Therefore, the Lasso linear regression model is used for prediction.

[0103] S5. Use Progressive Layered Extraction multi-task learning to perform multivariate load prediction on the periodic term to obtain the periodic term prediction result, and add the trend term prediction result and the periodic term prediction result to obtain the final prediction result;

[0104] It should be noted that for the periodic term component with strong non-stationary characteristics, it represents the periodic change law of each load and the change pattern is relatively complex. It is necessary to comprehensively consider the complex coupling relationship between loads and the independent characteristics of each load, and use Progressive Layered Extraction multi-task learning for multivariate load prediction.

[0105] Specifically, the step S5 includes steps S51 to S53:

[0106] S51. Based on Progressive Layered Extraction, use a multi-layer expert network and a gating network to extract the deep information in the periodic term;

[0107] S52. Use a fully connected layer as a learnable weight matrix, and use the Softmax function to assign weights to each layer of the expert network;

[0108] S53. Integrate the deep information extracted by each layer of the expert network after weight assignment to output the periodic prediction result;

[0109] It can be understood that shared experts and task-specific experts are used to reduce the interference of harmful parameters in general and specific information. Progressive Layered Extraction uses multi-layer experts and gating networks to extract more in-depth information from the underlying expert network and better separate task-specific parameters at the higher layer. A fully connected layer is used as a learnable weight matrix, and then the Softmax function is used to assign weights to each expert subnet. The output towers of each task integrate the information to output the prediction results of each task. Finally, the trend term prediction result and the periodic term prediction result are added to obtain the final prediction result, as specifically shown in Figure 3and Figure 4 as shown

[0110] In specific implementation, IES electricity, cooling, and heating load data from 00:00 on January 1, 2020 to 23:00 on December 31, 2020 in a certain area are collected. The data granularity is 1h, and the data length is 8784. The data is divided into the first half and the second half of the year, and 2000 data are respectively reconstructed in PatchTST according to the coupling between data and seasonal characteristics. Then, the reconstructed data and the original data for several days are respectively used to form complete data for January, February, and March as the winter training set, and data for July, August, and September as the summer training set to train the model. The data distributions in February and August 2021 are used as the winter test set and the summer test set to learn the comprehensive energy consumption patterns in different seasons. The performance of the model is evaluated through the test set, and the results of the test set are used as the prediction accuracy. The meteorological data is collected through a certain official website, including 7 categories such as temperature, wind speed, relative humidity, direct normal irradiance (DNI), global horizontal irradiance (GHI), dew point temperature, and air pressure. At the same time, five calendar rules including the number of months, days, hours, week, and whether it is a holiday are considered as input features. To comprehensively evaluate the model prediction results, root mean square error (RMSE), coefficient of determination ( ), and mean absolute percentage error (MAPE) are selected as the evaluation indicators for each load.

[0111]

[0112]

[0113]

[0114] In the formula, , , respectively represent root mean square error, coefficient of determination, and mean absolute percentage error. represents the number of samples, that is, the total number of observed values or data points in the dataset. represents the predicted value of the T-th sample, that is, the prediction result of the model for the T-th data point. represents the actual value of the T-th sample, that is, the true observed result of the T-th data point. T represents the index of the sample, that is, the T-th data point in the dataset. At the same time, a balanced accuracy (BMA) is defined to measure the overall multi-load, and the model is evaluated considering both accuracy and balance. BMA is defined as:

[0115] ;

[0116] ;

[0117] Wherein, , , are the MAPE of the electrical load, the MAPE of the cooling load, and the MAPE of the heating load respectively, , , are the weights of the electrical load, the weights of the cooling load, and the weights of the heating load respectively. Considering that the research should balance accuracy and balance, the weight coefficients in the formula are taken as 0.3, 0.4, and 0.3 respectively for winter prediction, and 0.3, 0.3, and 0.4 respectively for summer prediction. The smaller the value, the better the model performance.

[0118] To fully reflect the coupling between data and give full play to the performance of the program, at the beginning of the experiment, the electrical, cooling, and heating loads are all converted to KW. The conversion formula is:

[0119] ;

[0120] ;

[0121] After the experiment, it is converted back to the original unit for performance evaluation.

[0122] The specific implementation process is as follows. First, the original multi-load sequence is divided into blocks, and the original IES multi-load sequence is divided into several subsequences (Patches). Then, each Patch is regarded as a token, and three encoding operations are performed. They are token encoding, position encoding, and time encoding. Among them, token encoding is used to extract preliminary time features of each token based on the time convolutional layer, position encoding is used to assign position information to each token using the sine-cosine encoding method of the Transformer, and time encoding is used to provide the time stamp information of the data for the model. The encoded tokens that are independent of each other are sent to the Transformer encoder, and a global dependence relationship is established based on the multi-head attention mechanism of the Transformer. Nonlinear feature extraction of the local information in each token is performed through a position feed-forward fully connected neural network. Finally, BiLSTM is used to capture bidirectional time dependence relationships, acting as a decoder, and projected to the prediction window dimension through a fully connected layer to output the reconstruction result.

[0123] When dividing the load data into Patches, two consecutive Patch regions can be overlapping or non-overlapping, which is determined by the length of the Patch and the step size (the non-overlapping region between two consecutive Patches). Finally, the load sequence is divided into Subsequences are used as tokens for feature extraction. This way of dividing the load data into subsequences reduces the number of input tokens from the sequence length to , reducing the time and space complexity of the traditional Transformer. On the other hand, it allows the model to model longer historical load sequences and retain local information of the load. Multiple experiments have proven that in this paper's experiments, when the length of the load input sequence is the past six months, the length of the Patch is 10, and the step size is 10, the effect is better. That is, it is processed in a non-overlapping manner, with each Patch being local load sequence information for 10 hours. The three columns of electrical load, cooling load, and heating load are used as the input sequence for coupled reconstruction, and the remaining columns are used as the feature sequence. Each Patch constitutes the global multi-load sequence information for the past six months.

[0124] Traditional signal decomposition methods such as VMD use fixed tuning parameters for decomposition, which may lead to a decrease in decomposition accuracy when dealing with non-stationary signals. The TTAO optimization algorithm improves the accuracy and stability of the decomposition results by introducing a dynamic tuning mechanism that enables the tuning parameters to adaptively adjust with the changes in the signal. The core of the TTAO optimization algorithm lies in dynamically adjusting the parameters in the decomposition process through time-varying parameter tuning and adaptive optimization strategies.

[0125] Assume that the signal to be decomposed is , the TTAO optimization algorithm optimizes the decomposition process by dynamically adjusting the central frequency and the bandwidth parameter . The adjustment formula for the central frequency is:

[0126] ;

[0127] The adjustment formula for the bandwidth parameter is:

[0128] ;

[0129] where, is the objective function in the decomposition process (such as the reconstruction error of the signal or the frequency bandwidth constraint), and are the step sizes of the central frequency and the bandwidth parameter respectively, which are adaptively adjusted according to the time-varying characteristics of the signal.

[0130] The objective function of the TTAO optimization algorithm can be the minimization of the signal reconstruction error , where, is the th intrinsic mode function (IMF), is The derivative represents the frequency bandwidth of the signal.

[0131] During the optimization process, the TTAO optimization algorithm dynamically adjusts the step size according to the current signal characteristics and :

[0132] ;

[0133] ;

[0134] Among them, and are parameters that control the adjustment amplitude of the step size. The function represents the sign of the derivative and is used to control the increase or decrease of the step size. Then, the Refined Composite Multiscale Sample Entropy is introduced. The Refined Composite Multiscale Sample Entropy optimizes the selection of the similarity threshold through a finer-grained scale decomposition, combines multiple entropy metrics to enhance the stability of the results, and can more accurately evaluate the complexity of time series data. Progressive Layered Extraction divides the expert subnets into two types: one is the shared expert subnet (Shared), and the other is the task-specific expert subnet for each sub-task (Task-Specific). Therefore, this design retains the transfer learning ability between tasks through the shared expert subnet and can avoid the noise interference of harmful parameters when the correlation between different tasks is weak, thus avoiding the negative transfer phenomenon.

[0135] Since there are too many influencing factors composed of weather factors and calendar rules in the trend term, which may have a counterproductive effect on the final prediction result, the MIV model is added for feature screening. The MIV model is based on the electrical load, heat load, and cooling load. By perturbing the numerical values of the input meteorological data and calendar rules, the MIV values of each input feature are obtained, and the absolute value represents the importance of the input feature to the output variables of electrical load, cooling load, and heat load. The Lasso regression model effectively compresses the model coefficients by minimizing the loss function containing the L1 norm penalty, realizes the automatic elimination of unimportant features, and thus constructs a more concise and interpretable model. This sparsity not only improves the generalization ability of the model but also reduces the risk of overfitting.

[0136] In the trend term, considering that there are too many influencing factors composed of weather factors and calendar rules, which may have a counterproductive effect on the final prediction result, the MIV model is added for feature screening. The MIV model is based on the electrical load, heat load, and cooling load. By perturbing the numerical values of the input meteorological data and calendar rules, the MIV values of each input feature are obtained, and the absolute value represents the importance of the input feature to the output variables of electrical load, cooling load, and heat load. The Lasso regression model effectively compresses the model coefficients by minimizing the loss function containing the L1 norm penalty, realizes the automatic elimination of unimportant features, and thus constructs a more concise and interpretable model. This sparsity not only improves the generalization ability of the model but also reduces the risk of overfitting.

[0137] The comparison results of the prediction models are shown in Table 1:

[0138] Table 1

[0139]

[0140] As can be seen from Table 1, the proposed model has the highest accuracy in predicting heat load, cooling load and electrical load whether in winter or summer, and the value in the mean absolute percentage error of the final total balance accuracy is also the smallest, fully indicating that the method proposed in this paper has the best effect both in prediction accuracy and balance.

[0141] In summary, the short-term prediction method for multiple loads of integrated energy based on quadratic decoupling in the above embodiments of the present invention applies the PatchTST model to multiple load prediction. By splitting the time series into blocks and improving the model's ability to handle long-term dependencies, it effectively captures the coupling between multiple loads, and considers the seasonal characteristics of meteorological factors and calendar rules to reconstruct some data, thus achieving a significant performance improvement in the multi-variable time series prediction task; The TTAO-VMD algorithm effectively improves the time-frequency analysis performance in signal processing by combining the triangular topology aggregation optimizer and variational mode decomposition technology, realizes the accurate decomposition of non-linear and non-stationary signals, and demonstrates excellent anti-noise ability and time-frequency resolution; A feature processing method is proposed, which divides the input features into two groups for differential processing according to the difference of refined composite multi-scale sample entropy (refined composite multi-scale sample entropy) to pay more attention to historical load data. Through this method, feature redundancy and data noise in the feature extraction process can be reduced, thereby improving the prediction accuracy of the model; Construct an improved multi-task learning PatchTST-PLE model. In MMoE, multiple tasks share all experts, and different tasks control different expert coefficients through gate weights. In PatchTST-PLE, in addition to shared experts, multiple tasks also have private experts, and consider the interaction features of different experts, so as to achieve both balance while improving prediction accuracy.

[0142] In the description of this specification, the descriptions referring to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0143] The above-described embodiments merely represent several implementation manners of the present invention. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention shall be subject to the appended claims.

Claims

1. A short-term forecasting method for comprehensive energy multi-load based on secondary decoupling, characterized in that: The method comprises: Collecting a multivariate load data set, and dividing the multivariate load data set into a first half data set and a second half data set according to seasonal characteristics and data coupling relationships, and reconstructing a winter data set and a summer data set in the multivariate load data set based on the first half data set and the second half data set to perform a decoupling; The step of reconstructing the winter data set and the summer data set in the multivariate load data set based on the first half data set and the second half data set comprises: The multivariate load data set includes spring data, summer data, autumn data and winter data, the spring data, summer data, autumn data and winter data are arranged, and the spring data, summer data, autumn data and winter data are divided into first half data and second half data sets according to seasonal characteristics and data coupling relationships; Extracting local information in the multivariate load sequence in the multivariate load data set based on PtachTST, and mining the time dependency of the multivariate load sequence through Multi-Head Attention to obtain a feature vector; Obtaining attention distributions of different subspaces of the multivariate load sequence based on a plurality of independent attention mechanisms; Performing bidirectional modeling on the feature vector based on BiLSTM; Training PatchTST-BiLSTM based on the training set to obtain a trained PatchTST-BiLSTM, and adjusting hyperparameters on the training set to obtain a load reconstruction result under an optimal hyperparameter configuration; Reconstruct the trained PatchTST-BiLSTM based on the test set to obtain the final IES multi-cloud load reconstruction result; Decomposing the reconstructed winter data set and the reconstructed summer data set based on TTAO-VMD to obtain decomposed data; Dividing the decomposed data into periodic items and trend items by fine composite multi-scale sample entropy to perform secondary decoupling; The trend items are screened by using MIV feature screening, and the trend items after screening are predicted by using Lasso linear regression model to obtain trend item prediction results; Progressive Layered Extraction multi-task learning is used to perform multivariate load forecasting on the periodic item to obtain a periodic item prediction result, and the trend item prediction result and the periodic item prediction result are added together to obtain a final prediction result.

2. The method for short-term forecasting of comprehensive energy multi-load based on secondary decoupling according to claim 1 is characterized in that: The step of dividing the multivariate load data set into a first half data set and a second half data set according to seasonal characteristics and data coupling relationships comprises: Dividing the load sequence in the multivariate load data set into a plurality of subsequences; A random mask is used to cover part of the data in some of the subsequences to capture the coupling relationship between the multivariate loads for feature extraction and reconstruction.

3. The method for short-term forecasting of comprehensive energy multi-load based on secondary decoupling according to claim 1 is characterized in that: The step of decomposing the reconstructed winter data set and the reconstructed summer data set based on TTAO-VMD comprises: Introducing dynamic tuning mechanism into TTAO algorithm; The reconstructed winter data set and the reconstructed summer data set are decomposed according to the TTAO algorithm after the dynamic tuning mechanism is introduced.

4. The method for short-term forecasting of comprehensive energy multi-load based on secondary decoupling according to claim 3 is characterized in that: The adjustment formula of the center frequency of the TTAO algorithm is: ； In the formula, represents the center frequency, Indicates frequency components, Indicates that at time step hour, Indicates that at time step The The center frequency of a frequency component, represents the partial derivative, represents the objective function in the decomposition process, represents the step size of the center frequency; The adjustment formula of the bandwidth parameter of the TTAO algorithm is: ； In the formula, Indicates bandwidth parameter, Indicates that at time step The The value of the bandwidth parameter, Indicates the step size of the bandwidth parameter.

5. The method for short-term forecasting of comprehensive energy multi-load based on secondary decoupling according to claim 4 is characterized in that: The expression of the objective function in the decomposition process is: ； In the formula, The expression representing the objective function in the decomposition process, Indicates The intrinsic mode functions, for The derivative of represents the frequency bandwidth of the signal, Indicates that at time step The The value of the bandwidth parameter.

6. The method for short-term forecasting of comprehensive energy multi-load based on secondary decoupling according to claim 1 is characterized in that: The step of dividing the decomposed data into periodic items and trend items by fine composite multi-scale sample entropy comprises: Selecting a similarity threshold for the decomposed data based on fine composite multi-scale sample entropy and using a fine-grained scale, and evaluating it in combination with several entropy metrics; Fine scale decomposition is introduced to capture the complex dynamic features in the decomposed data, so as to divide the decomposed data into periodic items and trend items.

7. The method for short-term forecasting of comprehensive energy multi-load based on secondary decoupling according to claim 1 is characterized in that: The step of using MIV feature screening to screen the trend items and using Lasso linear regression model to predict the screened trend items includes: The meteorological data and the numerical values ​​of the calendar rules in the trend item are disturbed by the MIV model based on the electric load, the heating load and the cooling load to obtain the MIV value; The Lasso linear regression model is adopted and the preset features in the MIV value are eliminated by minimizing the loss function of the L1 norm penalty.

8. The method for short-term forecasting of comprehensive energy multi-load based on secondary decoupling according to claim 7 is characterized in that: The expression of the loss function that minimizes the L1 norm penalty is: ； In the formula, represents the loss function that minimizes the L1 norm penalty, Indicates The actual value of the samples, Indicates The predicted value of samples, represents the regression coefficient of the MIV model, Indicates The regression coefficient corresponding to the feature is represents the number of samples, represents the number of features, represents the regularization parameter.

9. The method for short-term forecasting of comprehensive energy multi-load based on secondary decoupling according to claim 1 is characterized in that: The step of using Progressive Layered Extraction multi-task learning to perform multivariate load forecasting on the periodic item to obtain a periodic item prediction result includes: Based on Progressive Layered Extraction, a multi-layer expert network and a gating network are used to extract the depth information in the periodic term; A fully connected layer is used as a learnable weight matrix, and a Softmax function is used to assign weights to each layer of the expert network; The depth information extracted by each layer of the expert network is integrated to output a periodic prediction result.

Images