Comprehensive energy system multi-element load hybrid prediction method, system, device and medium

Abstract

The present application relates to the technical field of multi-element load prediction, and provides a comprehensive energy system multi-element load hybrid prediction method, system, device and medium, which comprises splicing the to-be-analyzed multi-element load sequence and the influence factor sequence generated by processing the multi-element load sequence and the influence factor sequence to obtain a first multi-element load characteristic sequence; analyzing the first multi-element load characteristic sequence based on a correlation evaluation model to generate a correlated load group and an independent load group; splicing the reconstructed multi-element load sequence obtained by singular spectrum analysis of the to-be-analyzed multi-element load sequence with the to-be-analyzed multi-element load sequence and the influence factor sequence to obtain a second multi-element load characteristic sequence; and based on the correlated load group and the independent load group, performing multi-element load hybrid prediction on the multi-element load prediction input characteristics obtained according to a double-attention mechanism to obtain a multi-element load prediction result. The present application can fully capture the complex dynamic coupling characteristics among multi-element loads, and improve the accuracy and robustness of load prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of integrated energy system load prediction, in particular to an integrated energy system multi-element load hybrid prediction method, system, device and medium. BACKGROUND

[0002] Multi-element load prediction is a key support for integrated energy system (IES) to improve energy utilization efficiency and energy supply reliability through multi-energy complementation and collaborative optimization. However, there is a complex coupling relationship between the various heterogeneous loads such as electricity, heat and gas in IES, making it difficult to accurately predict multi-element load.

[0003] In order to effectively utilize the coupling relationship between multi-element loads, the existing multi-element load prediction method usually uses the maximum information coefficient to analyze the correlation of multi-element loads in advance, and statically divides the loads into "strongly correlated" groups and "weakly correlated" loads. Then, multi-task learning (MTL) is used to capture the coupling features between the loads in the "strongly correlated" group, and single-task learning (STL) is used to process specific "weakly correlated" loads. However, the correlation of multi-element loads in the actual integrated energy system is not constant, but is affected by various factors such as seasonal changes, day type differences, sudden events and operation mode adjustments, showing dynamic changes. For example, two kinds of loads may be highly correlated in summer, but their correlation may significantly weaken in winter, or even produce negative correlation, which will lead to the fact that the loads originally considered to be "weakly correlated" may produce new and unexpected interference with other loads in certain periods, and the coupling relationship of the loads originally considered to be "strongly correlated" may weaken in certain periods. In this case, the pre-set static grouping cannot adapt to such dynamic changes, so that MTL not only cannot effectively play a collaborative role, but also introduces noise, thereby reducing the prediction accuracy and model robustness. SUMMARY

[0004] The purpose of the present application is to provide an integrated energy system multi-element load hybrid prediction method, which can effectively avoid the negative impact of inaccurate load coupling relationship identification on overall prediction performance by introducing periodic dynamic load correlation evaluation to fully capture the dynamic coupling feature changes between multi-element loads and adaptively adjusting the load collaborative prediction strategy, thereby improving the accuracy and robustness of integrated energy system multi-element load prediction.

[0005] In order to achieve the above-mentioned purpose, it is necessary to provide an integrated energy system multi-element load hybrid prediction method, system, device and medium in view of the above technical problems.

[0006] In a first aspect, an embodiment of the present application provides a comprehensive energy system multi-element load hybrid prediction method, the method comprising:

[0007] Periodically acquire multi-element load sequences and influence factor sequences and preprocess to generate multi-element load sequences and influence factor sequences to be analyzed;

[0008] Time-align and splice the multi-element load sequences to be analyzed and the influence factor sequences to obtain a first multi-element load feature sequence;

[0009] According to the first multi-element load feature sequence, perform load correlation analysis based on a pre-constructed correlation evaluation model to generate a load correlation matrix, and according to the load correlation matrix, generate a correlated load group and an independent load group;

[0010] Perform singular spectrum decomposition and reconstruction on the multi-element load sequences to be analyzed to obtain reconstructed multi-element load sequences, and time-align and splice the reconstructed multi-element load sequences, the multi-element load sequences to be analyzed, and the influence factor sequences to obtain a second multi-element load feature sequence;

[0011] Respectively perform double-attention mechanism fusion analysis on each load feature sequence in the second multi-element load feature sequence to obtain multi-element load prediction input features;

[0012] According to the multi-element load prediction input features, perform multi-element load hybrid prediction based on the correlated load group and the independent load group to obtain multi-element load prediction results.

[0013] Further, the correlation evaluation model comprises a feature extraction layer and a similarity analysis layer connected in sequence; the feature extraction layer comprises two feature extraction sub-networks sharing weights; the feature extraction sub-networks comprise convolution layers and recurrent encoding layers.

[0014] Further, the similarity analysis layer is used for performing cosine similarity analysis on two load feature vectors output by the feature extraction layer to obtain corresponding load correlation scores.

[0015] Further, the step of generating a correlated load group and an independent load group according to the load correlation matrix comprises:

[0016] According to the load correlation matrix, obtain a load having a load correlation score with other loads less than a preset correlation score threshold to generate the independent load group; the preset correlation score threshold is adaptively adjusted and optimized based on a grid search method;

[0017] According to the multi-element load, generate the correlated load group from the load remaining in the independent load group.

[0018] Further, the step of performing singular spectrum decomposition and reconstruction on the to-be-analyzed poly-element load sequence to obtain a reconstructed poly-element load sequence comprises:

[0019] performing singular spectrum decomposition on each to-be-analyzed load sequence in the to-be-analyzed poly-element load sequence to obtain a corresponding plurality of frequency load subsequences;

[0020] reconstructing the frequency load subsequence of each to-be-analyzed load sequence based on the variance contribution degree of the subsequence to obtain a corresponding reconstructed load sequence;

[0021] collecting all the reconstructed load sequences to obtain the reconstructed poly-element load sequence.

[0022] Further, the step of performing poly-element load mixed prediction based on the associated load group and the independent load group according to the poly-element load prediction input features to obtain a poly-element load prediction result comprises:

[0023] performing collaborative prediction based on a pre-constructed multi-task learning model according to all load prediction input features corresponding to the associated load group in the poly-element load prediction input features to obtain a predicted value of each load in the associated load group;

[0024] inputting each load prediction input feature corresponding to the independent load group in the poly-element load prediction input features into a corresponding single-task learning model for separate prediction to obtain a predicted value of each load in the independent load group;

[0025] collecting the predicted values of each load in the associated load group and the predicted values of each load in the independent load group to obtain the poly-element load prediction result.

[0026] Further, the step of performing collaborative prediction based on a pre-constructed multi-task learning model according to all load prediction input features corresponding to the associated load group in the poly-element load prediction input features to obtain a predicted value of each load in the associated load group comprises:

[0027] performing feature encoding on each load prediction input feature corresponding to the associated load group based on a pre-set shared encoder to obtain a corresponding load encoded feature;

[0028] performing weighted summation on all related load encoded features corresponding to each load encoded feature based on the corresponding load correlation score in the load correlation matrix to obtain a corresponding collaborative encoded feature;

[0029] The load coding features and the corresponding collaborative coding features are spliced and fused to obtain corresponding load fusion features, and the load fusion features are input into a corresponding decoder for load prediction to obtain a predicted value of the corresponding load.

[0030] In a second aspect, an embodiment of the present application provides a comprehensive energy system multi-element load hybrid prediction system, which comprises:

[0031] A data acquisition module is configured to periodically acquire and pre-process multi-element load sequences and influence factor sequences to generate multi-element load sequences and influence factor sequences to be analyzed;

[0032] A first splicing module is configured to align and splice the multi-element load sequences to be analyzed and the influence factor sequences to obtain a first multi-element load feature sequence;

[0033] A correlation analysis module is configured to perform correlation analysis between loads based on a pre-constructed correlation evaluation model according to the first multi-element load feature sequence, generate a load correlation matrix, and generate a correlated load group and an independent load group according to the load correlation matrix;

[0034] A second splicing module is configured to perform singular spectrum decomposition and reconstruction on the multi-element load sequences to be analyzed to obtain reconstructed multi-element load sequences, and align and splice the reconstructed multi-element load sequences, the multi-element load sequences to be analyzed, and the influence factor sequences to obtain a second multi-element load feature sequence;

[0035] A fusion analysis module is configured to perform double-attention mechanism fusion analysis on each load feature sequence in the second multi-element load feature sequence to obtain multi-element load prediction input features;

[0036] A load prediction module is configured to perform multi-element load hybrid prediction based on the correlated load group and the independent load group according to the multi-element load prediction input features to obtain multi-element load prediction results.

[0037] In a third aspect, an embodiment of the present application further provides a computer device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program.

[0038] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the steps of the above method.

[0039] The application provides a comprehensive energy system multi-element load mixed prediction method, system, device and medium, through the method, periodic multi-element load sequence and influence factor sequence are acquired and pretreated to generate to-be-analyzed multi-element load sequence and influence factor sequence, then the to-be-analyzed multi-element load sequence and the influence factor sequence are time-aligned and spliced to obtain a first multi-element load characteristic sequence, according to the first multi-element load characteristic sequence, a correlation evaluation model is constructed in advance, correlation analysis between loads is performed to generate a load correlation matrix, and according to the load correlation matrix, a related load group and an independent load group are generated, singular spectrum decomposition and reconstruction are performed on the to-be-analyzed multi-element load sequence to obtain a reconstructed multi-element load sequence, the reconstructed multi-element load sequence, the to-be-analyzed multi-element load sequence and the influence factor sequence are time-aligned and spliced to obtain a second multi-element load characteristic sequence, each load characteristic sequence in the second multi-element load characteristic sequence is subjected to double-attention mechanism fusion analysis to obtain a multi-element load prediction input feature, and according to the multi-element load prediction input feature, a multi-element load mixed prediction is performed based on the related load group and the independent load group to obtain a multi-element load prediction result. Compared with the prior art, the comprehensive energy system multi-element load mixed prediction method can effectively avoid the negative influence of inaccurate load coupling relationship identification on the overall prediction performance, and improve the accuracy and robustness of the comprehensive energy system multi-element load prediction. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of the comprehensive energy system multi-element load mixed prediction method in the embodiment of the application;

[0041] Figure 2 is a prediction result comparison diagram of the prediction method proposed in the embodiment of the application and the multi-task learning architecture prediction model based on LSTM or GRU for predicting and analyzing cold load, gas load, heat load and electric load in the comprehensive energy system;

[0042] Figure 3 is a structural diagram of the comprehensive energy system multi-element load mixed prediction system in the embodiment of the application;

[0043] Figure 4 is an internal structure diagram of the computer device in the embodiment of the application;

[0044] Among them, the reference signs are:

[0045] 1, data acquisition module; 2, first splicing module; 3, correlation analysis module; 4, second splicing module; 5, fusion analysis module; 6, load prediction module. DETAILED DESCRIPTION

[0046] In order to make the purposes, technical solutions and beneficial effects of the present application clearer, further detailed description will be given below in combination with the drawings and examples. Obviously, the following described examples are a part of the embodiments of the present application and are only used to illustrate the present application, but not to limit the scope of the present application. Based on the examples in the present application, all other examples obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0047] The comprehensive energy system multi-load hybrid prediction method provided by the present application can be understood as a multi-load hybrid prediction method for periodically and dynamically capturing the complex coupling relationship between multi-loads and adaptively adjusting the load collaborative prediction strategy, based on the application status that the existing comprehensive energy system multi-load prediction adopts static grouping, which is difficult to dynamically capture the complex coupling relationship between multi-loads, resulting in poor load prediction effect. The comprehensive energy system multi-load hybrid prediction method can ensure the efficiency of multi-load prediction while significantly improving the accuracy and robustness of load prediction. The following examples will describe the comprehensive energy system multi-load hybrid prediction method in detail.

[0048] In one embodiment, as shown in Figure 1 A comprehensive energy system multi-load hybrid prediction method is provided, comprising the following steps:

[0049] S11, periodically acquire multi-load sequence and influence factor sequence and pre-process to generate multi-load sequence and influence factor sequence to be analyzed; wherein the multi-load sequence can be understood as the historical load time series data of the comprehensive energy system based on the current prediction time and the prediction time window (the current prediction time is backtracked to the latest historical time series data, for example, in the scenario of predicting every hour, the prediction time point is t, and the prediction time window can be set as [t-24, t], that is, the past 24 hours of data are used for subsequent dynamic analysis), which can include cold load sequence, heat load sequence, gas load sequence and electric load sequence, etc. The size of the prediction time window can be determined according to the required data time interval length of the actual load prediction task; the corresponding influence factor sequence can be understood as the load-related influence factor sequence that will affect the actual load, which can include meteorological data sequence (including temperature, humidity, air pressure and wind speed, etc.), timestamp sequence, holiday information sequence and running state sequence, etc., and the running state includes the start / stop state of the refrigeration / heating equipment.

[0050] The acquisition process of the multivariate load sequence and the influence factor sequence to be analyzed in the embodiment can be understood as performing data cleaning, missing value filling, abnormal value correction, time window alignment, and normalization on each sequence in the multivariate load sequence and the influence factor sequence to obtain a multidimensional sequence for subsequent analysis. The specific processing process can be implemented by referring to related prior art. It should be noted that the time stamp information in the influence factor sequence can be represented in the form of one-hot encoding, and the holiday information can be represented in the form of binary, for example, 1 represents a holiday, and 0 represents a non-holiday.

[0051] S12, time aligning and splicing the multivariate load sequence to be analyzed and the influence factor sequence to obtain a first multivariate load feature sequence; wherein the first multivariate load feature sequence includes a plurality of load feature sequences, and each load feature sequence can be understood as a data sequence obtained by horizontally splicing the data at the same time in each load sequence in the multivariate load sequence to be analyzed and each factor sequence in the influence factor sequence. The complete feature data at each time in the data sequence includes a load feature value and a plurality of influence factor values.

[0052] S13, according to the first multivariate load feature sequence, performing load correlation analysis based on a pre-constructed correlation evaluation model to generate a load correlation matrix, and generating a correlated load group and an independent load group according to the load correlation matrix; wherein the load correlation matrix can be understood as a matrix for dynamically maintaining the results of load correlation analysis, which is obtained by performing correlation analysis on the load feature sequences in the first multivariate load feature sequence in pairs. In order to ensure the efficiency and accuracy of the load correlation analysis, the embodiment preferably proposes a load correlation evaluation model based on deep learning, and considering the characteristics of the relationship between load changes and influence factors, the load influence factors are introduced for correlation analysis to improve the comprehensiveness and reliability of the load correlation analysis.

[0053] In practical applications, the load feature sequences in the first multi-load feature sequence are paired two by two to input a correlation evaluation model to evaluate the correlation between two loads, until the correlation evaluation results of all loads are obtained to generate a required load correlation matrix. Specifically, the correlation evaluation model includes a feature extraction layer and a similarity analysis layer connected in sequence. The feature extraction layer includes two feature extraction sub-networks sharing weights, and each feature extraction sub-network is used for feature vector extraction of a load feature sequence. The feature extraction sub-network includes a convolution layer and a recurrent encoding layer, which are used to extract local and global time sequence patterns of load task features. The convolution layer can use a stacked one-dimensional convolutional neural network (1D-CNN) to enhance the local feature extraction capability, and use a sliding window method to extract load change patterns along the time axis. The parameters can be adjusted and optimized according to the complexity of the actual task or the characteristics of the data. That is, the number of stacked one-dimensional convolutional neural networks can be set according to the actual application requirements. For example, when two layers of one-dimensional convolutional neural networks are stacked, the convolution kernel size can be set to 3 and 64, respectively. In addition, a ReLU activation function and a normalization layer are used after each convolution layer to improve the nonlinear modeling capability and improve the model convergence efficiency. The recurrent encoding layer can be an LSTM recurrent layer including a multi-layer stacked structure, and each LSTM recurrent layer is a complete LSTM network layer, which is used to capture long-term dependencies in load time series. The number of layers and the hidden state dimension can be adjusted according to the complexity of the task and the modeling requirements. For example, when a two-layer stacked structure is used, the number of neurons in each hidden layer can be set to 128, and the corresponding activation function can be tanh. The state vector output by the last hidden layer is used to generate the required load feature vector.

[0054] The similarity analysis layer in the correlation evaluation model can be understood as a correlation analysis of the two load feature vectors output by the two feature extraction sub-networks in the feature extraction layer of the correlation evaluation model, to obtain a correlation score between 0 and 1, to reflect the correlation strength of the two load prediction tasks in the current period, and to effectively capture complex nonlinear dynamic coupling relationships. Specifically, the similarity analysis layer is used to perform cosine similarity analysis on the two load feature vectors output by the feature extraction layer to obtain the corresponding load correlation score.

[0055] By introducing multiple load influencing factors and using the correlation evaluation model to evaluate the correlation between loads, the embodiment can efficiently and accurately capture the dynamic coupling relationship between loads, and provide a reliable basis for subsequent load prediction task grouping.

[0056] The load correlation matrix obtained through the above method steps is dynamically adjusted based on the actual obtained multivariate load sequence and influence factor sequence of each period, to guide dynamic grouping division of the load prediction task. Specifically, the step of generating the associated load group and the independent load group according to the load correlation matrix comprises:

[0057] According to the load correlation matrix, loads with a load correlation score less than a preset correlation score threshold value with other loads are obtained to generate the independent load group; the preset correlation score threshold value is adaptively adjusted and optimized based on a grid search method; in principle, the preset correlation score threshold value can be set according to experience, but considering that unreasonable fixed experience value setting may lead to actual application failure in actual application, in order to ensure the reliability of dynamic load task grouping, the preset correlation score threshold value is preferably set to an experience value (such as 0.3) in initial application, and is adaptively adjusted and optimized based on the grid search method in subsequent periodic prediction analysis. Specifically, the step of optimizing the preset correlation score threshold value using the grid search method can comprise: setting a candidate threshold set Θ = {0.1, 0.2,..., 0.9}, and setting a training set and a validation set comprising multiple groups of multivariate load sequences, influence factor sequences and corresponding multivariate load prediction results; for each candidate threshold θ ∈ Θ, a complete training and prediction process is run on the training set and the validation set, and the average MSE (Mean Square Error) comprehensive evaluation index of the corresponding load prediction is recorded respectively, to select the threshold with the optimal comprehensive index on the validation set as the preset correlation score threshold value used for dividing the associated load group and the independent load group in the final prediction process.

[0058] According to the loads remaining after excluding the loads in the independent load group from the multivariate loads, the associated load group is generated.

[0059] In actual application, whether the correlation between each load and all other loads belongs to the category of weak correlation or no correlation is judged according to the load correlation scores in the load correlation matrix in turn; if the load correlation score between a certain load and all other loads is less than the preset correlation score threshold value, it is considered that the coupling correlation between the load and other loads is weak, and the load needs to be independently predicted and added to the independent load group for separate load prediction in the current prediction period; if the load correlation score between a certain load and any other load is greater than or equal to the preset correlation score threshold value, it is considered that the load needs to be co-predicted with other loads in the current prediction period and added to the associated load group.

[0060] The example can capture dynamic coupling characteristic changes among multiple loads through periodic analysis, intelligently and dynamically divide associated load groups and independent load groups, and realize adaptive adjustment of load collaborative prediction strategies based on dynamic coupling characteristic changes among loads, thereby effectively avoiding performance degradation caused by incorrect load prediction task association binding, and further ensuring the reliability of multiple load grouping prediction.

[0061] S14, singular spectrum decomposition and reconstruction are performed on the to-be-analyzed multiple load sequence to obtain a reconstructed multiple load sequence, and the reconstructed multiple load sequence, the to-be-analyzed multiple load sequence, and the influence factor sequence are time-aligned and spliced to obtain a second multiple load characteristic sequence; the reconstructed multiple load sequence can be understood as a multiple-dimensional sequence obtained by performing singular spectrum analysis (SSA) on each to-be-analyzed load sequence in the to-be-analyzed multiple load sequence, selecting important subsequences, and reconstructing, and simultaneously including multiple reconstructed load sequences.

[0062] Specifically, the step of performing singular spectrum decomposition and reconstruction on the to-be-analyzed multiple load sequence to obtain a reconstructed multiple load sequence includes:

[0063] Each to-be-analyzed load sequence in the to-be-analyzed multiple load sequence is subjected to singular spectrum decomposition to obtain a plurality of frequency load subsequences corresponding thereto; the singular spectrum decomposition can be understood as decomposing each to-be-analyzed load sequence to obtain trend items, periodic items, and residual items, and the specific implementation process can refer to the implementation of related prior art, which is not described in detail here.

[0064] Each frequency load subsequence of the to-be-analyzed load sequence is screened and reconstructed based on subsequence variance contribution, to obtain a corresponding reconstructed load sequence; the subsequence variance contribution can be understood as the variance contribution of each subsequence to the original to-be-analyzed load sequence; in actual application, the variance contribution of each frequency load subsequence to the corresponding to-be-analyzed load sequence is calculated, and then the subsequence is arranged in descending order according to the variance contribution, and the subsequence with a cumulative variance contribution rate exceeding a certain proportion (such as 90%) is retained for sequence reconstruction, so as to enhance the ability to capture the inherent change characteristics of the load; for example, if the cumulative variance contribution rate of the first four subsequences reaches 90%, the first four subsequences are reconstructed, and the remaining subsequences are discarded.

[0065] All the reconstructed load sequences are summarized to obtain the reconstructed multiple load sequence.

[0066] The embodiment can effectively separate the trend item and the periodic item by adopting the singular spectrum decomposition reconstruction method, significantly reduce the noise interference, and further improve the stationarity and predictability of the sequence, thereby providing clearer structural feature support for the subsequent learning process. Meanwhile, by splicing and fusing the obtained reconstructed multivariate load sequence with the to-be-analyzed multivariate load sequence and the influence factor sequence as the input of the subsequent prediction model, the model can not only capture the evolution law of the load itself through the reconstructed multivariate load sequence, but also perceive the influence of external factors on the load, thereby achieving more comprehensive and accurate load prediction.

[0067] S15, each load feature sequence in the second multivariate load feature sequence is subjected to double-attention mechanism fusion analysis to obtain multivariate load prediction input features.

[0068] The double-attention mechanism can be understood as being composed of a temporal attention mechanism (TAM) and a feature attention mechanism (FAM), and is used for dynamically weighting each load feature sequence from the time and feature dimensions to capture important information in the time sequence or the feature. The temporal attention mechanism automatically identifies and weights key time points that have a significant influence on the prediction result by calculating the correlation between each time point and the prediction time, which helps the model focus on the information of the key historical time points, thereby better grasping the load change trend. The feature attention mechanism automatically assigns appropriate weights by evaluating the contribution of each feature in the load feature sequence to the prediction result, thereby highlighting the role of important features to enhance the discrimination ability of the subsequent prediction model for the input features. It should be noted that the processing processes of the above-mentioned temporal attention mechanism and feature attention mechanism can be implemented by referring to related prior art, and the features obtained by processing each load feature sequence through the temporal attention mechanism and the feature attention mechanism are spliced and fused to obtain the corresponding load prediction input features, which are then combined to form the required multivariate load prediction input features.

[0069] S16, according to the multivariate load prediction input features, performing multivariate load mixed prediction based on the associated load group and the independent load group to obtain a multivariate load prediction result; wherein the multivariate load mixed prediction can be understood as a multivariate load prediction strategy of performing multi-task collaborative prediction on all loads in the associated load group and performing single-task independent prediction on each load in the independent load group; specifically, the step of performing multivariate load mixed prediction based on the associated load group and the independent load group according to the multivariate load prediction input features to obtain a multivariate load prediction result includes:

[0070] According to all the load prediction input features corresponding to the associated load group in the multi-element load prediction input features, collaborative prediction is performed based on a pre-constructed multi-task learning model to obtain the predicted values of each load in the associated load group; in principle, the multi-task learning model can adopt an existing multi-task prediction model, but in order to better handle the complex and dynamic coupling relationship between multiple load prediction tasks, the embodiment preferably adopts a collaborative prediction model based on a temporal convolutional network (TCN) as a basic prediction structure design, learns the common change mode and coupling relationship features of each related load in the associated load group through a shared TCN encoder, and further learns the individualized features of each load prediction task through a dedicated decoder to finally realize efficient modeling and analysis of the complex sequences of each load in the associated load group.

[0071] Specifically, the step of performing collaborative prediction based on a pre-constructed multi-task learning model according to all the load prediction input features corresponding to the associated load group in the multi-element load prediction input features to obtain the predicted values of each load in the associated load group includes:

[0072] The feature of each load prediction input feature corresponding to the associated load group is encoded based on a preset shared encoder to obtain the corresponding load encoding feature; the preset shared encoder can be understood as a TCN encoder pre-trained based on related load feature data, and the specific process of encoding the feature of each load prediction input feature corresponding to the associated load group using the encoder can be realized by referring to the related application technology of the existing TCN encoder, which will not be described in detail here.

[0073] All the related load encoding features corresponding to each load encoding feature are weighted and summed based on the corresponding load correlation scores in the load correlation matrix to obtain the corresponding collaborative encoding feature; the collaborative encoding feature can be understood as a shared feature that each load in the associated load group needs to obtain from all the loads with which it has a strong correlation; considering that the load correlation scores between the same load and different loads are different, the corresponding collaborative values are also different, the embodiment preferably adopts a weighted collaborative modeling mechanism driven by the load correlation matrix to obtain the collaborative encoding feature corresponding to each load prediction task, that is, the corresponding load correlation scores in the load correlation matrix are used to guide different load prediction tasks to share features to different degrees. The implementation process of the weighted collaborative modeling mechanism driven by the load correlation matrix will be described below by taking the collaborative encoding feature acquisition process of a certain load i as an example:

[0074] Suppose that each load prediction input feature of the current associated load group S1 is represented as h1, h2,..., h after feature encoding by a preset shared encoder, and the load correlation score between load i and relevant load j in the associated load group is represented as k . k represents the number of loads in the associated load group S1.

[0075] The load correlation score between load j and load i is normalized based on the following formula to obtain data as the shared weight provided by the load coding feature of load j for the prediction task of load i:

[0076]

[0077] In the formula, represents the load correlation score between load j and load i in the associated load group S1. represents the load correlation score between load l and load i in the associated load group S1. represents the shared weight provided by the load coding feature of load j for the prediction task of load i.

[0078] Based on all the shared weights calculated by the above method, the load coding features of all the loads related to load i in the associated load group are weighted and fused to obtain the collaborative coding feature of load i as:

[0079]

[0080] In the formula, represents the collaborative coding feature of load i.

[0081] The weighted collaborative modeling mechanism based on the load correlation matrix driving provided by the embodiment can adaptively adjust the feature sharing degree based on the dynamic correlation between loads according to the principle that the greater the correlation, the greater the collaborative prediction contribution, so as to ensure that the multi-task learning model can accurately perceive and use the correlation between loads for reliable collaborative prediction, thereby effectively improving the accuracy of each associated load prediction.

[0082] The load fusion feature is obtained by splicing and fusing each load coding feature and the corresponding collaborative coding feature, and the load fusion feature is input into the corresponding decoder for load prediction to obtain the predicted value of the corresponding load. The load fusion feature can be understood as a feature obtained by splicing the load coding feature of each load i in the associated load group and the corresponding collaborative coding feature. , which can be represented as: .

[0083] After obtaining the load fusion features of each load in the associated load group through the above method steps, the load fusion features can be sent to the corresponding decoder of the load prediction task for processing and analysis to obtain the corresponding prediction result. is represented as: , represents the decoder corresponding to the load i prediction task, which can be obtained by pre-training based on relevant load data.

[0084] Each load prediction input feature corresponding to the independent load group in the multi-element load prediction input feature is input into the corresponding single-task learning model for individual prediction to obtain the prediction value of each load in the independent load group. The single-task learning model is preferably an independent load task prediction model obtained by modeling an independent TCN structure, which effectively avoids interference with the prediction effect caused by the introduction of low correlation information to ensure the accuracy of the prediction results of each load in the independent load group. It should be noted that the model structure of the single-task learning model corresponding to each load in the independent load group is the same, and the parameters can be different. Both can use relevant load data to pre-train and construct based on the TCN network. The specific training process can be realized by referring to related prior art.

[0085] The prediction values of each load in the associated load group and the prediction values of each load in the independent load group are summarized to obtain the multi-element load prediction result.

[0086] The multi-task learning mechanism of the TCN enables the multi-task learning model to fully utilize the dynamic correlation information between loads while taking into account the individual differences of the loads, effectively improving the accuracy of load collaborative prediction. At the same time, the single-task learning mechanism of the TCN effectively avoids information interference between load prediction tasks, enabling the single-task learning model to focus on a single load prediction task and ensuring the reliability of each independent load prediction, thereby effectively improving the overall prediction performance of the comprehensive energy system multi-element load prediction.

[0087] The technical scheme of the embodiment of the present application comprises the following steps: periodically acquiring and preprocessing a multivariate load sequence and an influencing factor sequence to generate a to-be-analyzed multivariate load sequence and an influencing factor sequence; performing time alignment splicing on the to-be-analyzed multivariate load sequence and the influencing factor sequence to obtain a first multivariate load characteristic sequence; performing correlation analysis among loads based on a pre-constructed correlation evaluation model according to the first multivariate load characteristic sequence to generate a load correlation matrix and generate a correlated load group and an independent load group according to the load correlation matrix; performing singular spectrum decomposition and reconstruction on the to-be-analyzed multivariate load sequence to obtain a reconstructed multivariate load sequence; performing time alignment splicing on the reconstructed multivariate load sequence, the to-be-analyzed multivariate load sequence, and the influencing factor sequence to obtain a second multivariate load characteristic sequence; performing double-attention mechanism fusion analysis on each load characteristic sequence in the second multivariate load characteristic sequence to obtain a multivariate load prediction input feature; and performing multivariate load hybrid prediction based on the correlated load group and the independent load group according to the multivariate load prediction input feature to obtain a multivariate load prediction result. The multivariate load hybrid prediction mechanism based on the introduction of periodic dynamic load correlation evaluation can fully capture the dynamic coupling characteristic changes among multivariate loads and adaptively adjust the load collaborative prediction strategy, which can effectively avoid the negative impact of inaccurate load coupling relationship recognition on the overall prediction performance, improve the accuracy and robustness of multivariate load prediction of the comprehensive energy system, and has good popularization prospect and engineering application value.

[0088] In order to verify the application performance of the multivariate load hybrid prediction method of the comprehensive energy system proposed in the present application, based on the actual operation data (covering cold load, heat load, gas load, electric load, and related meteorological factors and time stamp information) of a certain park comprehensive energy system from January 2020 to December 2020, and after dividing the data set constructed after related preprocessing according to the proportion of 80% training set, 10% verification set, and 10% test set, the prediction method proposed in the present application and the prediction model based on the mult-task learning architecture (MTL) of LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit) are used for multivariate load prediction comparison experiment, and the analysis results shown in Tables 1-2 are obtained. Figure 2

[0089] Table 1 Root Mean Square Error Statistics Table

[0090]

[0091] Table 2 Average Absolute Percentage Error Statistics Table

[0092]

[0093] As Figure 2 ​As shown, the method of the present application predicts the overall trend of the cold load, gas load, heat load and electric load in the integrated energy system, which is obviously superior to the prediction effect of the multi-task learning model based on LSTM and GRU. At the same time, as shown in Tables 1 and 2, the present application has significant advantages in the two evaluation indexes of root mean square error (RMSE) and mean absolute percentage error (MAPE), with an average accuracy improvement of 68.26% and 83.60% respectively, and the present application shows stronger generalization ability and stability, which is suitable for complex scenarios of multi-type load collaborative prediction.

[0094] It should be noted that although each step in the above flowchart is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps has no strict order limitation, and these steps can be executed in other order.

[0095] In one embodiment, as shown in Figure 3 An integrated energy system multi-element load hybrid prediction system is provided, comprising:

[0096] A data acquisition module 1 is configured to periodically acquire and preprocess multi-element load sequences and influence factor sequences to generate multi-element load sequences and influence factor sequences to be analyzed;

[0097] A first splicing module 2 is configured to align and splice the multi-element load sequences to be analyzed and the influence factor sequences to obtain a first multi-element load feature sequence;

[0098] A correlation analysis module 3 is configured to perform correlation analysis between loads based on a pre-constructed correlation evaluation model according to the first multi-element load feature sequence, generate a load correlation matrix, and generate a correlated load group and an independent load group according to the load correlation matrix;

[0099] A second splicing module 4 is configured to perform singular spectrum decomposition and reconstruction on the multi-element load sequences to be analyzed to obtain reconstructed multi-element load sequences, and align and splice the reconstructed multi-element load sequences, the multi-element load sequences to be analyzed and the influence factor sequences to obtain a second multi-element load feature sequence;

[0100] A fusion analysis module 5 is configured to perform double-attention mechanism fusion analysis on each load feature sequence in the second multi-element load feature sequence to obtain multi-element load prediction input features;

[0101] A load prediction module 6 is configured to perform multi-element load hybrid prediction based on the correlated load group and the independent load group according to the multi-element load prediction input features to obtain multi-element load prediction results.

[0102] The specific definition of the comprehensive energy system multi-element load hybrid prediction system can refer to the definition of the comprehensive energy system multi-element load hybrid prediction method, and the corresponding technical effect can also be obtained equally, which will not be repeated here. Each module in the above comprehensive energy system multi-element load hybrid prediction system can be realized by software, hardware and combination thereof in whole or in part. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to call and execute the operations of the above modules by the processor.

[0103] Figure 4 An internal structure diagram of a computer device in an embodiment is shown, which can be a terminal or a server. As shown in the figure, Figure 4 The computer device includes a processor, a memory, a network interface, a display, a camera and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is used to communicate with external terminals through network connection. The computer program can implement the comprehensive energy system multi-element load hybrid prediction method when executed by the processor. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0104] Those skilled in the art can understand, Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme, and does not constitute a limitation on the computer device to which the scheme is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0105] In one embodiment, a computer device is provided, which includes a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program.

[0106] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program implements the steps of the above method when executed by the processor.

[0107] In conclusion, the multi-element load hybrid prediction method, system, device and medium provided by the embodiment of the application can capture the dynamic coupling characteristic changes among the multi-element loads based on the introduction of the periodic dynamic load correlation, and adaptively adjust the multi-element load hybrid prediction mechanism of the load collaborative prediction strategy, so that the negative influence of inaccurate load coupling relationship identification on the overall prediction performance can be effectively avoided, the accuracy and robustness of the multi-element load prediction of the comprehensive energy system are improved, and the method has good popularization prospect and engineering application value.

[0108] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts of each of the embodiments can be referred to each other, and each of the embodiments mainly describes the difference from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, it is described more simply, and the related parts can be referred to the part of the description of the method embodiment. It should be noted that each of the technical features of the above-mentioned embodiments can be combined arbitrarily, in order to make the description simple, each of the technical features of the above-mentioned embodiments is not described in all possible combinations, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the description.

[0109] The above-mentioned embodiments only express several preferred embodiments of the application, the description is more specific and detailed, but it should not be understood as the limitation of the patent scope of the application. It should be pointed out that for ordinary skilled in the art, without departing from the technical principles of the application, a number of improvements and replacements can be made, and these improvements and replacements should be regarded as the protection scope of the application. Therefore, the protection scope of the patent of the application should be subject to the protection scope of the claims.

Claims

1. A method for comprehensive energy system multi-element load hybrid prediction, characterized in that, The method comprises: Periodically acquiring and preprocessing a multivariate load sequence and an influencing factor sequence to generate a to-be-analyzed multivariate load sequence and an influencing factor sequence; Time-aligning and splicing the to-be-analyzed multivariate load sequence and the influencing factor sequence to obtain a first multivariate load characteristic sequence; According to the first multivariate load characteristic sequence, performing load correlation analysis based on a pre-constructed correlation evaluation model to generate a load correlation matrix, and generating a correlated load group and an independent load group according to the load correlation matrix; Performing singular spectrum decomposition and reconstruction on the to-be-analyzed multivariate load sequence to obtain a reconstructed multivariate load sequence, and time-aligning and splicing the reconstructed multivariate load sequence, the to-be-analyzed multivariate load sequence, and the influencing factor sequence to obtain a second multivariate load characteristic sequence; Respectively performing double-attention mechanism fusion analysis on each load characteristic sequence in the second multivariate load characteristic sequence to obtain a multivariate load prediction input feature; According to the multivariate load prediction input feature, performing multivariate load mixed prediction based on the correlated load group and the independent load group to obtain a multivariate load prediction result, comprising: According to all load prediction input features corresponding to the correlated load group in the multivariate load prediction input feature, performing collaborative prediction based on a pre-constructed multi-task learning model to obtain predicted values of loads in the correlated load group; Respectively inputting each load prediction input feature corresponding to the independent load group in the multivariate load prediction input feature into a corresponding single-task learning model to perform individual prediction and obtain predicted values of loads in the independent load group; Summarizing the predicted values of loads in the correlated load group and the predicted values of loads in the independent load group to obtain the multivariate load prediction result.

2. The integrated energy system multi-element load hybrid prediction method of claim 1, wherein, The correlation evaluation model comprises a feature extraction layer and a similarity analysis layer connected in sequence; the feature extraction layer comprises two feature extraction sub-networks sharing weights; the feature extraction sub-networks comprise convolution layers and recurrent encoding layers. 3.The integrated energy system multi-element load hybrid prediction method of claim 2, wherein, The similarity analysis layer is used for performing cosine similarity analysis on two load feature vectors output by the feature extraction layer to obtain corresponding load correlation scores. 4.The integrated energy system multi-element load hybrid prediction method of claim 1, wherein, The step of generating a correlated load group and an independent load group according to the load correlation matrix comprises: According to the load correlation matrix, obtaining loads having load correlation scores with other loads less than a preset correlation score threshold to generate the independent load group; the preset correlation score threshold is adaptively adjusted and optimized based on a grid search method; Generating the correlated load group from loads remaining after excluding loads in the independent load group from the multivariate loads. 5.The integrated energy system multi-element load hybrid prediction method of claim 1, wherein, The step of performing singular spectrum decomposition and reconstruction on the to-be-analyzed multivariate load sequence to obtain a reconstructed multivariate load sequence comprises: Respectively performing singular spectrum decomposition on each to-be-analyzed load sequence in the to-be-analyzed multivariate load sequence to obtain a plurality of frequency load subsequences; Respectively screening and reconstructing frequency load subsequences of each to-be-analyzed load sequence based on subsequence variance contribution degrees to obtain corresponding reconstructed load sequences; The reconstruction multi-element load sequence is obtained by aggregating all the reconstruction load sequences. 6.The integrated energy system multi-element load hybrid prediction method of claim 1, wherein, The step of performing collaborative prediction on all load prediction input features corresponding to the associated load group based on a pre-constructed multi-task learning model to obtain the predicted value of each load in the associated load group comprises: Each load prediction input feature corresponding to the associated load group is respectively encoded based on a preset shared encoder to obtain a corresponding load encoding feature; All related load encoding features corresponding to each load encoding feature are weighted and summed based on the corresponding load correlation score in the load correlation matrix to obtain a corresponding collaborative encoding feature; Each load encoding feature and the corresponding collaborative encoding feature are spliced and fused to obtain a corresponding load fusion feature, and the load fusion feature is input into a corresponding decoder for load prediction to obtain the predicted value of the corresponding load.

7. A comprehensive energy system multi-element load hybrid prediction system, characterized in that, The integrated energy system multi-element load hybrid prediction method of claim 1, wherein the system comprises: a data acquisition module for periodically acquiring and preprocessing multi-element load sequences and influence factor sequences to generate multi-element load sequences and influence factor sequences to be analyzed; a first splicing module for aligning and splicing the multi-element load sequences to be analyzed and the influence factor sequences to obtain a first multi-element load feature sequence; a correlation analysis module for performing inter-load correlation analysis based on a pre-constructed correlation evaluation model according to the first multi-element load feature sequence to generate a load correlation matrix, and generating an associated load group and an independent load group according to the load correlation matrix; a second splicing module for singular spectrum decomposition and reconstruction of the multi-element load sequences to be analyzed to obtain a reconstruction multi-element load sequence, and aligning and splicing the reconstruction multi-element load sequence, the multi-element load sequences to be analyzed, and the influence factor sequences to obtain a second multi-element load feature sequence; a fusion analysis module for performing double attention mechanism fusion analysis on each load feature sequence in the second multi-element load feature sequence to obtain multi-element load prediction input features; a load prediction module for performing multi-element load hybrid prediction based on the associated load group and the independent load group according to the multi-element load prediction input features to obtain multi-element load prediction results.

8. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images