Multi-energy coupling comprehensive energy system multi-element load forecasting method and system
By combining the WGCNA and BiLSTM-MTL frameworks, the accuracy and efficiency issues of multi-load forecasting in integrated energy systems at the park level were solved, enabling joint forecasting and accurate analysis of cooling, heating, and power loads, thereby improving energy utilization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANDONG JIANZHU UNIV
- Filing Date
- 2022-08-02
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies cannot effectively address the volatility and coupling of multiple loads in integrated energy systems at the park level, resulting in low load forecasting accuracy and efficiency. Traditional methods cannot deeply explore the nonlinear relationships and coupling characteristics between multiple loads.
We used weighted gene co-expression network analysis (WGCNA) to screen out suitable input features, and combined them with a multi-task learning (MTL) framework and a bidirectional long short-term memory neural network (BiLSTM) as a shared layer to perform joint prediction of cold, heat and electricity loads, thereby achieving information sharing and accurate prediction among multiple loads.
It improves the accuracy and efficiency of multivariate load forecasting, effectively uncovers the nonlinear correlation between loads, and enhances the energy utilization rate of the park.
Smart Images
Figure CN115310355B_ABST
Abstract
Description
Technical Field
[0001] This disclosure belongs to the field of load forecasting technology for integrated energy systems in industrial parks, and particularly relates to a multi-electrode load forecasting method and system for integrated energy systems that considers multi-energy coupling. Background Technology
[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.
[0003] In recent years, energy and environmental issues have attracted widespread attention from countries and society worldwide. Integrated Energy Systems (IES), as the physical carrier of next-generation energy, can achieve cascaded utilization of energy sources, effectively improving overall energy efficiency and reducing carbon emissions in industrial parks. However, regional integrated energy systems (RIES) at the park level have relatively small user scales, poor system robustness, and large load fluctuations. Accurate load forecasting is a key factor in the operation, scheduling, and energy management of combined cooling, heating, and power (CCHP) systems. Therefore, effectively addressing the volatility and coupling of diverse loads and accurately achieving multi-energy load forecasting has become a current research hotspot.
[0004] The inventors discovered that while current short-term load forecasting technology for power systems is relatively mature, the deep coupling and mutual influence of various energy sources within RIES (Restricted Energy Systems) and their vastly different dynamic characteristics mean that load forecasting methods for single-energy sources cannot be extended to multi-energy forecasting. Furthermore, single-energy load forecasting methods cannot accurately describe the strong coupling relationships between multiple energy sources, significantly compromising forecasting results. Additionally, excessive input factors reduce the computational efficiency of the model during load forecasting. Existing methods often use correlation coefficients to select input factors, but these methods are highly dependent on the amount of data and cannot deeply explore the nonlinear characteristics between variables. Therefore, the model may miss important features or even select unrelated features, generating excessive noise and affecting forecast accuracy. Summary of the Invention
[0005] To address the aforementioned problems, this disclosure provides a method and system for predicting multi-electrode loads in an integrated energy system that considers multi-energy coupling. The scheme employs WGCNA to analyze the correlation between features and select suitable input features; it uses the MTL framework, with BiLSTM as the shared layer of the MTL, allowing each subtask to share information and perform joint prediction of cooling, heating, and power loads to obtain prediction results. This scheme not only effectively uncovers the nonlinear correlations between multi-electrode loads but also improves prediction accuracy and efficiency through joint prediction of multi-electrode loads, thus helping to enhance energy utilization in industrial parks.
[0006] According to a first aspect of the embodiments of this disclosure, a multi-energy load forecasting method for an integrated energy system considering multi-energy coupling is provided, comprising:
[0007] Obtain historical load forecast data from the park's integrated energy system and perform corresponding preprocessing.
[0008] Based on the historical load prediction data, the weighted gene co-expression network analysis method is used to explore the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors, and to identify the influencing factors that are strongly correlated with different loads.
[0009] Historical load data for different loads and the features corresponding to their strongly correlated influencing factors are simultaneously input into a pre-trained load prediction model to obtain load prediction results for different loads. The load prediction model adopts the MTL framework and uses BiLSTM as the shared layer of the MTL. Prediction tasks under different loads share information through the shared layer.
[0010] Furthermore, the method of using weighted gene co-expression network analysis to mine the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors specifically involves: based on the obtained historical load prediction data, using weighted gene co-expression network analysis to obtain influencing factors strongly correlated with different loads, including but not limited to meteorological factors and calendar factors strongly correlated with loads.
[0011] Furthermore, the load prediction model includes a sequentially connected input layer, a shared layer, and an output layer, wherein the shared layer employs a BiLSTM neural network.
[0012] Furthermore, the corresponding preprocessing specifically includes outlier removal, missing value filling, and data normalization.
[0013] Furthermore, the historical load forecasting data includes historical load data corresponding to different load conditions, as well as corresponding meteorological and calendar factors.
[0014] Furthermore, the different loads include cooling loads, heating loads, and electrical loads.
[0015] Furthermore, for pre-trained load prediction models, the root mean square error, mean absolute error, or R² is used to evaluate their performance.
[0016] According to a second aspect of the present disclosure, a multi-energy load forecasting system for an integrated energy system considering multi-energy coupling is provided, comprising:
[0017] The data acquisition unit is used to acquire historical load forecast data related to the park's integrated energy system and perform corresponding preprocessing.
[0018] The influencing factor determination unit is used to mine the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors based on the historical load prediction data, and to determine the influencing factors that are strongly correlated with different loads.
[0019] The load forecasting unit is used to input the historical load data of different loads and the features corresponding to their strong influencing factors into a pre-trained load forecasting model to obtain load forecasting results for different loads. The load forecasting model adopts the MTL framework and uses BiLSTM as the shared layer of the MTL. The forecasting tasks under different loads share information through the shared layer.
[0020] According to a third aspect of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and running on the memory, wherein the processor executes the program to implement a multi-energy load forecasting method for an integrated energy system considering multi-energy coupling.
[0021] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements a multi-element load forecasting method for an integrated energy system considering multi-energy coupling.
[0022] Compared with the prior art, the beneficial effects of this disclosure are:
[0023] (1) This disclosure provides a multi-load forecasting method and system for integrated energy systems that considers multi-energy coupling. The scheme uses WGCNA to analyze the correlation between features and selects suitable input features; it adopts the MTL framework and uses BiLSTM as the shared layer of MTL, with each sub-task sharing information to perform joint forecasting of cooling, heating and electricity, and obtains the forecast results. The scheme can not only effectively explore the nonlinear correlation between multi-loads, but also perform joint forecasting of multi-loads, which improves the forecasting accuracy and efficiency and helps the park improve energy utilization.
[0024] (2) The scheme described in this disclosure takes into account that the load is a time series and has autocorrelation. The BiLSTM algorithm is used to effectively learn the inherent laws of forward and backward historical data and improve the prediction accuracy. At the same time, based on the constructed MTL framework, multi-load joint prediction can be carried out to achieve simultaneous prediction of cooling, heating and electricity loads, which improves the prediction efficiency.
[0025] Advantages of this disclosure in some respects will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of this disclosure. Attached Figure Description
[0026] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.
[0027] Figure 1 This is a basic WGCNA analysis flowchart as described in the embodiments of this disclosure;
[0028] Figure 2 This is a flowchart illustrating the execution process of the BiLSTM-MTL model as described in this embodiment.
[0029] Figure 3 This is a flowchart of a multi-energy system load forecasting method considering multi-energy coupling, as described in an embodiment of this disclosure. Detailed Implementation
[0030] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.
[0031] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.
[0032] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0033] Where there is no conflict, the embodiments and features described herein can be combined with each other.
[0034] Terminology Explanation:
[0035] Module (Model): A highly interconnected cluster of genes;
[0036] Adjacency matrix: A matrix composed of weighted correlation values between genes;
[0037] TOM (Topological overlap matrix): Converts the adjacency matrix into a topological overlap matrix to reduce noise and spurious correlations. This information can be used to build networks or draw TOM graphs.
[0038] LSTM: Long Short-Term Memory Neural Network, is a type of neural network used to process time series data, which has the ability to remember historical information.
[0039] Example 1:
[0040] The purpose of this embodiment is to provide a multi-electrode load forecasting method for integrated energy systems that considers multi-energy coupling.
[0041] A multi-energy load forecasting method for integrated energy systems considering multi-energy coupling includes:
[0042] Obtain historical load forecast data from the park's integrated energy system and perform corresponding preprocessing.
[0043] Based on the historical load prediction data, the weighted gene co-expression network analysis method is used to explore the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors, and to identify the influencing factors that are strongly correlated with different loads.
[0044] Historical load data for different loads and the features corresponding to their strongly correlated influencing factors are simultaneously input into a pre-trained load prediction model to obtain load prediction results for different loads. The load prediction model adopts the MTL framework and uses BiLSTM as the shared layer of the MTL. Prediction tasks under different loads share information through the shared layer.
[0045] Furthermore, the method of using weighted gene co-expression network analysis to mine the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors specifically involves: based on the obtained historical load prediction data, using weighted gene co-expression network analysis to obtain influencing factors strongly correlated with different loads, including but not limited to meteorological factors and calendar factors strongly correlated with loads.
[0046] Furthermore, the load prediction model includes a sequentially connected input layer, a shared layer, and an output layer, wherein the shared layer employs a BiLSTM neural network.
[0047] Furthermore, the corresponding preprocessing specifically includes outlier removal, missing value filling, and data normalization.
[0048] Furthermore, the historical load forecasting data includes historical load data corresponding to different load conditions, as well as corresponding meteorological and calendar factors.
[0049] Furthermore, the different loads include cooling loads, heating loads, and electrical loads.
[0050] Furthermore, for pre-trained load prediction models, the root mean square error, mean absolute error, or R² is used to evaluate their performance.
[0051] Specifically, for ease of understanding, the following detailed description of the solution in this embodiment is provided in conjunction with the accompanying drawings:
[0052] In order to solve the problems existing in the current technology:
[0053] (1) The traditional correlation coefficient method cannot be applied to the nonlinear relationship between multiple loads and various influencing factors in the integrated energy system of the park, and cannot screen the truly suitable input factors, resulting in low model prediction accuracy.
[0054] (2) Load forecasting based on a single energy source cannot accurately describe the deep coupling characteristics between multiple loads, resulting in low model forecasting efficiency.
[0055] This embodiment provides a multi-energy load forecasting method for integrated energy systems considering multi-energy coupling. The main technical concept of this method is as follows: WGCNA is used to analyze the correlation between features and select suitable input features; an MTL framework is employed, with BiLSTM as the shared layer of the MTL, allowing each subtask to share information for joint forecasting of cooling, heating, and power, while simultaneously obtaining the forecast results. This method not only improves forecast accuracy but also enhances forecast efficiency; therefore, it effectively solves the problems existing in the aforementioned background techniques. Specifically, as... Figure 3 As shown, the method described in this embodiment includes the following steps:
[0056] Step 1: Preprocess the raw data;
[0057] First, the load data of the park's integrated energy system is large, and anomalies are prone to occur during data measurement and storage. In order to improve the reliability of the data and reduce unnecessary noise generated by abnormal data during model training, it is necessary to identify and fill out outliers and missing values.
[0058] Secondly, to prevent excessive differences in the magnitude of variables from causing bias in model learning, data normalization is applied to the original dataset, scaling it to the [0,1] interval, thereby making the prediction method more effective. The normalization formula is as follows:
[0059]
[0060] In the formula: y′ represents the normalized data; y represents the original data; y max y minThese are the maximum and minimum values of the data.
[0061] Step 2: Considering the multi-energy coupling of the park's integrated energy system, WGCNA is used to screen features and determine the input / output feature set;
[0062] Due to the deep coupling of multiple loads in the park's integrated energy system, traditional correlation analysis methods cannot delve into the complex nonlinear relationships between loads and between loads and other factors. Therefore, Weighted Gene Co-expression Network Analysis (WGCNA) is used as a means to explore the correlation of multiple loads in the park's integrated energy system.
[0063] WGCNA is a systems biology method used to describe gene association patterns among different samples, and can be used to identify gene sets with highly co-variable expression. It is an algorithm that identifies genes with similar expression patterns based on differences in gene expression patterns, defining them as modules. The WGCNA analysis flowchart is shown below. Figure 1 First, the soft threshold for the weighted correlation gene network is selected. A specific value β (soft threshold) is taken as the power of the correlation coefficient between any two pairs of genes i and j, forming an adjacency matrix. The calculation formula is as follows:
[0064] a ij =|cor(i,j)| β (1)
[0065] In the formula: a ij The correlation between genes a ij i and j represent two different genes; β is the weight, with a default value of 1-30.
[0066] After selecting an appropriate β value, a weighted co-expression network and module identification are constructed. The similarity between two genes is calculated using topological overlap (TOM), and the calculation formula is as follows:
[0067]
[0068] In the formula: u is the iterator, which iterates through all genes in the gene list except i and j, and performs the above calculation. TOM ij =0 means that the networks of genes i and j do not share any neighboring genes; TOM ij =1 means that genes i and j have the exact same network neighbor genes.
[0069] Finally, the gene modules are associated with external information. WGCNA provides a function to visualize their correlations, and the final result is a correlation coefficient matrix between each module and external features, from which modules that are highly correlated with external features can be identified.
[0070] WGCNA has been successfully applied in the fields of biology, genetics, etc., and is less applied in the engineering field. In this invention, WGCNA is applied to multi-load prediction to screen features that have a strong impact on load changes. The sample points of each feature in the dataset are used to form a gene set, and the process of weighted gene co-expression network analysis is implemented using the WGCNA package in R language.
[0071] After obtaining the visualization results, the correlation coefficient of each module is W ij Select the constant q as the division criterion. When W ij <q, it is considered weakly correlated and is not used as an input factor; if W ij ≥q, it is considered strongly correlated and is used as an input factor. Through this step, the input feature set for each task is determined as k n ={k1,k2,…,k l}, where l is the number of input factors and n is the number of subtasks. The input feature set usually includes historical data of cooling, heating, and power loads, meteorological factors strongly correlated with the load, and calendar factors, and the output feature set is the cooling, heating, and power loads to be predicted.
[0072] Step 3: Multi-load prediction model based on BiLSTM-MTL
[0073] The BiLSTM-MTL prediction model proposed in this embodiment is divided into an input layer, a shared layer, and an output layer. The model flow chart is as Figure 2 shown.
[0074] Input layer: After performing Step 1 and Step 2, determine the input feature set for each subtask as the input layer of the model input={k1,…,k n}, where n = 3;
[0075] Shared layer: Send it to the shared layer of the model. In the shared layer, a BiLSTM neural network layer is jointly composed of multiple LSTM neurons, and then linearly connected to multiple BiLSTM network layers with the same structure to form a multi-layer shared network in MTL;
[0076] Output layer: Output the label values of each subtask. The output layer may include a dropout layer and a dense layer, etc., and the output is the load values of cooling, heating, and power output={o1,…,o n}, where n = 3.
[0077] The principles of multi-task learning (MTL) and bidirectional long short-term memory network (BiLSTM) in the model are as follows.
[0078] (1) MTL
[0079] Compared to single-task learning, multi-task learning (MTL) enables multiple different tasks to be completed in parallel, and improves the generalization ability of each subtask by leveraging the correlation between tasks. Employing the hard parameter sharing mechanism in MTL can reduce the amount of data and the overall parameter size of the model, thereby making the model more efficient.
[0080] (2) BiLSTM
[0081] When using a unidirectional neural network for comprehensive energy load forecasting in a park, training is performed from front to back according to the time series. However, load series exhibit autocorrelation, and unidirectional neural networks have low utilization rates for long-term data, failing to effectively mine the inherent characteristics of the data. BiLSTM, composed of forward and backward LSTMs, can better capture the dependencies between time series. Therefore, to further mine the correlation information between past and future times of the load and further improve the model's prediction accuracy, BiLSTM is used as a shared layer in multi-task learning to share information parameters.
[0082] The network parameters of LSTM are calculated as follows:
[0083] f t =σ(W f [h t-1 ,x t ]+b f (4)
[0084] i t =σ(W i [h t-1 ,x t ]+b i (5)
[0085] o t =σ(W o [h t-1 ,x t ]+b o (6)
[0086] c t =f t c t-1 +i t tanh(W c [h t-1 ,x t ]+b c (7)
[0087] h t =σ(o t tanhc t (8)
[0088] In the formula: f t i t o t c t These represent the states of the forget gate, input gate, output gate, and state unit at time t; h t-1 The state of the hidden layer at the previous moment; x t This is the input at time t; W f W i W o W c and b f b i b o b c These are the corresponding weight coefficient matrix and bias term, respectively; σ represents the Sigmoid activation function.
[0089] The hidden layer state h of BiLSTM t The input quantity a at the current moment t The hidden layer output state h at the previous moment before forward propagation t-1 And the output state h of the previous moment before backpropagation i-1 Composition. The hidden layer state is shown in the following formula.
[0090] h t =LSTM(x t ,h t-1 (9)
[0091] h i =LSTM(x t ,h i-1 (10)
[0092] h t =a t h t +b t h i +c t (11)
[0093] In the formula: LSTM is the operation process of formulas (4)-(8); h t This represents the state of the forward hidden layer; h i This is the state of the backward hidden layer; a t Output weights for the hidden layer of the forward propagation unit; b t Output weights for the hidden layer of the backpropagation unit; c t Optimize the hidden layer bias parameters for the current moment.
[0094] Step 4: Model Evaluation
[0095] The model's output of the combined cooling, heating, and power prediction results was compared with the actual values, using root mean square error (RMSE), mean absolute error (MAE), and R². 2 The formula for evaluating the model is as follows:
[0096]
[0097]
[0098]
[0099] In the formula: n is the number of samples; y i Let i represent the actual value at time i; Let i represent the actual value at time i; This represents the average value of the sample.
[0100] When using RMSE and MAE as error assessment metrics, smaller values are better; when using R... 2 When used as a performance evaluation metric for a model, a higher value indicates a better fit.
[0101] Example 2:
[0102] The purpose of this embodiment is to provide a multi-electrode load forecasting system for integrated energy systems that considers multi-energy coupling.
[0103] A multi-energy load forecasting system for integrated energy systems considering multi-energy coupling includes:
[0104] The data acquisition unit is used to acquire historical load forecast data related to the park's integrated energy system and perform corresponding preprocessing.
[0105] The influencing factor determination unit is used to mine the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors based on the historical load prediction data, and to determine the influencing factors that are strongly correlated with different loads.
[0106] The load forecasting unit is used to input the historical load data of different loads and the features corresponding to their strong influencing factors into a pre-trained load forecasting model to obtain load forecasting results for different loads. The load forecasting model adopts the MTL framework and uses BiLSTM as the shared layer of the MTL. The forecasting tasks under different loads share information through the shared layer.
[0107] Furthermore, the system described in this embodiment corresponds to the method described in Embodiment 1, and its technical details have been described in detail in Embodiment 1, so they will not be repeated here.
[0108] In further embodiments, the following is also provided:
[0109] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.
[0110] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.
[0111] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.
[0112] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.
[0113] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.
[0114] Those skilled in the art will recognize that the units, i.e., algorithm steps, of the various examples described in connection with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.
[0115] The multi-electrode load forecasting method and system for integrated energy systems that considers multi-energy coupling provided in the above embodiments can be implemented and has broad application prospects.
[0116] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.
Claims
1. A multi-energy load forecasting method for integrated energy systems considering multi-energy coupling, characterized in that, include: Obtain historical load forecast data from the park's integrated energy system and perform corresponding preprocessing. The historical load forecasting data includes historical load data corresponding to different loads, as well as corresponding meteorological and calendar factors. Based on the historical load prediction data, the weighted gene co-expression network analysis method is used to explore the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors, and to identify the influencing factors that are strongly correlated with different loads. The method of using weighted gene co-expression network analysis to explore the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors is as follows: Based on the obtained historical load prediction data, the influencing factors strongly correlated with different loads are obtained from the weighted gene co-expression network analysis method. The influencing factors include, but are not limited to, meteorological factors and calendar factors strongly correlated with the load. The weighted gene co-expression network analysis method provides a function for visualizing its correlations, and the final result is the correlation coefficient matrix between each module and the external features, from which modules that are highly correlated with the external features can be identified; Historical load data for different loads and the features corresponding to their strongly correlated influencing factors are simultaneously input into a pre-trained load prediction model to obtain load prediction results for different loads. The load prediction model adopts the MTL framework and uses BiLSTM as the shared layer of the MTL. Prediction tasks under different loads share information through the shared layer. The different loads include cooling loads, heating loads, and electrical loads; The load prediction model includes an input layer, a shared layer, and an output layer connected in sequence. The shared layer adopts a BiLSTM neural network. Multiple BiLSTM network layers with the same structure are linearly connected to form a multi-layer shared network in MTL.
2. The multi-energy load forecasting method for integrated energy systems considering multi-energy coupling as described in claim 1, characterized in that, The corresponding preprocessing specifically includes outlier removal, missing value filling, and data normalization.
3. The multi-energy load forecasting method for integrated energy systems considering multi-energy coupling as described in claim 1, characterized in that, For pre-trained load prediction models, the root mean square error, mean absolute error, or R² is used to evaluate their performance.
4. A multi-energy load forecasting system for an integrated energy system considering multi-energy coupling, employing the multi-energy load forecasting method for an integrated energy system considering multi-energy coupling as described in any one of claims 1-3, characterized in that, include: The data acquisition unit is used to acquire historical load forecast data related to the park's integrated energy system and perform corresponding preprocessing. The influencing factor determination unit is used to mine the nonlinear relationships between multiple loads and between loads and their corresponding influencing factors based on the historical load prediction data, and to determine the influencing factors that are strongly correlated with different loads. The load forecasting unit is used to input the historical load data of different loads and the features corresponding to their strong influencing factors into a pre-trained load forecasting model to obtain load forecasting results for different loads. The load forecasting model adopts the MTL framework and uses BiLSTM as the shared layer of the MTL. The forecasting tasks under different loads share information through the shared layer.
5. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and running thereon, characterized in that, When the processor executes the program, it implements a multi-element load forecasting method for an integrated energy system considering multi-energy coupling as described in any one of claims 1-3.
6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements a multi-element load forecasting method for an integrated energy system considering multi-energy coupling, as described in any one of claims 1-3.