Multi-energy load prediction method and system under zero historical data of integrated energy system

Abstract

The invention belongs to the technical field of load prediction, and discloses a multi-energy load prediction method and system under zero historical data of an integrated energy system, and the method comprises the steps: obtaining the meteorological characteristics of a target park and the cold, heat, electricity and gas historical data of a source domain group park, and carrying out the preprocessing of the obtained data; performing park cross-correlation and generalization ability analysis on the preprocessed cold, heat, electricity and gas historical data of the source domain group park, and determining proper source domain data; constructing a multi-energy load prediction model, and training the model based on the source domain data by using a Metas training strategy to obtain a trained prediction model; and inputting the preprocessed meteorological characteristics of the target park into the prediction model to obtain a prediction result. According to the invention, long-term accurate prediction of the multi-energy load in the target integrated energy system under the condition of zero historical load data is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of load forecasting, and in particular to a multi-energy load forecasting method and system under zero historical data of an integrated energy system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the rapid development of the Integrated Energy System (IES), accurate prediction of multi-energy loads has become a core requirement for the optimal operation of the system. However, in practical applications, obtaining comprehensive multi-energy load historical data often requires a lot of time and financial costs, and due to data privacy and other reasons, some information cannot be made public. The size of the available data set is limited, and traditional deep learning methods face serious data dependency bottlenecks, which brings huge challenges to IES load forecasting. In response to this problem, many scholars at home and abroad have adopted the method of transfer learning to deal with this problem.

[0004] One of the most important key steps before transfer learning is to assist in selecting a suitable source domain by performing correlation analysis on the load sequence. Some scholars consider the complex linear and nonlinear characteristics of load data. Improvements have been made to its correlation analysis, further reducing the amount of target domain data required. However, in the extreme case where the park is not built, the target system cannot obtain any historical load data. Traditional source domain selection relies on target domain correlation analysis, and zero data conditions cause feature comparison to fail completely. Some scholars have proposed transfer learning solutions based on generative adversarial networks (GAN), fine-tuning strategies, and meta-learning frameworks, but they still require a small amount of target domain data, and have obvious defects in complex time series feature extraction and multi-source domain collaborative optimization.

[0005] In summary, domestic and foreign scholars have proposed a variety of solutions for high-precision multi-energy load forecasting under the condition of extremely scarce load data. However, under the condition of zero historical load data in newly built parks, the existing transfer learning technology still faces two bottlenecks: (1) Difficulty in selecting the source domain: The similarity analysis between the source domain and the target domain in transfer learning strongly depends on the target domain data. Traditional correlation analysis methods fail under zero-data conditions.

[0006] (2) Insufficient cross-domain generalization capability: Cross-domain parameter migration does not fully consider the dynamic differences of multiple source domains, resulting in low parameter migration efficiency and large prediction errors. Summary of the invention

[0007] In order to solve the above problems, the present invention proposes a multi-energy load prediction method and system under zero historical data of an integrated energy system. Based on the Tnet algorithm and an improved meta-learning strategy, a multi-energy load prediction model based on transfer learning is constructed, which realizes the long-term and accurate prediction of the multi-energy load in the target integrated energy system under the condition of zero historical load data.

[0008] In order to achieve the above object, the present invention adopts the following technical solution: In a first aspect, the present invention provides a multi-energy load forecasting method for an integrated energy system with zero historical data, comprising the following steps: Obtain the meteorological characteristics of the target park and the historical data of heating, cooling and electricity of the source domain group parks, and pre-process the acquired data; For the pre-processed historical data of heating, cooling and electricity in the source domain group, we conduct cross-correlation and generalization analysis to determine the appropriate source domain data. Construct a multi-energy load prediction model, use the Metas training strategy, train the model based on source domain data, adjust the gradient weight according to the verification loss of the inner loop task and the source domain generalization probability, and obtain a trained prediction model; The preprocessed meteorological characteristics of the target park are input into the prediction model to obtain the prediction results.

[0009] As an optional implementation, the Metas training strategy is specifically: Considering the parameter updates of different source domains, the loss value and weight value of each training task in the inner layer are fully applied to the update function of the outer layer.

[0010] As an optional implementation, the multi-energy load prediction model consists of an encoder module and a decoder module.

[0011] As an optional implementation, the encoder uses the CNN layer to perform feature extraction and dimensionality reduction through convolution operations, and passes the compressed data to the LSTM to further capture the long-term dependencies in the time series.

[0012] As an optional implementation, a multi-head attention mechanism is used to dynamically weight the relationships at different time steps, giving different hidden state probability weights to the LSTM, focusing on the impact of important information related to the cold and hot electrical loads.

[0013] As an optional implementation method, the park inter-correlation and generalization ability analysis is as follows: The Time2vec algorithm is used to transform multiple load time series data of each source domain into periodic and linear parts, which are processed by the MIC algorithm and the MD algorithm respectively. K-means clustering analysis is constructed to stratify the source domains for generalization potential, and low-potential clusters are eliminated to reduce noise interference. MD and MIC are weightedly fused, and grid search is used to find the optimal parameter range, thus obtaining the cross-correlation score of the source domain group and a reasonable source domain range.

[0014] In a second aspect, the present invention provides a multi-energy load forecasting system under zero historical data of an integrated energy system, comprising: The data acquisition and preprocessing module is configured to: acquire the meteorological characteristics of the target park and the historical data of heating, cooling and electricity of the source domain group parks, and preprocess the acquired data; The source domain data determination module is configured to: perform park cross-correlation and generalization capability analysis on the pre-processed cooling, heating and electrical historical data of the source domain group parks to determine appropriate source domain data; The model building and training module is configured to: build a multi-energy load prediction model, use the Metas training strategy to train the model based on the source domain data, adjust the gradient weight according to the verification loss of the inner loop task and the source domain generalization probability, and obtain a trained prediction model; The model output module is configured to: input the pre-processed meteorological characteristics of the target park into the prediction model to obtain the prediction result.

[0015] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.

[0017] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in the first aspect.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The multi-energy load forecasting method and system for the integrated energy system with zero historical data of the present invention constructs a transfer learning-driven multi-head attention optimization encoding-decoding model based on the Tnet algorithm and the improved meta-learning strategy, achieves prediction breakthroughs through three-level cascade optimization, realizes long-term prediction of cold, heat, electricity and gas loads in the target domain, and significantly improves the load prediction accuracy in zero-data scenarios.

[0019] The multi-energy load forecasting method and system under zero historical data of the integrated energy system of the present invention proposes a source domain selection algorithm (Tnet) for the scenario of zero historical load data in the target domain, which transforms the traditional source domain similarity analysis into a probabilistic evaluation problem of the model generalization ability. Without any target domain historical data, the most suitable source domain set is screened out through Tnet to realize load forecasting, breaking through the difficulty of initialization of transfer learning under zero data constraints.

[0020] The multi-energy load forecasting method and system for the integrated energy system with zero historical data of the present invention designs an improved meta-learning training algorithm (Metas), introduces the loss value and weight information of each inner loop task when updating the outer loop of meta-learning, makes full use of the multi-source domain knowledge provided by Tnet, breaks through the parameter migration barrier in time series prediction, and improves the generalization performance of the prediction model.

[0021] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0023] Figure 1 A flowchart of a multi-energy load forecasting method under zero historical data for an integrated energy system provided in Example 1 of the present invention; Figure 2 A flowchart of the Tnet algorithm provided in Example 1 of the present invention; Figure 3 A schematic diagram of the Metas algorithm provided in Example 1 of the present invention; Figure 4 This is a framework diagram of a multi-energy load prediction model provided in Example 1 of the present invention, wherein: Figure 4 (a) is the overall framework diagram of the model. Figure 4 (b) is the encoder-decoder framework diagram of the model. DETAILED DESCRIPTION

[0024] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0026] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0028] Example 1 like Figure 1 As shown, this embodiment provides a multi-energy load forecasting method for an integrated energy system with zero historical data, comprising the following steps: Obtain the meteorological characteristics of the target park and the historical data of heating, cooling and electricity of the source domain group parks, and pre-process the acquired data; For the pre-processed historical data of heating, cooling and electricity in the source domain group, we conduct cross-correlation and generalization analysis to determine the appropriate source domain data. Construct a multi-energy load prediction model, use the Metas training strategy, train the model based on source domain data, adjust the gradient weight according to the verification loss of the inner loop task and the source domain generalization probability, and obtain a trained prediction model; The preprocessed meteorological characteristics of the target park are input into the prediction model to obtain the prediction results.

[0029] In order to solve the extreme problem of no historical load data in the target domain, this application proposes a new multi-energy load forecasting method. This method includes a new source domain selection algorithm, an improved meta-learning training strategy, and an encoding-decoding prediction model that integrates a multi-head attention mechanism. It achieves a prediction breakthrough through three-level cascade optimization, thereby achieving long-term and accurate prediction of various types of loads such as cold, hot, electric, and gas in the target integrated energy system. It is suitable for the planning and scheduling of integrated energy systems in newly built parks or those lacking historical data.

[0030] like Figure 1As shown in the figure. First, the original data of the source domain group and the target park are preprocessed, including outlier detection, data normalization, feature contribution judgment, etc. Secondly, the Tnet algorithm is used to analyze the park correlation and generalization ability of the source domain group, convert the similarity analysis problem into a classification problem, and find the park suitable as the source domain. Then, combined with the data preprocessing of the target park, autocorrelation analysis is constructed to determine the type and period of the input data. Subsequently, the data of multiple parks are input into the prediction model, and the feature extraction and shared encoding and decoding modules composed of convolutional neural network (CNN), long short-term memory network (LSTM), and multi-head attention mechanism (MHA) are used to decompose and extract the coupling information to obtain the joint prediction result. Metas is used to apply the loss value and weight value of each task in the inner layer to the function of the outer layer to update the model parameters. Finally, the weather characteristics of the target park are brought into the updated model to realize the multi-energy prediction of zero historical data of cold, hot and electrical load data.

[0031] Among them, the Tnet algorithm is used to analyze the park inter-correlation and generalization ability of the source domain group to find the park suitable as the source domain. Specifically: The Tnet algorithm reconstructs the source domain selection paradigm through a five-stage probabilistic reasoning framework, using five main steps to find suitable source domain data: time vector embedding (Time to Vector, Time2vec), Mahalanobis distance (MahalanobisDistance, MD), maximum information coefficient (Maximal Information Coefficient, MIC), K-means clustering (K-means), and Bayesian Weighted Probability Averaging Method (Bayesian Weighted Probability Averaging Method, BWPA). Figure 1 middle They respectively refer to the best source domain groups selected by the Tnet algorithm. The best source domain group is a collection of parks, and each park has its own historical load data and weather data. The data of these best source domain groups are brought into the prediction model to train the model parameters, and the model parameters are updated through the loss between the data predicted by the prediction model and the real data. The loss cannot be used directly, and the Metas algorithm needs to be used to convert the loss into parameters. It represents the weighted loss, which is only part of the Metas training strategy. The other part is Figure 3After the model parameters are updated, the model is transferred to the target park to be predicted through transfer learning, and the load data of the park is predicted through the weather data of the target park. Figure 1 In It refers to the data of each prediction for 24 hours, that is, the first round of prediction is 24 hours, and the second round is also 24 hours, which adds up to 48 hours, which is a stacked graph (indicating that the predicted load is constantly increasing).

[0032] Figure 2 A flow chart is given. Since the historical data of the target park's cold, hot, electric and gas loads are unknown, it is crucial to find a similar park group in the source domain group as the source domain. In view of the periodicity and linear characteristics of the load data used, the single consideration of linear and nonlinear factors and the direct use of linear analysis and nonlinear analysis will inevitably be affected by opposing factors. Therefore, the Time2vec algorithm is used to transform the multiple load time series data of each source domain into periodic and linear parts, and map the time characteristics to a high-dimensional space. The periodic part mainly considers the seasonal changes in load forecasting, while the linear characteristics capture the baseline trend of the load data. As shown in formula (1). Among them, represents the original time series, represents the time step, Represents the embedding dimension, 0 dimension represents linear features, and 1 to k dimensions represent nonlinear features.

[0033] (1) In order to conduct linear and nonlinear joint analysis in high-dimensional space, the MD algorithm is used to introduce the covariance matrix to adjust the weights of different features so that features of different scales and different correlations can be treated fairly during calculation. The MIC algorithm is used to divide different grids for different nonlinear variables, and the maximum amount of information is found through different discretization strategies. As shown in formulas (2) and (3). Then, MD and MIC are used to construct K-means clustering analysis to stratify the generalization potential of the source domain, and low-potential clusters are eliminated to reduce noise interference. If a cluster contains multiple parks, if they are used uniformly, the exact generalization ability of each park for the target park cannot be considered, and the prediction error will be large. Therefore, formula (4) is used to weightedly fuse MD and MIC and use grid search to find the optimal parameter range to obtain the cross-correlation score of the source domain group in this cluster and a reasonable source domain range.

[0034] (2) (3) (4) (5) Finally, in order to make full use of the contribution of different source domains to the target domain in the prediction model, the weighted mutual information entropy is introduced to construct a Bayesian weighted model. The reasonable source domain is used as the label and the formula (4) is used to calculate the weighted mutual information entropy. Construct a parameter model and bring it into formula (5), taking into account the "uncertainty" of each data point to assign a weight to it. This converts the traditional probability prediction problem into a classification problem to find the most suitable park cluster. and These are the heating, cooling and electrical data of two different parks, all in 4D. Expressed as The inverse matrix of represents the correlation between the features of the data points, which is a The matrix of . The data is divided into The grid, Is a grid The mutual information estimate under the condition is the maximum value among all Select from the grid (usually , is the sample size), It is a model The prior probability of For the predicted results, are input features.

[0035] Therefore, in this application, this algorithm can be regarded as the maximum generalization ability problem after standardizing the four different scale data of cooling, heating and electricity. At the same time, the source domain group is analyzed using the Tnet algorithm, and the probability that the source domain can represent the target domain is obtained through BWPA (Bayesian weighted probability), which also provides a valuable reference for the optimization method in the following prediction model (the inner layer of the Metas algorithm below needs to use this probability value for weighting).

[0036] Improved meta-learning training strategy, specifically: Meta-learning optimizes the model’s initialization parameters by performing gradient descent updates on each task and minimizing the test loss of all tasks using a meta-optimizer, allowing it to quickly adapt to new tasks and improve the generalization ability of the prediction model. Although Tnet effectively solves the key problem of source domain selection in meta-learning and meets the stringent requirements of meta-learning for high-quality prior tasks, it still has limitations in the parameter update mechanism. Specifically, the meta-layer update optimizes the initialization parameter states between tasks through inner-layer gradient descent. This design may cause dual contradictions in time series prediction tasks: on the one hand, the inner-layer update does not adequately capture the dynamic correlation of time series features, making it difficult for model parameters to fully fit complex patterns in the time dimension; on the other hand, the initialization parameters shared between tasks lack adaptive adjustment to source domain differences during cross-domain migration, ultimately leading to a significant attenuation in the model’s prediction accuracy. Inspired by the idea of ​​meta-learning, an improved meta-learning parameter optimization method is proposed. Figure 3 As shown in Figure 2, three source domains are selected as representatives. represents the initial parameters in the prediction model, It means that the prediction model gets the final parameters after m+1 rounds of updates, where x 1. x 2 to x 3 represents the probability of a certain park being the source domain. 1 represents the loss corresponding to Park 1. 1 is to perform gradient update, 1 represents the internal update step size. Similarly, 1 and 1 are the parameters corresponding to Park 2 and Park 3 respectively. The Metas algorithm uses two layers of updates (including inner and outer layers) to make the two layers of updates act on the hyperparameters of the model. The inner layer is the loss value of each task and the source domain weight information calculated by Tnet, which is dynamically integrated into the outer layer parameter update to achieve differentiated gradient adjustment. The loss value and weight value of each task in the inner layer are fully applied to the update function of the outer layer, and the "inner layer loss-outer layer update" is realized through the dynamic gradient weighting mechanism. In this way, the parameter updates of different source domains can be fully considered to improve the prediction accuracy. As shown in formula (6). Where is the model vector parameter, For the task The training loss parameter is For the task The validation loss parameter is The task weight calculated by the Tnet algorithm, is the internal gradient update step size, is the outer learning rate, is the batch size, For input The model predicts the output, Task Hessian matrix of the training loss.

[0037] (6) Based on the above analysis, this application proposes a multi-energy load forecasting model based on transfer learning (Metas-Tnet-CNN-MHA-LSTM), such as Figure 4 As shown, the model is an m2m multi-step load forecasting model. Figure 4 middle, i （1） to i （n） These represent the input, o （1） to （n） It represents the predicted output. y It represents the true value and is used to calculate the loss with the predicted value. The final output of LSTM (Long Short-Term Memory Network) , t represents the time step, It represents the output vector of the long-term and short-term time network at a certain moment, and passes it to the multi-head attention mechanism (MHA). The principle of MHA is to analyze the local features of the time series and focus on the features to find the truly weighted content. Divided into two parts ( ) is because: Perform MHA analysis, the above use To weight, the following Used for calculation Here is how much. It is the head of MHA, that is, the head of the multi-head attention mechanism.

[0038] in, Figure 4 (a) is the overall framework of the model, and the number of source domain parks is set to , the number of features is , the number of output features is The prediction model is an m2m structure, such as Figure 4As shown in (b), it consists of an encoder module and a decoder module. The input is different weather characteristics, and the output is the cold, hot and electrical power forecast value. In the encoder, the CNN layer is used to extract features and reduce dimensions through convolution operations, and the compressed data is passed to LSTM to further capture the long-term dependencies in the time series. The multi-head attention mechanism is used to dynamically weight the relationships at different time steps, and different hidden state probability weights are given to LSTM, focusing on the impact of important information related to cold, hot and electrical loads. The global modeling ability of the model is further optimized. Finally, the hidden layer relationship obtained by the encoder module is input into the decoder module with LSTM as the core to output the final load forecast value.

[0039] So far, a multi-energy load forecasting model with zero historical load data based on transfer learning has been established. To verify the effectiveness of the proposed method, three case experiments are carried out to illustrate it.

[0040] Case 1: Business Park Forecasting.

[0041] The source domain data selected the electricity, heat, gas, and cooling load data of 9 parks (sampling interval is 1 hour) and 6 meteorological characteristics (temperature, humidity, wind speed, etc.) for normalization and outlier processing. The target domain data only inputs 6 types of meteorological data, and no historical load records are required. The Tnet algorithm is executed to select the appropriate source domains 1, 7, and 8. Then other source domains are used for comparative experiments, and the results are shown in Table 1: Table 1 Comparison of prediction accuracy of different source domains in business parks;

[0042] From the experimental data in Table 1, it can be seen that the Tnet source domain selection algorithm shows significant advantages in selecting source domains. By comparing the Metas-CNN-MHA-LSTM model with different source domains, the performance differences of each source domain under different target domains can be clearly observed. In particular, for the target park 0, the Tnet algorithm successfully selected source domain groups 1, 7, and 8 as the best source domain groups. This selection significantly improved the model's fit and prediction accuracy. Further analysis shows that the prediction effect in source domains 5 and 9 is the worst, which is completely consistent with the result that the probability of 5 and 9 selected by Tnet is low. Secondly, the prediction effect of source domains 1, 4, 7, and 8 is slightly worse, because the probability of including source domain 4 is <80%, resulting in a decrease in accuracy. The two source domains with the highest probability are not selected because we hope to make full use of the three source domains with the highest probability for prediction, and their probabilities are similar. The final prediction results are also very similar, which proves the reliability of the probability value output by Tnet. Through these experimental analyses, the advantage of the Tnet algorithm is that it can accurately select the best source domain group from different combinations of source domains and target domains, helping the model to achieve better performance. Especially in multi-source domain and multi-target domain tasks, the Tnet algorithm obviously has strong adaptability and effect optimization capabilities.

[0043] Case 2: Office Park Forecasting.

[0044] In order to further verify the generalization ability of the proposed prediction model in different scenarios, this experiment conducts cross-regional and cross-application scenario tests, using the same model architecture: Metas-Tnet-CNN-MHA-LSTM, focusing on analyzing the model's adaptive ability in different scenarios to verify its cross-scenario robustness. The office park is selected as the new target domain. Its cooling load is significantly affected by the periodicity of the working day, which is in sharp contrast to the commercial park in Section 3.4. The Tnet algorithm selects a high-probability source domain group (1, 5, 9) for the office park. The results in Table 2 show that when the preferred source domain group (5, 9) is used, the MAPE of electricity, heat, and cooling loads are reduced to 10.166%, 10.341%, and 9.342%, respectively, and the R² values ​​are all over 85%, which is significantly better than other combinations. It is worth noting that when low-probability source domains (such as 2, 4) are included, the R² value of cooling load prediction is even negative (-45.2%), indicating that heterogeneous feature interference causes model failure. Tnet excludes such interference source domains (probability <60%) through Bayesian weighting to ensure that the model focuses on subgroups with high generalization capabilities. In comparison, the cold load MAPE of traditional single-source domain migration (such as using only Park 5) is 12.557%, and the error is 34.4% higher than the Tnet preferred combination, highlighting the necessity of multi-source domain collaboration. In addition, source domain 3 is highly matched with the target domain in terms of time periodic components and local nonlinear trends, and is given a higher weight by the Tnet algorithm (probability 70%~80%). Its heat load prediction contributes key information (MAPE=15.585%, R²=67.7%), indicating that Tnet can adaptively adjust the source domain priority based on multi-dimensional features.

[0045] The Metas-Tnet-CNN-MHA-LSTM model dynamically captures cross-domain spatiotemporal dependencies through a multi-head attention mechanism, alleviating the distribution shift problem. For example, in gas load forecasting, the linear correlation between the target domain and source domains 5 and 9 is lower than that of other load forecasts, but the model further extracts cross-domain nonlinear dependencies (such as load fluctuation patterns) through attention weights, making the MAPE (16.421%) significantly lower than the full source domain combination (19.372%). In addition, the Metas optimization strategy considers the adaptability of different source domain features to the target domain during the gradient update process, allowing the model to learn the optimal parameter configuration faster and further improve the prediction accuracy. In similar scenarios, the Tnet algorithm effectively identifies source domain groups with heterogeneous distribution but complementary functions through spatiotemporal feature decoupling and probabilistic optimization; Metas-Tnet-CNN-MHA-LSTM realizes the directional reinforcement of cross-domain nonlinear features through the attention mechanism. The two work together to keep the model high in similar generalization tasks (MAPE<10.15%, R²>83.7%). At the same time, through the experimental verification in this section, the proposed model is not only applicable to specific target domains, but can also be extended to different application scenarios, providing a theoretical basis for the standardized deployment of park-level integrated energy systems.

[0046] Table 2 Predicted values ​​for office parks;

[0047] Case 3: Fine-tuning and optimizing predictions with small samples.

[0048] On the basis of the above zero data migration, in order to further explore the fine-tuning effect of a small amount of target domain data on the prediction model, this section introduces a small sample transfer learning strategy and analyzes the impact of different fine-tuning data ratios on model accuracy. The experiment uses 3.3% (12h), 10% (36h), and 16.7% (60h) of target domain data for model fine-tuning, and adopts a layered freezing strategy (Layer-wise Freezing Strategy, LFS), that is, freezing the first two layers of CNN and LSTM, and only unfreezing the last layer of LSTM and the multi-head attention layer. In this way, the general features learned in the source domain can be retained, while also adapting to the specific patterns of the target domain, improving the prediction accuracy. Table 3 shows the prediction results under different migration data amounts.

[0049] Table 3 shows that after introducing 3.33% of the target domain data (12 hours), the cooling load MAPE dropped from 9.329% to 7.094%, and R² increased by 6.9 percentage points to 85.5%. When the fine-tuning data increased to 16.7% (60 hours), the power load MAPE was optimized to 7.916% (R²=89.6%), and the error was reduced by 13.2% compared with the zero sample. This improvement is due to the dynamic adjustment of the attention layer: the model updates the multi-head attention weights to strengthen the feature extraction of key time steps in the target domain (such as the peak of the working day), effectively alleviating the local distribution difference. The R² value of gas load prediction increased to 86.1% after fine-tuning, indicating that the model can quickly correct the feature mapping relationship with a small amount of data. It is worth noting that although a small amount of data (such as 3.3%) can significantly improve the prediction accuracy, when the amount of data is further increased (such as 16.7%), the improvement tends to converge.

[0050] This shows that the prediction model based on the Tnet-Metas framework has strong generalization capabilities and can make full use of source domain knowledge for accurate predictions even when the amount of fine-tuning data is small. In addition, the experimental results also verify the effectiveness of the layered freezing strategy in capturing the unique laws of the target domain by retaining the global features of the source domain and adjusting local parameters. It avoids the overfitting problem caused by excessive adjustment of the underlying feature extraction layer during the fine-tuning process. This method has important application value in scenarios where data can be acquired incrementally.

[0051] Table 3 Small sample fine-tuning prediction results;

[0052] Example 2 This embodiment provides a multi-energy load forecasting system under zero historical data of an integrated energy system, including: The data acquisition and preprocessing module is configured to: acquire the meteorological characteristics of the target park and the historical data of heating, cooling and electricity of the source domain group parks, and preprocess the acquired data; The source domain data determination module is configured to: perform park cross-correlation and generalization capability analysis on the pre-processed cooling, heating and electrical historical data of the source domain group parks to determine appropriate source domain data; The model building and training module is configured to: build a multi-energy load prediction model, use the Metas training strategy to train the model based on the source domain data, adjust the gradient weight according to the verification loss of the inner loop task and the source domain generalization probability, and obtain a trained prediction model; The model output module is configured to: input the pre-processed meteorological characteristics of the target park into the prediction model to obtain the prediction result.

[0053] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0054] In further embodiments, there is also provided: An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.

[0055] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0056] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0057] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.

[0058] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.

[0059] A computer program product includes a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0060] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer executable instructions, such as instructions included in a program module, which are executed in a device on a real or virtual processor of the target to perform the process / method as described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform specific tasks or implement specific abstract data types. In various embodiments, the functions of program modules can be combined or divided between program modules as needed. Machine executable instructions for program modules can be executed in local or distributed devices. In distributed devices, program modules can be located in local and remote storage media.

[0061] The computer program code for implementing the method of the present invention can be written in one or more programming languages. These computer program codes can be provided to the processor of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the computer or other programmable data processing device, causes the function / operation specified in the flow chart and / or block diagram to be implemented. The program code can be executed completely on a computer, partially on a computer, as an independent software package, partially on a computer and partially on a remote computer or completely on a remote computer or server.

[0062] In the context of the present invention, computer program codes or related data may be carried by any appropriate carrier to enable a device, apparatus or processor to perform the various processes and operations described above. Examples of carriers include signals, computer readable media, and the like. Examples of signals may include electrical, optical, radio, acoustic or other forms of propagation signals, such as carrier waves, infrared signals, and the like.

[0063] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0064] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A multi-energy load forecasting method for an integrated energy system with zero historical data, characterized in that: The following steps are involved: Obtain the meteorological characteristics of the target park and the historical data of heating, cooling and electricity of the source domain group parks, and pre-process the acquired data; For the pre-processed historical data of heating, cooling and electricity in the source domain group, we conduct cross-correlation and generalization analysis to determine the appropriate source domain data. Construct a multi-energy load prediction model, use the Metas training strategy, train the model based on source domain data, adjust the gradient weight according to the verification loss of the inner loop task and the source domain generalization probability, and obtain a trained prediction model; The preprocessed meteorological characteristics of the target park are input into the prediction model to obtain the prediction results.

2. The multi-energy load forecasting method under zero historical data of the integrated energy system according to claim 1 is characterized in that: Metas training strategy, specifically: Considering the parameter updates of different source domains, the loss value and weight value of each training task in the inner layer are fully applied to the update function of the outer layer.

3. The multi-energy load forecasting method under zero historical data of the integrated energy system according to claim 1 is characterized in that: The multi-energy load prediction model consists of an encoder module and a decoder module.

4. The multi-energy load forecasting method under zero historical data of the integrated energy system according to claim 3 is characterized in that: The encoder uses the CNN layer to perform feature extraction and dimensionality reduction through convolution operations, and passes the compressed data to the LSTM to further capture the long-term dependencies in the time series.

5. The multi-energy load forecasting method under zero historical data of the integrated energy system according to claim 4 is characterized in that: The multi-head attention mechanism is used to dynamically weight the relationships at different time steps and assign different hidden state probability weights to the LSTM, focusing on the impact of important information related to the cold and hot electrical loads.

6. The multi-energy load forecasting method under zero historical data of the integrated energy system according to claim 1, characterized in that: The park's mutual correlation and generalization ability analysis is as follows: The Time2vec algorithm is used to transform multiple load time series data of each source domain into periodic and linear parts, which are processed by the MIC algorithm and the MD algorithm respectively. K-means clustering analysis is constructed to stratify the source domains for generalization potential, and low-potential clusters are eliminated to reduce noise interference. MD and MIC are weightedly fused, and grid search is used to find the optimal parameter range, thus obtaining the cross-correlation score of the source domain group and a reasonable source domain range.

7. A multi-energy load forecasting system with zero historical data for an integrated energy system, characterized in that: include: The data acquisition and preprocessing module is configured to: acquire the meteorological characteristics of the target park and the historical data of heating, cooling and electricity of the source domain group parks, and preprocess the acquired data; The source domain data determination module is configured to: perform park cross-correlation and generalization capability analysis on the pre-processed cooling, heating and electrical historical data of the source domain group parks to determine appropriate source domain data; The model building and training module is configured to: build a multi-energy load prediction model, use the Metas training strategy to train the model based on the source domain data, adjust the gradient weight according to the verification loss of the inner loop task and the source domain generalization probability, and obtain a trained prediction model; The model output module is configured to: input the pre-processed meteorological characteristics of the target park into the prediction model to obtain the prediction result.

8. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 6 is completed.

9. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 6.

10. A computer program product, characterized in that The invention comprises a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images