Multi-element load and carbon emission prediction method, device, equipment and medium
By using local correlation analysis and Shapley and interpretation techniques to screen features, and combining a multi-task learning model with dilated causal convolution and multi-head sparse attention mechanisms, the weights are dynamically adjusted, solving the problems of feature screening and weight fixing in multivariate load and carbon emission prediction, and achieving high-precision and efficient joint prediction.
Patent Information
- Application Number
- CN202511992686.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-26
- Publication Date
- 2026-03-31
AI Technical Summary
Existing technologies for predicting multivariate loads and carbon emissions suffer from limitations in feature selection, insufficient ability of models to capture long-sequence dependencies, low prediction accuracy due to fixed weights, and insufficient generalization ability.
We employ local correlation analysis combined with Shapley and interpretation techniques to screen features, utilize a multi-task learning model with dilated causal convolution and multi-head sparse attention mechanisms, and dynamically adjust weights using gradient normalization algorithms to achieve joint prediction of multiple loads and carbon emissions.
It achieves accurate prediction of multiple loads and carbon emissions, improves prediction accuracy and generalization ability, reduces training time, and adapts to stable prediction in different scenarios.
Smart Images
Figure CN121767004A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of energy system data prediction technology, and in particular to a method, apparatus, equipment and medium for predicting multiple loads and carbon emissions. Background Technology
[0002] To address the severe challenges posed by global warming to human survival and development, reducing carbon emissions is an ongoing goal. Integrated energy systems (IES), as key carriers for efficient energy utilization and carbon emission control, require accurate prediction of their diverse loads and carbon emissions for optimized system scheduling and the development of scientific emission reduction strategies. However, current technologies face numerous challenges in achieving accurate predictions.
[0003] First, traditional feature selection methods have limitations. For example, they only focus on overall correlation across the entire time period, neglecting local differences in data across different dates and times within the week. For instance, when analyzing the relationship between electricity load and meteorological factors, they fail to consider the changes in correlation caused by different electricity consumption patterns during weekdays and nighttimes, and between weekends and weekdays. Alternatively, they may only select features from a statistical perspective, such as simply calculating correlation coefficients without deeply considering the actual contribution of features to the model's decision-making process. This results in a large amount of redundant information in the input data, increasing the computational burden on the model, interfering with the model's learning of effective features, and ultimately leading to low prediction accuracy.
[0004] Secondly, while existing recurrent neural network models such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) can process time-series data to some extent, their ability to capture long-term dependencies is limited, making it difficult to address the complex long-term dependencies of diverse loads and carbon emission data in integrated energy systems. Furthermore, a single model struggles to uncover the intricate coupling relationships between diverse loads (electricity, heat, and cooling loads) and between them and carbon emissions. For example, when analyzing the synergistic relationship between electricity and heat loads as they vary seasonally, a single model cannot comprehensively and deeply extract the characteristics of their mutual influence, leading to significant biases in prediction results.
[0005] Furthermore, fixed weights for each task are a common problem in multi-task prediction models. Different loads exhibit significantly different characteristics; for example, electrical loads are highly volatile due to factors such as user behavior and economic activities, while thermal loads are relatively stable and predictable due to factors like building insulation performance and outdoor temperature changes. Fixed weights prevent the model from dynamically adjusting to the actual difficulty and data characteristics of each task during training. This can easily lead to overtraining in some tasks (such as electrical load prediction) and underfitting in others (such as thermal load prediction), severely reducing the model's overall generalization ability and failing to provide stable and accurate prediction results in diverse scenarios. Summary of the Invention
[0006] The purpose of this application is to provide a method, apparatus, equipment, and medium for predicting multiple loads and carbon emissions, which can achieve accurate prediction of joint prediction of multiple loads and carbon emissions.
[0007] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for predicting multiple loads and carbon emissions, including: Data collection for the integrated energy system will use time series of each set time period within a set time period as the data source. Each set time period time series includes a target sequence and multiple feature sequences. The target sequence includes the multi-load and carbon emissions of the predicted date within the set time period, and the feature sequences include feature data of other dates within the set time period besides the predicted date. For each time series with a set time period, the correlation coefficient between the target series and each feature series is calculated, and feature series with correlation coefficients greater than a specific threshold are selected to obtain the selected feature series. Based on each target sequence and the corresponding filtered feature sequence, the Shapley and interpretation methods are used to filter the feature data to obtain the filtered feature data; the dataset consists of each target sequence and the corresponding filtered feature data; the input data for each sample data in the dataset includes the filtered feature data and time-series features, and the label data is the multivariate load and carbon emissions for the predicted date; The dataset is used to train the multi-task learning model to obtain a multivariate load and carbon emission prediction model; the multivariate load and carbon emission prediction model is used to predict the multivariate load and carbon emission amount for the prediction date within each set time period.
[0008] Optionally, the correlation coefficient is the Pearson correlation coefficient.
[0009] Optionally, the multi-load includes electrical load, heat load, and cooling load; The feature data in each feature sequence of the data source includes ambient temperature, humidity, and air pressure.
[0010] Optionally, the multi-task learning model includes a data input layer, a feature extraction layer, an information sharing layer, and a data output layer connected in sequence; The feature extraction layer uses dilated causal convolution for feature extraction. The information sharing layer employs a multi-head sparse attention mechanism and self-attention distillation for feature extraction.
[0011] Optionally, the time sequence features include hour, date, weekday, month, and season.
[0012] Optionally, the multi-task learning model is trained using the dataset to obtain a multivariate load and carbon emission prediction model, specifically including: A gradient normalization algorithm is used to adaptively adjust the parameter weights in the multi-task learning model; The gradient normalization algorithm is expressed as follows: ; in, Let be the overall loss during the t-th training iteration, and n be the number of prediction tasks. Let L be the gradient norm of the weighted loss function for prediction task i at training time t with respect to the parameters of the information sharing layer. for Expected value To predict the relative loss rate of task i, , , This represents the initial loss function value for prediction task i. To predict the loss function value of task i during the t-th training iteration, To predict the loss function value of task i during the t-th training iteration, for The expected value, where α is the adjustment factor. This represents the L1 norm.
[0013] Optionally, the time period can be set to one year, the time cycle to one week, and the prediction date to one day of the week.
[0014] Secondly, this application provides a multi-load and carbon emission prediction device, which applies the aforementioned multi-load and carbon emission prediction method, and the multi-load and carbon emission prediction device includes: The data source acquisition module is used to acquire data from the integrated energy system. It uses the time series of each set time period within a set time period as the data source. The time series of each set time period includes a target sequence and multiple feature sequences. The target sequence includes the multi-load and carbon emissions of the predicted date within the set time period. The feature sequences include feature data of other dates within the set time period besides the predicted date. The local filtering module is used to calculate the correlation coefficient between the target sequence and each feature sequence for each time series in a set time period, and filter out the feature sequences with a correlation coefficient greater than a specific threshold to obtain the filtered feature sequences. The global feature filtering module is used to filter feature data based on each target sequence and the corresponding filtered feature sequences, using the Shapley and interpretation methods to obtain filtered feature data; the dataset consists of each target sequence and the corresponding filtered feature data; the input data for each sample data in the dataset includes filtered feature data and time-series features, and the label data is the multivariate load and carbon emissions of the predicted date; The model training module is used to train the multi-task learning model using the dataset to obtain a multivariate load and carbon emission prediction model; the multivariate load and carbon emission prediction model is used to predict the multivariate load and carbon emission amount for the prediction date within each set time period.
[0015] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multi-load and carbon emission prediction method described in any one of the above.
[0016] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-load and carbon emission prediction method described in any one of the above.
[0017] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method, apparatus, equipment, and medium for predicting multiple loads and carbon emissions. It utilizes correlation coefficients to perform local correlation analysis, capturing the linear relationship between multiple loads and carbon emissions and influencing factors on different dates within a set time period. Then, it uses the Shapley and Interpretation Technique (SAET) to evaluate the contribution of global features. The results of both methods are combined to accurately select model input features, removing redundant and low-contribution features. During model training, the gradient normalization algorithm is used to dynamically adjust the weights of each prediction task, ensuring that each task progresses in a balanced manner during training and avoiding any one task dominating the training direction, thereby achieving accurate prediction of joint multiple loads and carbon emissions. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a flowchart illustrating a multi-element load and carbon emission prediction method provided in an embodiment of this application.
[0020] Figure 2 This is a schematic diagram illustrating the principle framework of a multi-element load and carbon emission prediction method provided in an embodiment of this application.
[0021] Figure 3 A comparison chart of spring multi-element load and carbon emission joint prediction results provided for an embodiment of this application.
[0022] Figure 4 This is a scatter plot of predicted and actual values for different models in spring, provided as an embodiment of this application.
[0023] Figure 5 A comparison chart of summer multi-load and carbon emission joint prediction results provided for an embodiment of this application.
[0024] Figure 6 A comparison chart of summer multi-load and carbon emission joint prediction results provided for an embodiment of this application.
[0025] Figure 7 A comparison chart of the joint prediction results of multiple loads and carbon emissions in autumn provided for an embodiment of this application.
[0026] Figure 8 A comparison chart of the joint prediction results of multiple loads and carbon emissions in autumn provided for an embodiment of this application.
[0027] Figure 9 A comparison chart of winter multi-load and carbon emission joint prediction results provided for an embodiment of this application.
[0028] Figure 10 A comparison chart of winter multi-load and carbon emission joint prediction results provided for an embodiment of this application.
[0029] Figure 11 This is a schematic diagram of the functional modules of a multi-load and carbon emission prediction device provided in an embodiment of this application.
[0030] Figure 12 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0032] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0033] In one exemplary embodiment, this application provides a method for predicting multiple loads and carbon emissions, such as... Figure 1 and Figure 2 As shown, the multi-load and carbon emission prediction method includes steps 101-104.
[0034] Step 101: Collect data from the integrated energy system. Use the time series of each set time period within the set time period as the data source. Each set time period time series includes a target sequence and multiple feature sequences. The target sequence includes the multi-load and carbon emissions of the predicted date within the set time period. The feature sequences include feature data of other dates within the set time period besides the predicted date.
[0035] Step 102: For each time series with a set time period, calculate the correlation coefficient between the target series and each feature series, and select the feature series with a correlation coefficient greater than a specific threshold to obtain the selected feature series.
[0036] Step 103: Based on each target sequence and the corresponding filtered feature sequence, the Shapley and interpretation methods are used to filter the feature data to obtain the filtered feature data; the dataset consists of each target sequence and the corresponding filtered feature data; the input data for each sample data in the dataset includes the filtered feature data and time-series features, and the label data is the multivariate load and carbon emissions for the predicted date.
[0037] Step 104: Train the multi-task learning model using the dataset to obtain a multivariate load and carbon emission prediction model; the multivariate load and carbon emission prediction model is used to predict the multivariate load and carbon emission amount for the prediction date within each set time period.
[0038] For example, if the forecast date is Friday, the characteristic data from the week preceding Friday and the time-series characteristics of Friday are input into the multivariate load and carbon emission forecasting model to obtain the predicted values of multivariate load and carbon emissions for Friday.
[0039] For example, to predict the multi-load and carbon emissions on Friday, the hourly characteristic data of the seven consecutive days before Friday (including characteristic data such as multi-load, carbon emissions, temperature, humidity, and air pressure) and the time-series characteristics of Friday (hour, date, week, month, and season) are input into the multi-load and carbon emission prediction model to obtain the predicted value of multi-load and carbon emissions for 24 hours on Friday.
[0040] In one exemplary embodiment, the time period is set to one year, the time cycle is set to one week, and the predicted date is one day of the week.
[0041] The correlation coefficient mentioned is the Pearson correlation coefficient.
[0042] Multi-load loads include electrical loads, heat loads, and cooling loads; The feature data in each feature sequence of the data source includes ambient temperature, humidity, and air pressure.
[0043] The time sequence features include hour, date, weekday, month, and season.
[0044] This application presents a method for predicting multiple loads and carbon emissions, which is a short-term joint prediction method for multiple loads and carbon emissions based on local-global feature screening and weight adaptive balancing. At the product level, it is intended to protect the software products, prediction systems, and related data processing tools developed around this method. These products can systematically integrate input data, perform feature screening, run prediction models, and output accurate prediction results to form a complete prediction solution that can be independently applied to integrated energy systems.
[0045] Steps 101-103 of this application implement local-global refined feature screening, extracting the time series of multivariate loads, carbon emissions and influencing factors on different days of the week, analyzing the local correlation between multivariate loads, carbon emissions and influencing factors on different days of the week, using model interpretability analysis methods to analyze the global contribution of input features of the joint prediction model, and combining the correlation analysis results with the actual contribution of features to refine the screening of input features of the joint prediction model.
[0046] 1) Local Correlation Analysis: Using a full-week time series over a year as the base data source, the multivariate load and carbon emission data for specific dates within the week are set as the target series. Simultaneously, the multivariate load and carbon emission data for other dates, along with various influencing factors such as ambient temperature, humidity, air pressure, socioeconomic factors, and energy price factors, are used as characteristic series. The Pearson correlation coefficient formula is applied. ; in, Let x be the Pearson correlation coefficient between sequence x and sequence y. and Let x and y be the values of sequence x and y at time n, respectively. and y and y are the average values of sequences x and y, respectively, and N is the sequence length.
[0047] Calculate the linear correlation between the target sequence and each feature sequence, screen out feature sequences whose absolute correlation coefficient is greater than a certain threshold, and preliminarily determine the features closely related to the prediction target.
[0048] In step 103, the core principle of Shapley's interpretation method is expressed by the formula: ; in, This refers to a data instance consisting of a specific load or carbon emission-related data from the multi-load analysis results screened in step 102. This is the predicted value of the machine learning model. Here, the machine learning model is a temporary model of the TCN-Informer architecture used for feature contribution evaluation. This temporary model is only used to simulate the decision logic of the subsequent joint prediction model and to build a mapping framework between features and prediction results. Its purpose is to remove low-contribution redundant features, rather than to achieve the final prediction. The output is for the linearly additive interpretable model. The baseline value is for the linearly additive explanatory model. Let be the Shapley value of the m-th input feature. It is a binary mapping vector. Indicates whether the feature exists. It can be 1 or 0, where M is the number of feature types in the feature data.
[0049] Calculate the Shapley value of the input features. In this process, the impact of different feature subset combinations on the prediction results is simulated to evaluate the contribution of each feature to the prediction results at the global level. For example, features are sorted from highest to lowest importance, and the performance improvement of the prediction model when each feature is added to different feature combinations is calculated, resulting in the Shapley value for each feature.
[0050] In one exemplary embodiment, the multi-task learning model includes a data input layer, a feature extraction layer (TCN), an information sharing layer (Informer), and a data output layer connected in sequence.
[0051] The feature extraction layer uses dilated causal convolution for feature extraction.
[0052] The information sharing layer employs a multi-head sparse attention mechanism and self-attention distillation for feature extraction.
[0053] The selected multivariate loads and carbon emission datasets (composed of target sequences and their corresponding selected feature data) are simultaneously input into a multi-task learning model. The input data features are extracted using a feature extraction layer (TCN) model, and the Informer is used to fully explore the coupling relationship between multivariate loads and carbon emissions. Furthermore, temporal features are introduced into the Informer input, combining hour, date, day of the week, month, and season to form temporal information, which is extracted using the Informer's internal embedding layer. Meteorological factors for the day to be predicted are then incorporated into the Informer decoder input data, and the multivariate load and carbon emission prediction results are obtained through a dense model in the data output layer.
[0054] Dilated causal convolution in the feature extraction layer is represented as: ; in, d The expansion coefficient, Before time t di Historical sequence values at time points For convolution operations, n is the filter size for one-dimensional convolution. This is the output of the dilated causal convolution at time t.
[0055] By combining convolutional kernels with different dilation rates, it is possible to expand the model's sensitivity to long-sequence historical data without increasing too many parameters.
[0056] In the information sharing layer, the multi-head sparse attention mechanism is represented as: ; In the formula: Q , K , V These are query and key-value matrix, respectively. , , . d The dimension of the key vector. This can prevent gradient vanishing. , and It represents the linear transformation matrix corresponding to the query, key, and value. For normalized exponential function operations, For input.
[0057] When processing long sequence data, the similarity between the query matrix and the key matrix is calculated, and only the data with high relevance to the current position is focused on, thereby reducing computational complexity.
[0058] Simultaneously employ attention distillation: ; By transferring knowledge from a complex model (teacher model) to a simpler model (student model), the number of model parameters is reduced while maintaining model performance, thus improving model training efficiency. These two methods allow for a deeper exploration of the complex coupling relationship between multiple loads and carbon emissions.
[0059] In the formula: This is an attention block that incorporates a sparse attention mechanism. Represents a one-dimensional convolution operation. Represents the activation function. Represents max pooling operation. This indicates the first [item] after attention block processing. j +1 layer output feature sequence. It is the +1 layer. j Layer feature sequence The next layer of feature representation is obtained after attention distillation operations consisting of "one-dimensional convolution (Conv1d) + ELU activation + max pooling".
[0060] In an exemplary embodiment, the dataset is used to train a multi-task learning model to obtain a multivariate load and carbon emission prediction model, specifically including: using a gradient normalization algorithm to adaptively adjust the parameter weights in the multi-task learning model.
[0061] Calculate the loss function for multiple load and carbon emission prediction tasks during the training of the multi-task learning model, construct the model weight update loss function based on the loss function, obtain the weight gradient of the function for each prediction task, and adaptively adjust the weights of each prediction task to ensure the training balance of the joint prediction model.
[0062] The gradient normalization algorithm is expressed as follows: ; in, Let be the overall loss during the t-th training iteration, and n be the number of prediction tasks. Let L be the gradient norm of the weighted loss function for prediction task i at training time t with respect to the parameters of the information sharing layer. for Expected value To predict the relative loss rate for task i, a smaller loss rate indicates that the task has been trained more thoroughly. , , This represents the initial loss function value for prediction task i. To predict the loss function value of task i during the t-th training iteration, To predict the loss function value of task i during the t-th training iteration, for The expected value, where α is a regulation factor that can adjust the model weight update rate. This represents the L1 norm. During model training, the weight gradients of the loss function for different prediction tasks can be calculated to update the weights. Then, multiply by the corresponding learning rate to achieve prediction task weight update and adaptive adjustment.
[0063] The loss function is the "data input" to the gradient normalization algorithm, which dynamically adjusts task weights based on the gradients of these losses.
[0064] Calculate the weights and update the loss function. During model training, dynamically adjust the weights based on the decay characteristics of the loss rate for each task.
[0065] This application first utilizes Pearson correlation coefficients for local correlation analysis to capture the linear relationship between multivariate loads and carbon emissions and influencing factors on different days within a week. Then, it employs Shapleyga and interpretation techniques to evaluate the global feature contribution, combining the results of both to accurately select model input features and remove redundant and low-contribution features. Next, a TCN-Informer multi-task learning model is constructed. The TCN layer uses dilated causal convolution and residual connections to mine effective features and long-term dependencies in the data, while the Informer layer deeply analyzes the coupling relationship between multivariate loads and carbon emissions through sparse attention mechanisms and attention distillation. Finally, the prediction results are output through a Dense layer. Furthermore, during the training of the multi-task learning model, an adaptive weight balancing strategy is used. A multi-task loss function is constructed based on mean squared error, and gradient normalization algorithms are used to dynamically adjust the weights of each prediction task, ensuring balanced progress of each task during training and preventing any one task from dominating the training direction. Ultimately, this achieves joint and accurate prediction of multivariate loads such as electricity, heat, and cooling, as well as carbon emissions in integrated energy systems, providing core technical support for efficient scheduling, planning, and carbon emission control of energy systems.
[0066] The technical effectiveness of the method described in this application has been fully verified through experiments using actual operational data from the integrated energy system at Arizona State University, Tempe Campus.
[0067] In terms of prediction accuracy, compared with traditional prediction methods, such as a single LSTM model, which struggles to effectively capture long-sequence dependencies and complex coupling relationships, the mean absolute percentage error (MAPE) of electricity load prediction in multivariate load forecasting is typically between 4% and 6%. The method in this application can reduce the MAPE of electricity load forecasting to 2.12%. Regarding carbon emission prediction, the MAPE of the traditional GRU model is between 3% and 5%, while the method in this application can reduce it to 1.15%, as shown in the spring section of Table 1. This is mainly due to the refined feature selection in step 102, which removes noise interference and provides high-quality input to the model; and the deep mining of the complex coupling relationship between multivariate loads and carbon emissions by the TCN-Informer model in step 104, enabling the model to learn data features and patterns more accurately. In Table 1, GN-FS-TCN-Informer is a multi-task learning architecture of gradient normalization (GN) + refined feature selection (FS) + TCN-Informer. FS-TCN-Informer contains only a refined feature selection (FS) + TCN-Informer multi-task learning architecture (without a gradient normalization module) and is used to verify the role of gradient normalization in multi-task balancing.
[0068] Table 1 Comparison of prediction performance of the TCN-Informer model in this application with other models In terms of computational efficiency, before feature selection, the model input features have a large dimension, for example, the original data dimension can reach [64, 168, 10] (representing the number of samples, time step, and number of features, respectively). After local-global feature selection, the dimension is reduced to [64, 24, 30]. This significantly reduces the amount of data processing during model training, shortening the training time per batch from 13 seconds in traditional methods to 2 seconds. Taking a model training task with 1000 samples and 100 training rounds as an example, the total training time of traditional methods is about 13000 seconds, while the method in this application only requires 2000 seconds, greatly shortening the model training cycle, improving the response speed of the prediction system, and enabling it to provide prediction results for actual decision-making more quickly.
[0069] In terms of generalization ability, traditional models use fixed weights, which cannot adapt to differences in workload characteristics, leading to underfitting of some tasks and large fluctuations in prediction performance across different scenarios. In contrast, the adaptive weight balancing method in this application dynamically adjusts the weights of each task, enabling the prediction model to maintain stable prediction performance in complex scenarios such as different seasons and different workday types. Figures 3-10 As shown, Figure 4The horizontal axis represents the actual value, and the vertical axis represents the model's predicted value. Different colored points represent different models (blue for GN-FS-TCN-Informer, green for FS-TCN-Informer, purple for TCN-Informer, yellow for TCN-Dense, and red for the actual load value). The red line is the y=x reference line. The closer a point is to the line, the smaller the deviation between the model prediction and the actual value. For example, in winter, the mean absolute error (MAE) of the cooling load prediction is reduced by 68.06% compared to the fixed-weight model. In cross-regional applications, the multi-task model of this application can quickly adapt to regions with different energy structures and climate conditions, maintaining high prediction accuracy and providing reliable assurance for the optimized scheduling of integrated energy systems in diverse scenarios.
[0070] Based on the same inventive concept, this application also provides a multi-load and carbon emission prediction device for implementing the multi-load and carbon emission prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the multi-load and carbon emission prediction device provided below can be found in the limitations of the multi-load and carbon emission prediction method described above, and will not be repeated here.
[0071] In one exemplary embodiment, such as Figure 11 As shown, a multi-load and carbon emission prediction device is provided, which applies the multi-load and carbon emission prediction method described above. The multi-load and carbon emission prediction device includes: The data source acquisition module is used to acquire data from the integrated energy system. It uses the time series of each set time period within a set time period as the data source. The time series of each set time period includes a target sequence and multiple feature sequences. The target sequence includes the multi-load and carbon emissions of the predicted date within the set time period. The feature sequences include feature data of other dates within the set time period besides the predicted date.
[0072] The local filtering module is used to calculate the correlation coefficient between the target sequence and each feature sequence for each time series in a set time period, and filter out the feature sequences with a correlation coefficient greater than a specific threshold to obtain the filtered feature sequences.
[0073] The global feature filtering module is used to filter feature data based on each target sequence and the corresponding filtered feature sequence, using the Shapley and interpretation methods to obtain filtered feature data; the dataset consists of each target sequence and the corresponding filtered feature data; the input data for each sample data in the dataset includes filtered feature data and time-series features, and the label data is the multivariate load and carbon emissions of the predicted date.
[0074] The model training module is used to train the multi-task learning model using the dataset to obtain a multivariate load and carbon emission prediction model; the multivariate load and carbon emission prediction model is used to predict the multivariate load and carbon emission amount for the prediction date within each set time period.
[0075] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 12 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores multi-variable load and carbon emission prediction data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a multi-variable load and carbon emission prediction method.
[0076] Those skilled in the art will understand that Figure 12 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.
[0077] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0078] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.
[0079] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0080] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).
[0081] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units, etc., and are not limited to these.
[0082] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0083] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for predicting multiple loads and carbon emissions, characterized in that, The multi-factor load and carbon emission prediction method includes: Data collection for the integrated energy system will use time series of each set time period within a set time period as the data source. Each set time period time series includes a target sequence and multiple feature sequences. The target sequence includes the multi-load and carbon emissions of the predicted date within the set time period, and the feature sequences include feature data of other dates within the set time period besides the predicted date. For each time series with a set time period, the correlation coefficient between the target series and each feature series is calculated, and feature series with correlation coefficients greater than a specific threshold are selected to obtain the selected feature series. Based on each target sequence and the corresponding filtered feature sequence, the Shapley and interpretation methods are used to filter the feature data to obtain the filtered feature data; the dataset consists of each target sequence and the corresponding filtered feature data; the input data for each sample data in the dataset includes the filtered feature data and time-series features, and the label data is the multivariate load and carbon emissions for the predicted date; The dataset is used to train the multi-task learning model to obtain a multivariate load and carbon emission prediction model; the multivariate load and carbon emission prediction model is used to predict the multivariate load and carbon emission amount for the prediction date within each set time period.
2. The method for predicting multiple loads and carbon emissions according to claim 1, characterized in that, The correlation coefficient mentioned is the Pearson correlation coefficient.
3. The method for predicting multiple loads and carbon emissions according to claim 1, characterized in that, Multi-load loads include electrical loads, heat loads, and cooling loads; The feature data in each feature sequence of the data source includes ambient temperature, humidity, and air pressure.
4. The method for predicting multiple loads and carbon emissions according to claim 1, characterized in that, The multi-task learning model includes a data input layer, a feature extraction layer, an information sharing layer, and a data output layer connected in sequence. The feature extraction layer uses dilated causal convolution for feature extraction. The information sharing layer employs a multi-head sparse attention mechanism and self-attention distillation for feature extraction.
5. The method for predicting multiple loads and carbon emissions according to claim 1, characterized in that, The time sequence features include hour, date, weekday, month, and season.
6. The method for predicting multiple loads and carbon emissions according to claim 4, characterized in that, The multi-task learning model is trained using the dataset to obtain a multivariate load and carbon emission prediction model, which specifically includes: A gradient normalization algorithm is used to adaptively adjust the parameter weights in the multi-task learning model; The gradient normalization algorithm is expressed as follows: ； in, Let be the overall loss during the t-th training iteration, and n be the number of prediction tasks. Let L be the gradient norm of the weighted loss function for prediction task i at training time t with respect to the parameters of the information sharing layer. for Expected value To predict the relative loss rate of task i, , , This represents the initial loss function value for prediction task i. To predict the loss function value of task i during the t-th training iteration, To predict the loss function value of task i during the t-th training iteration, for The expected value, where α is the adjustment factor. This represents the L1 norm.
7. The method for predicting multiple loads and carbon emissions according to claim 1, characterized in that, The time period is set to one year, the time cycle is set to one week, and the prediction date is one day of the week.
8. A multi-load and carbon emission prediction device, characterized in that, The multi-load and carbon emission prediction device applies the multi-load and carbon emission prediction method according to any one of claims 1-7, and the multi-load and carbon emission prediction device comprises: The data source acquisition module is used to acquire data from the integrated energy system. It uses the time series of each set time period within a set time period as the data source. The time series of each set time period includes a target sequence and multiple feature sequences. The target sequence includes the multi-load and carbon emissions of the predicted date within the set time period. The feature sequences include feature data of other dates within the set time period besides the predicted date. The local filtering module is used to calculate the correlation coefficient between the target sequence and each feature sequence for each time series in a set time period, and filter out the feature sequences with a correlation coefficient greater than a specific threshold to obtain the filtered feature sequences. The global feature filtering module is used to filter feature data based on each target sequence and the corresponding filtered feature sequences, using the Shapley and interpretation methods to obtain filtered feature data; the dataset consists of each target sequence and the corresponding filtered feature data; the input data for each sample data in the dataset includes filtered feature data and time-series features, and the label data is the multivariate load and carbon emissions of the predicted date; The model training module is used to train the multi-task learning model using the dataset to obtain a multivariate load and carbon emission prediction model; the multivariate load and carbon emission prediction model is used to predict the multivariate load and carbon emission amount for the prediction date within each set time period.
9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the multi-load and carbon emission prediction method according to any one of claims 1-7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the multi-load and carbon emission prediction method as described in any one of claims 1-7.