Heat supply system heat load prediction method and device based on VMD-STL-SBiLSTM-Attention

Abstract

The invention discloses a VMD-STL-SBiLSTM-Attention-based heat supply system thermal load prediction method and device, and the method comprises the following steps: obtaining the operation data of a heat supply system, and carrying out the preprocessing of the operation data, and obtaining the preprocessing data; processing the preprocessed data through variational mode decomposition and season-trend decomposition based on local weighted regression to obtain data features; fusing the new data features and the predicted weather forecast data features to obtain a thermal load prediction data set; and constructing a prediction model according to the thermal load prediction data set, and performing prediction through the prediction model to obtain a thermal load prediction result. According to the method, short-term prediction of the thermal load of the heat supply system is realized, and the precision and accuracy of thermal load prediction are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of heating systems, and in particular relates to a method and device for predicting heat load of a heating system based on VMD-STL-SBiLSTM-Attention. Background Art

[0002] The heating system is a key part of the infrastructure of modern society. It not only ensures the comfort and health of residents in cold seasons, but also provides the necessary temperature conditions for industrial production. Efficient heating systems can reduce energy consumption and costs, reduce environmental pollution, and combat climate change. In addition, the stability of the heating system is crucial to public safety and social stability, and it is also an important component of urban planning and development.

[0003] The stable operation of the heating system depends on the high accuracy and efficiency of short-term heat load forecasting. Accurate load forecasting is a key factor in ensuring that the heating system can operate efficiently and energy-efficiently. Therefore, in-depth understanding and analysis of the inherent characteristics of heating system operation data and improving forecast accuracy have become a challenging and extremely important topic in the field. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method and device for predicting heat load of a heating system based on VMD-STL-SBiLSTM-Attention to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides a heat load prediction method for a heating system based on VMD-STL-SBiLSTM-Attention, comprising the following steps:

[0006] Acquire the operation data of the heating system and perform preprocessing to obtain preprocessing data;

[0007] Processing the preprocessed data by variational mode decomposition and seasonal trend decomposition based on local weighted regression to obtain data features;

[0008] The new data features are combined with the predicted weather forecast data features to obtain a heat load prediction data set;

[0009] A prediction model is constructed according to the heat load prediction data set, and prediction is performed using the prediction model to obtain a heat load prediction result.

[0010] Preferably, the pretreatment comprises:

[0011] The duplicate data in the operation data are deleted; the missing data are filled with the mean of the existing values ​​before and after the missing point; and the 3-sigma method is used to judge and process the outliers of the operation data.

[0012] Preferably, the acquired data features include:

[0013] Processing the preprocessed data by variational mode decomposition to obtain an intrinsic mode function of a specific center frequency and bandwidth;

[0014] The primary transient power, primary water supply temperature, primary return water temperature, secondary water supply temperature, secondary return water temperature, outdoor maximum temperature and outdoor minimum temperature data in the preprocessed data are processed by seasonal-trend decomposition of local weighted regression to obtain trend components, periodic components and residual components;

[0015] The number of intrinsic mode functions and regularization parameters in variational mode decomposition are optimized by particle swarm optimization method.

[0016] Preferably, the processing of the preprocessed data by variational mode decomposition comprises:

[0017] The preprocessed data is denoised and optimized using Wiener filtering to obtain K estimated center frequencies, and then each modal function and its center frequency are updated by alternating direction multiplication, and the intrinsic modal function of a specific center frequency and bandwidth is obtained by minimizing the total estimated bandwidth.

[0018] Preferably, the optimization by particle swarm optimization method includes:

[0019] Particle swarm initialization, randomly generating a group of particles, each particle is a parameter combination in variational mode decomposition;

[0020] The envelope entropy is used as the objective function to evaluate the performance of each particle. The smaller the envelope entropy, the better the parameter.

[0021] Update the speed and position of each particle according to the rules of the PSO algorithm, where the new position of the particle is determined by the current position, the individual historical optimal position, and the global historical optimal position;

[0022] Iterate the above steps, and when the maximum number of iterations is reached, the optimal number of intrinsic mode functions and regularization parameters are obtained.

[0023] Preferably, the obtaining of the trend component, the period component and the residual component comprises:

[0024] The hyperparameter period of the periodic component is set, each position in the period is smoothed by local estimated scatter plot smoothing to obtain an initial periodic component, and a seasonal component is obtained from the original sequence. The seasonal component is subtracted from the initial periodic component to obtain the periodic component;

[0025] Set the trend component hyperparameter trend to a value range of 1.5 to 2.5 times the period component hyperparameter period. Smooth the period component using local estimated scatter plot smoothing to obtain the trend component.

[0026] Subtract the trend component from the period component to obtain the residual component.

[0027] Preferably, the detection step of the prediction model includes:

[0028] The sample data is input into a stacked bidirectional long short-term memory network for calculation and the calculation result is output; the query, key, and value of the multi-head attention mechanism layer are calculated according to the calculation result, the query, key, and value are linearly transformed, and divided into 8 multi-heads, and processed through layer normalization calculation and a fully connected layer to output the prediction result.

[0029] The present invention also provides a heat load prediction device for a heating system based on VMD-STL-SBiLSTM-Attention, comprising:

[0030] A data acquisition module is used to obtain the operating data of the heating system and perform preprocessing to obtain preprocessed data;

[0031] A feature reconstruction module, used for processing the preprocessed data through variational mode decomposition and seasonal-trend decomposition based on local weighted regression to obtain data features;

[0032] A data set construction module, used for fusing the new data features with the predicted weather forecast data features to obtain a heat load prediction data set;

[0033] The prediction module is used to build a prediction model according to the heat load prediction data set, and perform prediction through the prediction model to obtain a heat load prediction result.

[0034] The present invention also provides a heat load prediction system for a heating system based on VMD-STL-SBiLSTM-Attention, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes the above method when executed by the processor.

[0035] The present invention also provides a storage medium, on which a computer program is stored, and the computer program executes the above method when running.

[0036] Compared with the prior art, the present invention has the following advantages and technical effects:

[0037] The present invention discloses a method and device for predicting heat load of a heating system based on VMD-STL-SBiLSTM-Attention, comprising the following steps: obtaining and preprocessing the operation data of the heating system to obtain preprocessed data; processing the preprocessed data by variational mode decomposition and seasonal-trend decomposition based on local weighted regression to obtain data features; fusing the new data features with the predicted weather forecast data features to obtain a heat load prediction data set; constructing a prediction model according to the heat load prediction data set, and predicting by the prediction model to obtain a heat load prediction result. The present invention realizes the short-term prediction of the heat load of the heating system, and effectively improves the precision and accuracy of the heat load prediction.

[0038] The present invention performs variational mode and seasonal-trend decomposition based on local weighted regression on the heating load data to generate new data features. Compared with the traditional use of directly collected data for heat load prediction, the present invention integrates the new data features generated by VMD-STL and weather forecast data, constructs a new data set, and improves the quality of heat load prediction data. Compared with simple deep learning models, the present invention proposes a prediction model based on a stacked bidirectional long short-term memory network module and a multi-head attention module. The stacked bidirectional long short-term memory network considers information from both forward and backward sequences to learn complex relationships in time series, can effectively process time series data, and improves the overall performance of the model. The multi-head attention module allows the model to simultaneously pay attention to multiple information at different positions at the same level, so that it can model more complex dependencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0040] Figure 1 A schematic diagram of the structure of a device according to an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of constructing a prediction model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0043] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0044] Embodiment 1:

[0045] The embodiment of the present invention provides a method for predicting heat load of a heating system based on VMD-STL-SBiLSTM-Attention, comprising the following steps:

[0046] Acquire the operation data of the heating system and perform preprocessing to obtain preprocessing data;

[0047] The preprocessed data are processed by variational mode decomposition and seasonal-trend decomposition based on local weighted regression to obtain data features;

[0048] The new data features are integrated with the predicted weather forecast data features to obtain a heat load prediction data set;

[0049] A prediction model is constructed based on the heat load prediction data set, and prediction is performed through the prediction model to obtain the heat load prediction results.

[0050] The collected heating system operation data include primary transient power, primary transient flow, primary water supply temperature, primary return water temperature, primary water supply pressure, primary return water pressure, secondary water supply temperature, secondary return water temperature, secondary water supply pressure, secondary return water pressure, indoor temperature, outdoor temperature - maximum temperature, outdoor temperature - minimum temperature, outdoor humidity, outdoor wind speed and solar radiation value every hour.

[0051] Furthermore, the preprocessing includes:

[0052] Delete the duplicate data in the operation data; use the mean of the existing values ​​before and after the missing point to fill the missing data; use the 3-sigma method to judge and process the outliers of the operation data.

[0053] Furthermore, the data features obtained include:

[0054] The preprocessed heating system data is processed by VMD-STL, and new data features are generated by VMD decomposition and STL decomposition;

[0055] As an implementation method of the embodiment of the present invention, VMD-STL includes:

[0056] VMD decomposition is used to process 16 data features including primary transient power, primary transient flow, primary water supply temperature, primary return water temperature, primary water supply pressure, primary return water pressure, secondary water supply temperature, secondary return water temperature, secondary water supply pressure, secondary return water pressure, indoor temperature, outdoor temperature-maximum temperature, outdoor temperature-minimum temperature, outdoor humidity, outdoor wind speed and solar radiation value every hour using variational mode decomposition to obtain a series of intrinsic mode functions with specific center frequency and bandwidth.

[0057] The VMD method uses Wiener filtering for noise reduction and optimizes related parameters to obtain K estimated center frequencies. Then, each modal function and its center frequency are updated using alternating direction multiplication to minimize the total estimated bandwidth. The calculation formula is as follows:

[0058]

[0059] The above formula is subject to the constraint that k μ k =f. Among them, μ k ={μ1,μ2,...,μ k} is the decomposed modal components of k different frequencies, ω k ={ω1,ω2,...,ω k} is the center frequency of each mode, δ(t is the unit pulse, and * is the convolution operation.

[0060] PSO optimization is used to optimize the number k of IMFs and the penalty parameter α for each feature, which is obtained by particle swarm optimization (PSO), that is, k 一次瞬态功率 , k 一次瞬态流量 , k 一次供水温度 , k 一次回水温度 , k 一次供水压力 , k 一次回水压力 , k 二次供水温度 , k 二次回水温度 , k 二次供水压力 , k 二次回水压力 , k 室内温度 , k 室外温度-最高温 , k 室外温度-最低温 , k 室外湿度 , k 室外风速 , k 太阳辐射值 .

[0061] The particle swarm optimization (PSO) method is used to optimize the number k of optimal IMFs and the penalty parameter α for each feature of VMD.

[0062] The multi-feature fusion dataset construction unit, for each training sample, will fuse each feature data obtained by optimizing and decomposing VMD and STL through PSO, and add the outdoor temperature-highest temperature forecast for the next hour, the outdoor temperature-lowest temperature forecast for the next hour, the outdoor humidity forecast for the next hour, the outdoor wind speed forecast for the next hour, and the solar radiation data forecast for the next hour. The dimension of the fused new sample data is k 一次瞬态功率 +k 一次瞬态流量 +k 一次供水温度 +k 一次回水温度 +k 一次供水压力 +k 一次回水压力 +k 二次供水温度 +k 二次回水温度 +k 二次供水压力 +k 二次回水压力 +k 室内温度 +k 室外温度-最高温 +k 室外温度-最低温 +k 室外湿度 +k 室外风速 +k 太阳辐射值 +21+5.

[0065] Further, the preprocessed data is processed by variational mode decomposition including:

[0066] The preprocessed data is denoised using Wiener filtering and optimized to obtain K estimated center frequencies. Each modal function and its center frequency are then updated by alternating direction multiplication. The intrinsic modal function with a specific center frequency and bandwidth is obtained by minimizing the total estimated bandwidth.

[0067] Further, the optimization by particle swarm optimization method includes:

[0068] Initialization, particle swarm initialization, randomly generate a group of particles, each particle is a parameter combination in TVF-EMD (the number of IMFs k and the penalty parameter α), where the value range of k is 2-15, the value range of α is 0.01-0.5, and the number of particle swarms is 30.

[0069] Define the objective function and use envelope entropy as the objective function to evaluate the performance of each particle. The minimum envelope entropy indicates that the found parameters are optimal.

[0070] Update particles and update the speed and position of each particle according to the rules of the PSO algorithm. The new position of the particle is determined by the current position, the individual historical best position (pbest) and the global historical best position (gbest).

[0071] Iterative optimization, repeat the above steps, when the preset stop condition is met, that is, when the maximum number of iterations is 1000, stop optimizing the number of IMFs k and the penalty parameter α.

[0072] The optimal result output is the number k of the optimal IMF for each feature, that is, k 一次瞬态功率 , k 一次瞬态流量 , k 一次供水温度 , k 一次回水温度 , k 一次供水压力 , k 一次回水压力 , k 二次供水温度 , k 二次回水温度 , k 二次供水压力 , k 二次回水压力 , k 室内温度 , k 室外温度-最高温 , k 室外温度-最低温 , k 室外湿度 , k 室外风速 , k 太阳辐射值 .

[0073] Furthermore, obtaining trend components, period components and residual components includes:

[0074] STL decomposition is used to perform STL decomposition on seven data features, namely, primary transient power, primary water supply temperature, primary return water temperature, secondary water supply temperature, secondary return water temperature, outdoor temperature-maximum temperature and outdoor temperature-minimum temperature, to obtain three new data features of each feature, namely, trend component, periodic component and residual component.

[0075] Periodic component extraction: set the periodic component hyperparameter period to a value between [24, 48, 72, 96, 120, 144, 168], and use LOESS to smooth each position within the period to obtain the initial periodic component S t Remove the seasonal component from the original sequence and obtain the deseasonalized sequence Y t -S t .

[0076] Trend component extraction, set the trend component hyperparameter trend, the value range is 1.5 to 2.5 times the period component hyperparameter period, and the corresponding value range is [36, 72, 108, 144, 180, 216, 252, 288, 336, 384, 420]. t -S t Use LOESS smoothing to get the trend component T t .

[0077] Subtract the trend component from the series after periodization to obtain the residual component.

[0078] R t =Y t -T t -S t

[0079] Repeat the above steps and gradually improve the periodicity and trend components through 100 iterative calculations.

[0080] Furthermore, the detection steps of the prediction model include:

[0081] The sample data is input into the stacked bidirectional long short-term memory network for calculation and output the calculation results; the query, key, and value of the multi-head attention mechanism layer are calculated based on the calculation results, and the query, key, and value are linearly transformed and divided into 8 multi-heads. They are processed through layer normalization calculation and a fully connected layer to output the prediction results.

[0082] In this embodiment, MAPE (Mean Absolute Percentage Error) is used to evaluate the performance of the model in both the training and testing stages of the prediction model. The calculation formula of MAPE is as follows:

[0083]

[0084] Among them, y i is the actual value, is the predicted value, and n is the number of samples.

[0085] MAPE is an intuitive measure of model accuracy, ranging from 0 to 1. The closer the value is to 0, the higher the prediction accuracy of the model.

[0086] In this embodiment, the prediction model is trained with the training data set, and the performance of the prediction model is tested with the test set, and finally the MAPE value of the prediction model under the test set is output.

[0087] Embodiment 2:

[0088] like Figure 1 As shown, the embodiment of the present invention further provides a heat load prediction device for a heating system based on VMD-STL-SBiLSTM-Attention, comprising:

[0089] A data acquisition module is used to obtain the operating data of the heating system and perform preprocessing to obtain preprocessed data;

[0090] The feature reconstruction module is used to process the preprocessed data through variational mode decomposition and seasonal-trend decomposition based on local weighted regression to obtain data features;

[0091] A data set construction module is used to merge new data features with predicted weather forecast data features to obtain a heat load prediction data set;

[0092] The prediction module is used to build a prediction model based on the heat load prediction data set, and perform prediction through the prediction model to obtain the heat load prediction result.

[0093] Specifically, it can be divided into: data acquisition module, data preprocessing module, VMD-STL module, multi-feature fusion data set construction module, prediction model construction module based on SBiLSTM and Multi Head Attention, and heat load prediction module;

[0094] The data acquisition module is used to collect relevant operating data of heating.

[0095] The collected heating system operation data include primary transient power, primary transient flow, primary water supply temperature, primary return water temperature, primary water supply pressure, primary return water pressure, secondary water supply temperature, secondary return water temperature, secondary water supply pressure, secondary return water pressure, indoor temperature, outdoor temperature - maximum temperature, outdoor temperature - minimum temperature, outdoor humidity, outdoor wind speed and solar radiation value every hour.

[0096] The collected heating system data is preprocessed.

[0097] As an implementation method of the embodiment of the present invention, the data preprocessing module is used to preprocess the data collected by the heating system. The data preprocessing module includes three units:

[0098] The deletion subunit is used to delete duplicate data in the heating system data; in this embodiment, the data stored in the database is extracted, and the duplicate data collected multiple times at the same time are deleted.

[0099] The filling subunit is used to fill the missing data in the heating system data with the mean of the existing values ​​before and after the point; in this embodiment, for missing data, the missing value is filled with the mean of the previous and next non-missing values ​​of the missing value.

[0100] The outlier detection subunit is used to judge and process the outliers of the heating system data using the 3-sigma method. In this embodiment, the 3-sigma method is used to judge the outliers of each collected data. If it is an outlier, the outlier is replaced by the mean of the previous and next non-missing values ​​of the missing value.

[0101] VMD-STL module, used to process the preprocessed heating system data and generate new data features through VMD decomposition and STL decomposition;

[0102] As an implementation method of an embodiment of the present invention, the VMD-STL module includes:

[0103] The VMD decomposition unit is used to process 16 data features of primary transient power, primary transient flow, primary water supply temperature, primary return water temperature, primary water supply pressure, primary return water pressure, secondary water supply temperature, secondary return water temperature, secondary water supply pressure, secondary return water pressure, indoor temperature, outdoor temperature-maximum temperature, outdoor temperature-minimum temperature, outdoor humidity, outdoor wind speed and solar radiation value every hour using variational mode decomposition to obtain a series of intrinsic mode functions with specific center frequency and bandwidth.

[0104] The VMD method uses Wiener filtering for noise reduction and optimizes related parameters to obtain K estimated center frequencies. Then, each modal function and its center frequency are updated using alternating direction multiplication to minimize the total estimated bandwidth. The calculation formula is as follows:

[0105]

[0106] The above formula is subject to the constraint that k μ k =f. Among them, μ k ={μ1,μ2,...,μ k} is the decomposed modal components of k different frequencies, ω k ={ω1,ω2,...,ω k} is the center frequency of each mode, δ(t is the unit pulse, and * is the convolution operation.

[0107] The PSO optimization unit is used to optimize the number k of IMFs and the penalty parameter α for each feature through the particle swarm optimization method (PSO), that is, k 一次瞬态功率 , k 一次瞬态流量 , k 一次供水温度 , k 一次回水温度 , k 一次供水压力 , k 一次回水压力 , k 二次供水温度 , k 二次回水温度 , k 二次供水压力 , k 二次回水压力 , k 室内温度 , k 室外温度-最高温 , k 室外温度-最低温 , k 室外湿度 , k 室外风速 , k 太阳辐射值 .

[0108] The particle swarm optimization (PSO) method is used to optimize the number k of optimal IMFs and the penalty parameter α for each feature of VMD.

[0109] The PSO optimization unit also uses the particle swarm optimization method (PSO) to optimize the period and trend hyperparameters of STL within the value range to obtain the optimal period and trend hyperparameter values ​​for each feature.

[0110] The multi-feature fusion dataset construction unit, for each training sample, will fuse each feature data obtained by optimizing and decomposing VMD and STL through PSO, and add the outdoor temperature-highest temperature forecast for the next hour, the outdoor temperature-lowest temperature forecast for the next hour, the outdoor humidity forecast for the next hour, the outdoor wind speed forecast for the next hour, and the solar radiation data forecast for the next hour. The dimension of the fused new sample data is k 一次瞬态功率 +k 一次瞬态流量 +k 一次供水温度 +k 一次回水温度 +k 一次供水压力 +k 一次回水压力 +k 二次供水温度 +k 二次回水温度 +k 二次供水压力 +k 二次回水压力 +k 室内温度 +k 室外温度-最高温 +k 室外温度-最低温 +k 室外湿度 +k 室外风速 +k 太阳辐射值 +21+5.

[0113] As an implementation of an embodiment of the present invention, the data set construction module includes:

[0114] The input feature construction unit is used to decompose and calculate the IMF1, IMF2…IMF by VMD for the primary transient power, primary transient flow, primary water supply temperature, primary return water temperature, primary water supply pressure, primary return water pressure, secondary water supply temperature, secondary return water temperature, secondary water supply pressure, secondary return water pressure, indoor temperature, outdoor temperature-highest temperature, outdoor temperature-lowest temperature, outdoor humidity, outdoor wind speed and solar radiation value data of the previous 48 hours. k And three new data features, namely trend component, periodic component and residual component, are obtained by decomposing the primary transient power, primary water supply temperature, primary return water temperature, secondary water supply temperature, secondary return water temperature, outdoor temperature-maximum temperature and outdoor temperature-minimum temperature of the first 48 hours through STL, and the outdoor temperature-maximum temperature forecast for the next hour, the outdoor temperature-minimum temperature forecast for the next hour, the outdoor humidity forecast for the next hour, the outdoor wind speed forecast for the next hour and the solar radiation forecast for the next hour are added as the input features of the data set.

[0115] The prediction feature construction unit is used to use the secondary pipe network water supply temperature in the next hour as the prediction feature of the data set.

[0116] The data set construction unit is used to integrate input features and prediction features to construct a heat load prediction data set; and divide the heat load prediction data set into a training set, a validation set, and a test set.

[0117] The constructed multi-feature fused heat load dataset of the heating system is randomly shuffled, 80% of the data is used as the training and validation sets, and the remaining 20% ​​of the data is used as the test dataset.

[0118] like Figure 2 As shown in the figure, based on the prediction model building module of SBiLSTM and Multi Head Attention, a prediction model using stacked bidirectional long short-term memory network and multi-head attention mechanism is constructed;

[0119] In this embodiment, the prediction model based on SBiLSTM and Multi Head Attention

[0120] The stacked bidirectional long short-term memory network (SBiLSTM) module is composed of two layers of bidirectional long short-term memory networks (BiLSTM). For each training sample, the data dimension is 48*(k 一次瞬态功率 +k 一次瞬态流量 +k 一次供水温度 +k 一次回水温度 +k 一次供水压力 +

[0121] k 一次回水压力 +k 二次供水温度 +k 二次回水温度 +k 二次供水压力 +k 二次回水压力 +

[0122] k 室内温度 +k 室外温度-最高温 +k 室外温度-最低温 +k 室外湿度 +k 室外风速 +k 太阳辐射值 +21)+5. Input the training samples into the prediction model for calculation and output the calculation results.

[0123] The output module outputs the calculation results of the prediction model to obtain the predicted value of the secondary water supply temperature of the heating system in the next hour.

[0124] The heat load prediction module is used to predict the heat load of the heating system based on the constructed prediction model and the heat load prediction data set to obtain the heat load prediction result.

[0125] As an implementation of an embodiment of the present invention, the heat load prediction module includes:

[0126] The heat load prediction unit is used to input the test set into the trained prediction model based on SBiLSTM and Multi HeadAttention to predict the heat load of the heating system, obtain the heat load prediction results, and realize the heat load prediction of each test sample;

[0127] The model performance evaluation unit uses the mean absolute percentage error to evaluate the performance of the integrated learning model in the training phase and the test phase to obtain the evaluation results.

[0128] As an implementation of the embodiment of the present invention, a heating system control module is also included, which is used to control the heating system based on the result of the heat load prediction. In this embodiment, the secondary water supply temperature obtained by prediction in the next hour can be used as the adjustment parameter of the heat exchange station, so as to effectively control the thermal power station part of the heating system and reduce the energy consumption of the thermal power station.

[0129] Embodiment three:

[0130] An embodiment of the present invention also provides a heating system heat load prediction system based on VMD-STL-SBiLSTM-Attention, comprising: a memory and a processor, wherein the memory stores a computer program run by the processor, and the computer program executes a heating system heat load prediction method when run by the processor.

[0131] Embodiment 4:

[0132] An embodiment of the present invention further provides a storage medium, on which a computer program is stored, and the computer program executes a method for predicting heat load of a heating system when running.

[0133] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A heat load prediction method for a heating system based on VMD-STL-SBiLSTM-Attention, characterized in that: The following steps are involved: Acquire the operation data of the heating system and perform preprocessing to obtain preprocessing data; Processing the preprocessed data by variational mode decomposition and seasonal trend decomposition based on local weighted regression to obtain data features; The data features are combined with the predicted weather forecast data features to obtain a heat load prediction data set; A prediction model is constructed according to the heat load prediction data set, and prediction is performed using the prediction model to obtain a heat load prediction result.

2. The method according to claim 1, characterized in that The pre-processing comprises: The duplicate data in the operation data are deleted; the missing data are filled with the mean of the existing values ​​before and after the missing point; and the 3-sigma method is used to judge and process the outliers of the operation data.

3. The method according to claim 1, characterized in that The obtained data features include: Processing the preprocessed data by variational mode decomposition to obtain an intrinsic mode function of a specific center frequency and bandwidth; The primary transient power, primary water supply temperature, primary return water temperature, secondary water supply temperature, secondary return water temperature, outdoor maximum temperature and outdoor minimum temperature data in the preprocessed data are processed by seasonal-trend decomposition of local weighted regression to obtain trend components, periodic components and residual components; The number of intrinsic mode functions and regularization parameters in variational mode decomposition are optimized by particle swarm optimization method.

4. The method according to claim 3, characterized in that: The processing of the preprocessed data by variational mode decomposition includes: The preprocessed data is denoised and optimized using Wiener filtering to obtain K estimated center frequencies, and then each modal function and its center frequency are updated by alternating direction multiplication, and the intrinsic modal function of a specific center frequency and bandwidth is obtained by minimizing the total estimated bandwidth.

5. The method according to claim 3, characterized in that: Optimization by particle swarm optimization method includes: Particle swarm initialization, randomly generating a group of particles, each particle is a parameter combination in variational mode decomposition; The envelope entropy is used as the objective function to evaluate the performance of each particle. The smaller the envelope entropy, the better the parameter. Update the speed and position of each particle according to the rules of the PSO algorithm, where the new position of the particle is determined by the current position, the individual historical optimal position, and the global historical optimal position; Iterate the above steps, and when the maximum number of iterations is reached, the optimal number of intrinsic mode functions and regularization parameters are obtained.

6. The method according to claim 3, characterized in that The obtaining of trend component, period component and residual component comprises: The hyperparameter period of the periodic component is set, each position in the period is smoothed by local estimated scatter plot smoothing to obtain an initial periodic component, and a seasonal component is obtained from the original sequence. The seasonal component is subtracted from the initial periodic component to obtain the periodic component; Set the trend component hyperparameter trend to a value range of 1.5 to 2.5 times the period component hyperparameter period. Smooth the period component using local estimated scatter plot smoothing to obtain the trend component. Subtract the trend component from the period component to obtain the residual component.

7. The method according to claim 1, characterized in that The detection step of the prediction model includes: The sample data is input into a stacked bidirectional long short-term memory network for calculation and the calculation result is output; the query, key, and value of the multi-head attention mechanism layer are calculated according to the calculation result, the query, key, and value are linearly transformed, and divided into 8 multi-heads, and processed through layer normalization calculation and a fully connected layer to output the prediction result.

8. A heat load prediction device for a heating system based on VMD-STL-SBiLSTM-Attention, characterized in that: include: A data acquisition module is used to obtain the operating data of the heating system and perform preprocessing to obtain preprocessed data; A feature reconstruction module, used for processing the preprocessed data through variational mode decomposition and seasonal-trend decomposition based on local weighted regression to obtain data features; A data set construction module, used for fusing the data features with the predicted weather forecast data features to obtain a heat load prediction data set; The prediction module is used to build a prediction model according to the heat load prediction data set, and perform prediction through the prediction model to obtain a heat load prediction result.

9. A heat load prediction system for a heating system based on VMD-STL-SBiLSTM-Attention, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the method for predicting the heat load of a heating system according to any one of claims 1 to 7 is executed.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when running, executes the method for predicting the heat load of a heating system according to any one of claims 1 to 7.

Images