A method for establishing a secondary side heating time series model for a heating station in a central heating system

Abstract

The present invention belongs to the field of centralized heating, relates to data analysis technology, and is used to solve the problem that traditional methods have poor spatial feature learning ability of factors affecting the secondary side water supply temperature of a thermal power station. Specifically, it is a method for establishing a secondary side heating time series model of a thermal power station in a centralized heating system, and comprises the following steps: obtaining historical secondary side heating characteristics, meteorological characteristics and date characteristics that affect the secondary side water supply temperature of a thermal power station in a heating system; calculating the correlation between meteorological data and the secondary side water supply temperature of the thermal power station; the present invention determines meteorological characteristics that are strongly correlated with the secondary side water supply temperature of the thermal power station, and constructs a secondary side heating time series data set of the thermal power station in combination with historical secondary side heating characteristics and date characteristics; it can take into account that the secondary side heating of the thermal power station is affected by meteorological and holiday factors, and selecting appropriate characteristics can increase the prediction accuracy of a subsequent secondary side heating time series model of the thermal power station, and can reduce the time complexity and space complexity of the model.

Description

Technical Field

[0001] The present invention belongs to the field of centralized heating, relates to data analysis technology, and specifically is a method for establishing a secondary side heating time series model of a thermal power station in a centralized heating system. Background Art

[0002] With the rapid development of heating systems, heat users have higher and higher requirements for heating comfort, and the water supply temperature on the secondary side of the heating station has a significant impact on the comfort of heat users. Reasonable control of the water supply temperature on the secondary side of the heating station is an important factor in ensuring the normal operation of the heating system and meeting the heat needs of users. The water supply temperature on the secondary side of the heating station directly affects the regulation of indoor temperature. If the water supply temperature is too high, it may cause the indoor temperature to be too high, making users feel overheated, stuffy, or even hot and dry, which is not conducive to human health and comfort. On the contrary, if the water supply temperature on the secondary side of the heating station is too low, it may cause the indoor temperature to be too low, making users feel cold and unable to meet basic heating needs.

[0003] The invention patent with the announcement number CN116817357B discloses a heating control method and device based on a heating station, a building and a household. The method inputs the weather data within a predetermined time after the current time and the load data of the target heating station within the predetermined time into a time series load prediction model to predict the load data of the target heating station in the future predetermined time period under the target room temperature; based on the load data of the target heating station in the future predetermined time period and the current actual load of the target heating station, the load adjustment signal of the target heating station is determined; based on the household temperature data of the target heating station at the current time and the target room temperature, the household valve adjustment signal of the heating of the target heating station is determined;

[0004] However, the change of the secondary side water supply temperature of the thermal power station is not only affected by its own change law, but also by factors such as date and outdoor weather. The traditional method has poor spatial feature learning ability of the factors affecting the secondary side water supply temperature of the thermal power station, and cannot handle sequences with dynamic time dependence and cannot analyze the influence of other factors on the change of the secondary side water supply temperature of the thermal power station. How to solve the problem of spatiotemporal dependence of the factors affecting the secondary side heating of the thermal power station and realize accurate prediction of the secondary side water supply temperature of the thermal power station is an urgent problem to be solved. Summary of the invention

[0005] The purpose of the present invention is to provide a method for establishing a secondary side temperature supply time series model of a heating station in a centralized heating system, which is used to solve the problem that the traditional method has poor spatial feature learning ability of the factors affecting the secondary side water supply temperature of the heating station;

[0006] The technical problem to be solved by the present invention is: how to provide a method for establishing a secondary side heating timing model of a thermal power station in a centralized heating system that can process sequences with dynamic time dependence and analyze the influence of other factors on the change of the secondary side water supply temperature of the thermal power station.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] A method for establishing a secondary side heating time series model of a heating station in a centralized heating system comprises the following steps:

[0009] Step S1: Acquire historical secondary side heating characteristics, meteorological characteristics and date characteristics that affect the secondary side water supply temperature of the heating power station of the heating system; wherein the meteorological characteristics include outdoor temperature, light intensity, wind speed and humidity; the date characteristics include working days, holidays, months and weeks;

[0010] Step S2: Calculate the correlation between meteorological data and the secondary side water supply temperature of the thermal power station, determine the meteorological features that are strongly correlated with the secondary side water supply temperature of the thermal power station, and construct a secondary side heating time series data set of the thermal power station by combining the average, maximum, minimum and maximum difference of the meteorological features with the historical secondary side heating features and date features;

[0011] Step S3: Establish a graph neural network model including a graph learning module, a first graph convolution module, a second graph convolution module, a time series convolution module and an output module; extract the dynamic spatial relationship of the data through the graph learning module and output the graph adjacency matrix; extract the spatial dependency through the first graph convolution module; perform deep information extraction through the second graph convolution module and input the time-varying information into the time series; extract the time dependency through the TCN time series convolution neural network and the GRU gated recurrent unit network in the time series convolution module, and perform time dependency fusion; obtain the secondary side heating prediction result of the thermal power station through the output module;

[0012] Step S4: After training and learning the graph neural network model, a secondary side heating timing prediction model for the thermal power station is established.

[0013] Further, in step S2, the correlation between the meteorological data and the secondary side water supply temperature of the thermal power station is calculated, and the meteorological characteristics strongly correlated with the secondary side water supply temperature of the thermal power station are determined, specifically including:

[0014] Based on the Copula entropy correlation analysis method, the correlation between meteorological data and the secondary water supply temperature of the thermal power station was calculated. After sorting and analyzing the meteorological data variables according to the size of the Copula entropy, the outdoor temperature and light intensity were determined to be the key influencing factors.

[0015] Furthermore, in step S2, the secondary side heating time series data set of the thermal power station is constructed by combining the historical secondary side heating characteristics and date characteristics, which specifically includes:

[0016] The outdoor average temperature in the 24 hours before the day to be predicted, the maximum temperature on the day to be predicted, the minimum temperature on the day to be predicted and the maximum temperature difference are selected as outdoor temperature features;

[0017] The average light intensity in the 24 hours before the day to be predicted, the maximum light intensity on the day to be predicted, the minimum light intensity on the day to be predicted and the maximum light intensity difference are selected as light intensity features;

[0018] Based on the selected outdoor temperature characteristics and light intensity characteristics, combined with the historical secondary side heating characteristics and date characteristics one week before the predicted day, a secondary side heating time series dataset of the thermal power station is constructed.

[0019] Furthermore, in step S2, it also includes: identifying outdoor temperature patterns according to outdoor temperature characteristics, and dividing them into outdoor low temperature patterns, outdoor medium temperature patterns, and outdoor high temperature patterns; identifying light intensity patterns according to light intensity characteristics, and dividing them into low light intensity patterns, medium light intensity patterns, and high light intensity patterns; dividing meteorological characteristics into different temperature-light pattern categories according to outdoor temperature patterns and light intensity patterns, and constructing secondary heating time series data subsets of thermal power stations under different temperature-light pattern categories in combination with historical secondary heating characteristics and date characteristics; and, in step S3, it also includes: using the secondary heating time series data subsets of thermal power stations under different temperature-light pattern categories, establishing graph neural network models under different temperature-light pattern categories, and obtaining secondary heating prediction results of thermal power stations under different temperature-light pattern categories; and, in step S4, it also includes: using the secondary heating time series data subsets of thermal power stations under different temperature-light pattern categories to train and learn the corresponding graph neural network models, and then establishing secondary heating timing series prediction sub-models of thermal power stations under different temperature-light pattern categories.

[0020] Furthermore, in step S3, the first graph convolution module, the second graph convolution module and the temporal convolution module are used as the first spatiotemporal convolution module; the graph neural network model includes multiple spatiotemporal convolution modules; residual connections are added to the input end of the second graph convolution module and the output end of the temporal convolution module; the graph neural network model also includes a jump connection module for making the sequence lengths from input to output the same.

[0021] Furthermore, in step S3, the dynamic spatial relationship of the data is extracted through the graph learning module, and the graph adjacency matrix is ​​output, which specifically includes:

[0022] Heating data diagram for secondary side of thermal power station To describe the topological structure of multiple variables in secondary side heating; is a set of N variable nodes; is the set of edges that represent connectivity between nodes; It is a picture The adjacency matrix represents the connection weights between variable nodes;

[0023] The graph learning module adopts an asymmetric adjacency matrix, which is expressed as:

[0024] ;

[0025] ;

[0026] ;

[0027] , To initialize the node; , are model parameters; is the saturation rate of the activation function.

[0028] Furthermore, in step S3, the spatial dependency of features is processed by fusing the information of a node with the information of its neighbors through the first graph convolution module, which is expressed as:

[0029] ;

[0030] For the The node feature vector of the layer is composed of the node feature vector of the previous layer and the adjacency matrix Calculated; is the activation function; It is The weight matrix of the layer; , is a unit matrix of order N; is a diagonal matrix.

[0031] Furthermore, in step S3, deep information extraction is performed through a second graph convolution module, including:

[0032] In the graph convolutional network (GCN), each node performs a convolution operation with its neighboring nodes to fuse the information of the neighboring nodes into the feature representation of the current node. Through multiple convolution operations, the information of the entire graph can be fused into a global feature representation.

[0033] In addition to using the structural information of the graph's edges and vertex connections, GCN can also use attribute information attached to the graph structure, including node features and edge weights, to provide GCN with additional contextual information and determine the structure and relationships of the graph.

[0034] Furthermore, the TCN temporal convolutional neural network applies a convolution operation on a one-dimensional time series, capturing the dependencies in the time series by moving the convolution kernel in the time dimension; each hidden layer of the TCN temporal convolutional neural network contains a set of convolutional layers, each convolutional layer consists of a one-dimensional convolution operation and a nonlinear activation function, and the output of the one-dimensional convolution operation is:

[0035] ;

[0036] For the Layer Outputs; For the Layer The weight of the convolution kernel; is the length of the convolution kernel; is a nonlinear activation function; is the bias term; For the Layer Outputs;

[0037] The temporal dependency extracted by the gated recurrent unit network GRU is expressed as:

[0038] ;

[0039] is the hidden layer state at the previous moment t-1; is the secondary heating monitoring value of the thermal power station at the current time t; is an update gate used to control the amount of state information before entering the current state; To control the reset gate that forgets the state information of the previous moment, if the output is 0, the secondary heating sequence information of the heating station at the previous moment will be ignored; if the output is 1, the secondary heating sequence information of the heating station at the previous moment will be completely memorized and brought into the next moment; is the candidate hidden state at the current time t; is the output state at the current time t; , They are the weight matrix and bias vector of the update gate respectively; , are the weight matrix and bias vector of the reset gate respectively; , are the weight matrix and bias vector of the candidate hidden state respectively; is the activation function.

[0040] Furthermore, in step S3, performing time dependency fusion includes:

[0041] The attention mechanism is introduced to allocate the attention of the temporal dependency between the two, and they are fused through the fully connected layer to output the final temporal dependency result.

[0042] The present invention has the following beneficial effects:

[0043] 1. The present invention obtains the historical secondary side heating characteristics, meteorological characteristics and date characteristics that affect the secondary side water supply temperature of the heating station of the heating system; wherein the meteorological characteristics include outdoor temperature, light intensity, wind speed and humidity; the date characteristics include working days, holidays, months and weeks; calculates the correlation between meteorological data and the secondary side water supply temperature of the heating station, determines the meteorological characteristics that are strongly correlated with the secondary side water supply temperature of the heating station, and constructs the secondary side heating time series data set of the heating station in combination with the historical secondary side heating characteristics and date characteristics; it can consider that the secondary side heating of the heating station is affected by meteorological and holiday factors, and after reducing the input characteristics of the model, selecting appropriate characteristics can increase the prediction accuracy of the subsequent secondary side heating time series model of the heating station, and reduce the time complexity and space complexity of the model;

[0044] 2. The present invention can take into account the powerful ability of graph neural networks in capturing data relationships, apply graph neural networks to the secondary side heating prediction of thermal power stations to obtain dynamic dependencies of variables, and use graph learning modules to adaptively obtain dynamic adjacency matrices to better obtain the impact of other factors on the numerical changes of the secondary side heating of thermal power stations; in the calculation of complex spatial dependencies and dynamic time change sequence problems, the graph convolution module and the time series convolution module are used to capture spatial correlation and time correlation to realize the spatiotemporal prediction of graph neural networks, especially the TCN time series convolution neural network and the GRU gated recurrent unit network in the time series convolution module are used to extract time dependencies respectively, and perform time dependency fusion, which improves the accuracy of time dependency calculation and ensures the effectiveness and accuracy of the graph neural network in predicting the secondary side heating of thermal power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0047] The technical scheme of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] like Figure 1 As shown, a method for establishing a secondary side heating time series model of a heating station in a centralized heating system includes:

[0049] S1: Obtain the historical secondary side heating characteristics, meteorological characteristics, and date characteristics that affect the secondary side water supply temperature of the heating system thermal power station; the meteorological characteristics include outdoor temperature, light intensity, wind speed, and humidity; the date characteristics include working days, holidays, months, and weeks;

[0050] S2: Calculate the correlation between meteorological data and the secondary water supply temperature of the thermal power station, determine the meteorological characteristics that are strongly correlated with the secondary water supply temperature of the thermal power station, and construct the secondary water supply time series data set of the thermal power station by combining the average, maximum, minimum and maximum difference of the meteorological characteristics with the historical secondary water supply characteristics and date characteristics;

[0051] Calculate the correlation between meteorological data and the secondary water supply temperature of the thermal power station, and determine the meteorological characteristics that are strongly correlated with the secondary water supply temperature of the thermal power station, including:

[0052] Based on the Copula entropy correlation analysis method, the correlation between meteorological data and the secondary water supply temperature of the thermal power station is calculated. After sorting and analyzing the meteorological data variables according to the size of the Copula entropy, outdoor temperature and light intensity are determined to be the key influencing factors.

[0053] Combining the historical secondary side heating characteristics and date characteristics, a time series dataset of secondary side heating of thermal power stations is constructed, including:

[0054] The outdoor average temperature in the 24 hours before the day to be predicted, the maximum temperature on the day to be predicted, the minimum temperature on the day to be predicted and the maximum temperature difference are selected as outdoor temperature features;

[0055] The average light intensity in the 24 hours before the day to be predicted, the maximum light intensity on the day to be predicted, the minimum light intensity on the day to be predicted and the maximum light intensity difference are selected as light intensity features;

[0056] Based on the selected outdoor temperature characteristics and light intensity characteristics, combined with the historical secondary side heating characteristics and date characteristics of the week before the forecast date, a secondary side heating time series dataset of the thermal power station is constructed;

[0057] The correlation between meteorological data and the secondary water supply temperature of the thermal power station is calculated, the meteorological characteristics that are strongly correlated with the secondary water supply temperature of the thermal power station are determined, and the secondary heating time series data set of the thermal power station is constructed by combining the historical secondary heating characteristics and date characteristics. The secondary heating of the thermal power station is affected by meteorological and holiday factors, and after reducing the input features of the model, selecting appropriate features can increase the prediction accuracy of the subsequent secondary heating time series model of the thermal power station, and reduce the time complexity and space complexity of the model.

[0058] It should be noted that in step S2, it also includes: identifying the outdoor temperature pattern according to the outdoor temperature characteristics, and dividing it into an outdoor low temperature mode, an outdoor medium temperature mode and an outdoor high temperature mode; identifying the light intensity pattern according to the light intensity characteristics, and dividing it into a low light intensity mode, a medium light intensity mode and a high light intensity mode; dividing the meteorological characteristics into different temperature-light pattern categories according to the outdoor temperature pattern and the light intensity pattern, and constructing a secondary heating time series data subset of the thermal power station under different temperature-light pattern categories in combination with the historical secondary heating characteristics and the date characteristics; and, in step S3, it also includes: using the secondary heating time series data subset of the thermal power station under different temperature-light pattern categories, establishing a graph neural network model under different temperature-light pattern categories, and obtaining the secondary heating prediction results of the thermal power station under different temperature-light pattern categories; and, in step S4, it also includes: using the secondary heating time series data subset of the thermal power station under different temperature-light pattern categories to train and learn the corresponding graph neural network model, and then establishing a secondary heating time series prediction sub-model for the thermal power station under different temperature-light pattern categories.

[0059] S3: Establish a graph neural network model including a graph learning module, a first graph convolution module, a second graph convolution module, a time series convolution module and an output module; extract the dynamic spatial relationship between each feature and the secondary side heating of the thermal power station through the graph learning module, output the graph adjacency matrix, and use it as the input of the first graph convolution module; extract the spatial dependency relationship between multivariate variables in the secondary side heating time series data of the thermal power station through the first graph convolution module; perform deep information extraction through the second graph convolution module, and input the extracted time-varying information at different times into the time series; extract the time dependency relationship of the time series data with spatial features through the time series convolution module, specifically, extract the time dependency relationship through the TCN time series convolution neural network and the GRU gated recurrent unit network in the time series convolution module, and perform time dependency fusion; obtain the prediction result of the secondary side heating of the thermal power station by splicing the information of multiple modules through the output module;

[0060] The first graph convolution module, the second graph convolution module and the temporal convolution module are used as the first spatiotemporal convolution module; the graph neural network model includes multiple spatiotemporal convolution modules; a residual connection is added to the input end of the second graph convolution module and the output end of the temporal convolution module; the graph neural network model also includes a jump connection module for making the sequence length from input to output the same;

[0061] The dynamic spatial relationship between each feature and the secondary side heating of the thermal power station is extracted through the graph learning module, and the graph adjacency matrix is ​​output, which includes:

[0062] Heating data diagram for secondary side of thermal power station To describe the topological structure of multiple variables in secondary side heating; is a set of N variable nodes; is the set of edges that represent connectivity between nodes; It is a picture The adjacency matrix represents the connection weights between variable nodes;

[0063] The graph learning module uses an asymmetric adjacency matrix, expressed as:

[0064] ;

[0065] ;

[0066] ;

[0067] , To initialize the node; , are model parameters; is the saturation rate of the activation function.

[0068] It should be noted that the graph learning module can automatically extract the one-way relationship between variables and integrate the time series of other variables into the time series of secondary heating of the thermal power station. This processing method is based on the principle of Graph Neural Network (GNN), which uses graph structure data to extract the relationship between variables.

[0069] In a graph neural network, each node represents a variable and the edge represents the relationship between variables. By applying a series of graph convolution operations on the graph, the graph neural network can capture the complex relationships between nodes and learn the intrinsic structure and patterns of the data. In the secondary heating problem of the thermal power station, the graph learning module can use the time series of other variables as nodes of the graph and use the graph neural network to extract the unidirectional relationship between them and the secondary heating of the thermal power station.

[0070] In this way, the graph learning module can integrate the time series of other variables into the time series of secondary heating of the thermal power station. This helps to comprehensively consider the impact of multiple factors on the secondary heating of the thermal power station and further optimize the performance of the heating system. At the same time, this method also helps to discover potential variable relationships and patterns, providing important information for subsequent predictions and decisions.

[0071] Also, in the traditional adjacency matrix, the connection relationship between each node and other nodes is treated equally, resulting in storage and computational redundancy. The asymmetric adjacency matrix only focuses on directed edges or specific types of connection relationships, which greatly reduces the required storage space and computational complexity. Through the asymmetric adjacency matrix, the graph learning layer can process graph data more efficiently, focusing only on actual or meaningful connection relationships. This strategy can significantly reduce storage and computational costs, especially when processing large-scale graph data, and can significantly improve processing speed and resource utilization.

[0072] In addition, the asymmetric adjacency matrix also helps to better represent the intrinsic structure and pattern of graph data. By focusing only on specific connection relationships, the graph learning layer can more accurately capture the interactions and information flows between nodes, thereby better understanding the intrinsic laws and characteristics of graph data. In short, the use of an asymmetric adjacency matrix in the graph learning layer can significantly reduce computational costs and improve the efficiency of processing large-scale graph data.

[0073] The first graph convolution module processes the spatial dependency of features by fusing the information of a node with its neighbor information, which is expressed as:

[0074] ;

[0075] For the The node feature vector of the layer is composed of the node feature vector of the previous layer and the adjacency matrix Calculated; is the activation function; It is The weight matrix of the layer; , is a unit matrix of order N; is a diagonal matrix.

[0076] The second graph convolution module is used to extract deep information and input the extracted time-varying information at different moments into the time series, including:

[0077] In the graph convolutional network (GCN), each node performs convolution operations with neighboring nodes to fuse the information of neighboring nodes into the feature representation of the current node. Through multiple convolution operations, the information of the entire graph can be fused into a global feature representation;

[0078] In addition to using the structural information of the edges and vertices of the graph, GCN can also use the attribute information attached to the graph structure, including node features and edge weights, to provide GCN with additional contextual information and determine the structure and relationship of the graph;

[0079] The function of the first graph convolution module is to extract the spatial dependency between multivariate variables in the time series data of the secondary heating of the thermal power station. This module will use the ability of the graph convolution network to process graph structure data, taking the secondary heating of the thermal power station as a node and the relationship between multivariate variables as an edge, and extracting the feature representation of the node through convolution operation. This feature representation can capture the spatial dependency between multivariate variables and provide important information for subsequent time series analysis.

[0080] The second graph convolution module is used to extract deep information. This module also uses graph convolutional networks to perform deeper processing and analysis on the spatial dependencies extracted by the first module. Through deeper convolution operations, the second module can extract more complex patterns and relationships, further enhancing the understanding of the interactions between multivariate variables.

[0081] The TCN temporal convolutional neural network applies convolution operations on one-dimensional time series, and captures the dependencies in the time series by moving the convolution kernel in the time dimension; each hidden layer of the TCN temporal convolutional neural network contains a set of convolutional layers, each convolutional layer consists of a one-dimensional convolution operation and a nonlinear activation function, and the output of the one-dimensional convolution operation is:

[0082] ;

[0083] For the Layer Outputs; For the Layer The weight of the convolution kernel; is the length of the convolution kernel; is a nonlinear activation function; is the bias term; For the Layer Outputs;

[0084] The temporal dependency extracted by the gated recurrent unit network GRU is expressed as:

[0085] ;

[0086] is the hidden layer state at the previous moment t-1; is the secondary heating monitoring value of the thermal power station at the current time t; is an update gate used to control the amount of state information before entering the current state; To control the reset gate that forgets the state information of the previous moment, if the output is 0, the secondary heating sequence information of the heating station at the previous moment will be ignored; if the output is 1, the secondary heating sequence information of the heating station at the previous moment will be completely memorized and brought into the next moment; is the candidate hidden state at the current time t; is the output state at the current time t; , They are the weight matrix and bias vector of the update gate respectively; , are the weight matrix and bias vector of the reset gate respectively; , are the weight matrix and bias vector of the candidate hidden state respectively; is the activation function.

[0087] In step S3, performing time dependency fusion includes:

[0088] The attention mechanism is introduced to allocate the attention of the temporal dependency between the two, and they are fused through the fully connected layer to output the final temporal dependency result.

[0089] Considering the powerful ability of graph neural networks in capturing data relationships, graph neural networks are applied to the secondary side heating prediction of thermal power stations to obtain dynamic dependencies of variables. The graph learning module is used to adaptively obtain the dynamic adjacency matrix to better obtain the impact of other factors on the numerical changes of the secondary side heating of the thermal power station. In the calculation of complex spatial dependencies and dynamic time change sequences, the graph convolution module and the time series convolution module are used to capture spatial correlation and time correlation to realize the spatiotemporal prediction of graph neural networks. In particular, the TCN time series convolution neural network and the GRU gated recurrent unit network in the time series convolution module are used to extract time dependencies respectively, and the time dependencies are fused, which improves the accuracy of time dependency calculation and ensures the effectiveness and accuracy of graph neural networks in predicting the secondary side heating of thermal power stations.

[0090] S4: After training and learning the graph neural network model, a secondary side heating time series prediction model for the thermal power station is established;

[0091] In specific practical applications, the secondary heating time series data set of the thermal power station is divided into a training set and a test set. The data in the training set and the test set are used to train and learn the graph neural network model, and a secondary heating time series prediction model for the thermal power station based on the graph neural network is established.

[0092] The improved vulture search algorithm is used to optimize the weight parameters, bias parameters, activation function parameters and hyperparameters of the graph neural network. The vulture search algorithm simulates the hunting behavior of vultures, including the stages of selecting search space, searching for prey and diving to capture. It is simple to implement, has a wide application space and good performance. It can be used to optimize the parameters of the graph neural network model and improve the performance and generalization ability of the model by continuously iteratively updating the parameters.

[0093] Improvements to the vulture search algorithm, including:

[0094] Use elite reverse learning to initialize the population: randomly generate The initial position of each individual ,in ; , is the population size, For each population dimension; generate elite reverse population , is the elite reverse coefficient, For the elite group The minimum boundary on dimension, For the elite group The maximum boundary on dimension, if If the solution exceeds the boundary, that is, it is not within the feasible solution range, then a random value within the dynamic boundary is taken; the current population and the elite reverse population are merged, their fitness values ​​are sorted, and the selection is The individuals with better fitness are used as the new initial population.

[0095] The Levy flight strategy is introduced to optimize the selection search space, which can be expressed as:

[0096] ;

[0097] ;

[0098] For the A new location chosen by a vulture; , To control the position variation parameters and random parameters; The search space selected for the best location currently identified by the condor; is the average distribution position of vultures after the previous search was completed; For the The location of the vulture; is the Levy distribution function; is the step length; To obey Normal distribution of is the number of iterations.

[0099] A method for establishing a secondary side heating time series model of a heating station in a centralized heating system. When working, the method obtains historical secondary side heating characteristics, meteorological characteristics and date characteristics that affect the secondary side water supply temperature of the heating station in the heating system; wherein the meteorological characteristics include outdoor temperature, light intensity, wind speed and humidity; the date characteristics include working days, holidays, months and weeks; the correlation between meteorological data and the secondary side water supply temperature of the heating station is calculated, the meteorological characteristics that are strongly correlated with the secondary side water supply temperature of the heating station are determined, and the secondary side heating time series data set of the heating station is constructed by combining the historical secondary side heating characteristics and date characteristics; a graph neural network model including a graph learning module, a first graph convolution module, a second graph convolution module, a time series convolution module and an output module is established.

[0100] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

[0101] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0102] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for establishing a secondary side heating time series model for a heating station in a centralized heating system, characterized in that: The following steps are involved: Step S1: Acquire historical secondary side heating characteristics, meteorological characteristics and date characteristics that affect the secondary side water supply temperature of the heating power station of the heating system; wherein the meteorological characteristics include outdoor temperature, light intensity, wind speed and humidity; the date characteristics include working days, holidays, months and weeks; Step S2: Calculate the correlation between meteorological data and the secondary side water supply temperature of the thermal power station, determine the meteorological features that are strongly correlated with the secondary side water supply temperature of the thermal power station, and construct a secondary side heating time series data set of the thermal power station by combining the average, maximum, minimum and maximum difference of the meteorological features with the historical secondary side heating features and date features; Step S3: Establish a graph neural network model including a graph learning module, a first graph convolution module, a second graph convolution module, a time series convolution module and an output module; extract the dynamic spatial relationship between each feature and the secondary side heating of the thermal power station through the graph learning module, output the graph adjacency matrix, and use it as the input of the first graph convolution module; extract the spatial dependency relationship between multivariate variables in the secondary side heating time series data of the thermal power station through the first graph convolution module; perform deep information extraction through the second graph convolution module, and input the extracted time-varying information at different times into the time series; extract the time dependency relationship of the time series data with spatial features through the time series convolution module, specifically, extract the time dependency relationship through the TCN time series convolution neural network and the GRU gated recurrent unit network in the time series convolution module, and perform time dependency fusion; obtain the prediction result of the secondary side heating of the thermal power station by splicing the information of multiple modules through the output module; Step S4: After training and learning the graph neural network model, a secondary side heating timing prediction model for the thermal power station is established.

2. The method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 1, characterized in that: In step S2, the correlation between meteorological data and the secondary side water supply temperature of the thermal power station is calculated, and meteorological features strongly correlated with the secondary side water supply temperature of the thermal power station are determined, including: Based on the Copula entropy correlation analysis method, the correlation between meteorological data and the secondary water supply temperature of the thermal power station was calculated. After sorting and analyzing the meteorological data variables according to the size of the Copula entropy, the outdoor temperature and light intensity were determined to be the key influencing factors.

3. The method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 2, characterized in that: In step S2, the secondary side heating time series data set of the thermal power station is constructed by combining the historical secondary side heating characteristics and date characteristics, which specifically includes: The outdoor average temperature in the 24 hours before the day to be predicted, the maximum temperature on the day to be predicted, the minimum temperature on the day to be predicted and the maximum temperature difference are selected as outdoor temperature features; The average light intensity in the 24 hours before the day to be predicted, the maximum light intensity on the day to be predicted, the minimum light intensity on the day to be predicted and the maximum light intensity difference are selected as light intensity features; Based on the selected outdoor temperature characteristics and light intensity characteristics, combined with the historical secondary side heating characteristics and date characteristics one week before the predicted day, a secondary side heating time series dataset of the thermal power station is constructed.

4. The method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 3 is characterized in that: In step S2, it also includes: identifying the outdoor temperature pattern according to the outdoor temperature characteristics, and dividing it into an outdoor low temperature mode, an outdoor medium temperature mode and an outdoor high temperature mode; identifying the light intensity pattern according to the light intensity characteristics, and dividing it into a low light intensity mode, a medium light intensity mode and a high light intensity mode; dividing the meteorological characteristics into different temperature-light pattern categories according to the outdoor temperature pattern and the light intensity pattern, and constructing a thermal power station secondary heating time series data subset under different temperature-light pattern categories in combination with the historical secondary heating characteristics and the date characteristics; and, in step S3, it also includes: using the thermal power station secondary heating time series data subset under different temperature-light pattern categories, establishing a graph neural network model under different temperature-light pattern categories, and obtaining the thermal power station secondary heating prediction results under different temperature-light pattern categories; and, in step S4, it also includes: using the thermal power station secondary heating time series data subset under different temperature-light pattern categories to train and learn the corresponding graph neural network model, and then establishing a thermal power station secondary heating time series prediction sub-model under different temperature-light pattern categories.

5. The method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 4, characterized in that: In step S3, the first graph convolution module, the second graph convolution module and the temporal convolution module are used as the first spatiotemporal convolution module; the graph neural network model includes multiple spatiotemporal convolution modules; a residual connection is added to the input end of the second graph convolution module and the output end of the temporal convolution module; the graph neural network model also includes a jump connection module for making the sequence length from input to output the same.

6. A method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 5, characterized in that: In step S3, the dynamic spatial relationship of the data is extracted through the graph learning module, and the graph adjacency matrix is ​​output, which specifically includes: Heating data diagram for secondary side of thermal power station To describe the topological structure of multiple variables in secondary side heating; is a set of N variable nodes; is the set of edges that represent connectivity between nodes; It is a picture The adjacency matrix represents the connection weights between variable nodes; The graph learning module adopts an asymmetric adjacency matrix, which is expressed as: ; ； ； , To initialize the node; , are model parameters; is the saturation rate of the activation function.

7. A method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 6, characterized in that: In step S3, the spatial dependency of features is processed by fusing the information of a node with the information of its neighbors through the first graph convolution module, which is expressed as: ; For the The node feature vector of the layer is composed of the node feature vector of the previous layer and the adjacency matrix Calculated; is the activation function; It is The weight matrix of the layer; , is a unit matrix of order N; is a diagonal matrix.

8. The method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 7, characterized in that: In step S3, deep information extraction is performed through the second graph convolution module, including: In the graph convolutional network (GCN), each node performs a convolution operation with its neighboring nodes to fuse the information of the neighboring nodes into the feature representation of the current node. Through multiple convolution operations, the information of the entire graph can be fused into a global feature representation. In addition to using the structural information of the graph's edges and vertex connections, GCN can also use attribute information attached to the graph structure, including node features and edge weights, to provide GCN with additional contextual information and determine the structure and relationships of the graph.

9. A method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 8, characterized in that: The TCN temporal convolutional neural network applies convolution operations on one-dimensional time series, and captures the dependencies in the time series by moving the convolution kernel in the time dimension; each hidden layer of the TCN temporal convolutional neural network contains a set of convolutional layers, each convolutional layer consists of a one-dimensional convolution operation and a nonlinear activation function, and the output of the one-dimensional convolution operation is: ; For the Layer Outputs; For the Layer The weight of the convolution kernel; is the length of the convolution kernel; is a nonlinear activation function; is the bias term; For the Layer Outputs; The temporal dependency extracted by the gated recurrent unit network GRU is expressed as: ; is the hidden layer state at the previous moment t-1; is the secondary heating monitoring value of the thermal power station at the current time t; is an update gate used to control the amount of state information before entering the current state; To control the reset gate that forgets the state information of the previous moment, if the output is 0, the secondary heating sequence information of the heating station at the previous moment will be ignored; if the output is 1, the secondary heating sequence information of the heating station at the previous moment will be completely memorized and brought into the next moment; is the candidate hidden state at the current time t; is the output state at the current time t; , They are the weight matrix and bias vector of the update gate respectively; , are the weight matrix and bias vector of the reset gate respectively; , are the weight matrix and bias vector of the candidate hidden state respectively; is the activation function.

10. A method for establishing a secondary side heating time series model of a heating station in a centralized heating system according to claim 9, characterized in that: In step S3, performing time dependency fusion includes: An attention mechanism is introduced to allocate attention to the temporal dependency between the two, and they are fused through a fully connected layer to output the final temporal dependency result.

Images