A cold source system cold quantity intelligent prediction method and system and a storage medium

Abstract

The application relates to the technical field of cold intelligent prediction, and discloses a cold-source-system cold-intelligent-prediction method and system and a storage medium, the method constructs a graph structure with feature variables as nodes and the correlation between the feature variables as edges, can preliminarily divide the features through experience information and data information, improves the utilization rate of data by using the correlation between the feature variables, and accurately predicts the cold in real time.

Description

Technical Field

[0001] This invention relates to the field of intelligent cold load prediction technology, and in particular to an intelligent cold load prediction method, system and storage medium for a cold source system. Background Technology

[0002] In large buildings, the cooling supply of the HVAC system is crucial for maintaining indoor comfort and ensuring normal building operation. As the source of cooling energy, the HVAC system accounts for the largest share of building energy consumption. Therefore, establishing a high-precision cooling capacity prediction model and accurately predicting cooling capacity is an important foundation for achieving optimized control and ultimately realizing energy conservation and emission reduction.

[0003] Numerous factors influence cooling capacity, specifically outdoor and indoor factors. These factors are interconnected and highly coupled, making mechanistic thermodynamic analysis difficult, costly, and requiring repeated analysis for different building structures. Therefore, establishing a method to efficiently extract the relationships between cooling capacity influencing factors and accurately predict cooling capacity is of significant practical importance for optimizing central air conditioning cooling source systems and achieving building energy conservation and carbon reduction goals. With the promotion of smart building systems, real-time building data is more readily available, and data-driven prediction methods are gaining popularity in the building industry. However, traditional machine learning methods have limited utilization of the relationships between influencing factors, the implicit information provided by these relationships being crucial for cooling capacity prediction. Furthermore, due to the differences in units and magnitudes of data from different influencing factors, simple correlation analysis methods struggle to reflect the differences in relationships between data of different properties. Thus, existing cooling capacity prediction methods have low data utilization and low prediction accuracy. Summary of the Invention

[0004] This invention provides a method, system, and storage medium for intelligent prediction of cooling capacity in a cold source system, in order to solve the problems of low data utilization and low prediction accuracy in existing cooling capacity prediction methods.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] In a first aspect, the present invention provides a method for intelligent prediction of cooling capacity in a cold source system, comprising:

[0007] Obtain the historical operational dataset of the cold source system to be analyzed, identify the factors affecting the cooling capacity in the operational dataset as feature variables, and determine the data unit of the feature variables;

[0008] Based on the correlation between the feature variables, construct a correlation graph between different feature variables;

[0009] Based on the data unit and the feature variable, construct a set of similar features, and based on the properties of the feature variable, divide the set of similar features into homogeneous feature subsets;

[0010] A homogeneous feature mask matrix is ​​constructed based on the homogeneous feature subset, and the association relationships between nodes in the association graph are redistributed based on the homogeneous feature mask matrix to obtain a masked association graph.

[0011] Based on the mask association graph, a target graph neural network model is determined. The test data of the cold source system to be analyzed, which is acquired in real time, is input into the target graph neural network model to obtain the predicted cooling capacity output by the target graph neural network model.

[0012] Optionally, the correlation graph includes nodes and edges, wherein nodes represent feature variables used in cooling quantity prediction, and edges represent correlation coefficients between feature variables.

[0013] Optionally, the association graph satisfies the adjacency matrix A of the graph structure. cor The mathematical expression for satisfies the following formula:

[0014] A CORij =COR(f i ,f j );

[0015] Where COR represents the calculation of the i-th feature variable f i and the j-th feature variable f j The correlation coefficient between them, A CORij Let A be the adjacency matrix. COR The element in the i-th row and j-th column, where the value is the i-th feature variable f. i and the j-th feature variable f j The correlation coefficient between them is a positive real number ranging from 0 to 1. The closer the correlation coefficient is to 1, the stronger the correlation coefficient is. i and f j The stronger the correlation between them.

[0016] Optionally, dividing the set of similar features into homogeneous feature subsets based on the properties of the feature variables includes:

[0017] Based on the data source of the feature variables and the set of similar features to which the feature variables belong, it is determined whether each pair of feature variables is homogeneous. If they are homogeneous, the set of similar features is divided into homogeneous feature subsets. The steps for determining whether each pair of feature variables is homogeneous are as follows:

[0018] The two feature variables belong to the same set of features, and;

[0019] The data for both feature variables were obtained from the same type of equipment, and;

[0020] The two characteristic variables represent the same physical change process, or the two characteristics represent the same operating status index of the same type of equipment.

[0021] Optionally, the homogeneous feature mask matrix A M Satisfy the following formula:

[0022]

[0023] Among them, A Mij For homogeneous feature mask matrix A M The element in the i-th row and j-th column, where the value is the masking weight between the i-th and j-th feature variables, is used when the i-th and j-th feature variables are in the same homogeneous feature subset. Mij Take strong weights; when the i-th feature variable and the j-th feature variable are not in the same homogeneous feature subset, A Mij Use weak weights;

[0024] Both the strong weight and the weak weight are positive real numbers between 0 and 1. The value of the strong weight is always greater than that of the weak weight. The strong weight and the weak weight are used to redistribute the association relationships between the nodes in the association graph.

[0025] Optionally, the redistribution of relationships between nodes in the relationship graph is achieved using the following formula:

[0026] A MBij =A CORij ·A Mij ;

[0027] Where, matrix A MB For the mathematical representation of the mask association graph, A MBij The element in the i-th row and j-th column of the mathematical representation of the masked association graph is the value of the redistributed weight between the i-th feature and the j-th feature.

[0028] Optionally, determining the target graph neural network model based on the mask association graph includes:

[0029] The mask association graph is input into the constructed graph spatiotemporal neural network framework for iterative training until a target graph neural network model that meets the preset convergence condition is obtained. The constructed graph spatiotemporal neural network framework includes a graph convolutional network and a time series network.

[0030] Secondly, this application provides an intelligent prediction system for cooling capacity of a cold source system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method described in the first aspect above.

[0031] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0032] Beneficial effects:

[0033] The intelligent prediction method for cooling capacity of a cold source system provided by this invention constructs a graph structure with feature variables as nodes and the relationships between feature variables as edges. It can perform preliminary segmentation of features through experience and data information, improve data utilization by utilizing the relationships between feature variables, and make real-time and accurate predictions of cooling capacity. Attached Figure Description

[0034] Figure 1 This is a flowchart of a preferred embodiment of the present invention for an intelligent prediction method of cooling capacity in a cold source system;

[0035] Figure 2 A comparison diagram of the prediction results of the intelligent prediction method for cooling capacity of a cold source system according to a preferred embodiment of the present invention and existing prediction methods;

[0036] Figure 3 The diagram illustrates the prediction and actual effects of the intelligent cold capacity prediction method for a cold source system according to a preferred embodiment of the present invention. Detailed Implementation

[0037] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0039] Please see Figure 1 This application provides a method for intelligent prediction of cooling capacity in a cold source system, comprising:

[0040] Obtain the historical operational dataset of the cold source system to be analyzed, identify the factors affecting the cooling capacity in the operational dataset as feature variables, and determine the data unit of the feature variables;

[0041] Construct a relationship diagram between different feature variables based on the relationships between them;

[0042] Construct a set of similar features based on data units and feature variables, and divide the set of similar features into homogeneous feature subsets based on the properties of the feature variables;

[0043] A homogeneous feature mask matrix is ​​constructed based on a homogeneous feature subset, and the association relationships between nodes in the association graph are redistributed based on the homogeneous feature mask matrix to obtain a masked association graph.

[0044] The target graph neural network model is determined based on the mask association graph. The test data of the cold source system to be analyzed, which is acquired in real time, is input into the target graph neural network model to obtain the predicted cooling capacity output by the target graph neural network model.

[0045] In this embodiment, the historical operational dataset of the cold source system to be analyzed can refer to the operational dataset of the building refrigeration station management system, including but not limited to cold source system operational data, outdoor meteorological data, and time data. Specifically, in one example, the key data collected includes: condenser approach temperature of multiple chillers, evaporator approach temperature of multiple chillers, condensing pressure of multiple chillers, evaporating pressure of multiple chillers, power of multiple chillers, chilled water flow rate of multiple chillers, cooling capacity of multiple chillers, energy efficiency ratio of multiple chillers, chilled water inlet temperature of multiple chillers, chilled water outlet temperature of multiple chillers, cooling water inlet temperature of multiple chillers, cooling water outlet temperature of multiple chillers, outdoor wet-bulb temperature, outdoor humidity, outdoor temperature, outdoor dew point temperature, frequency of multiple cooling towers, total cooling capacity, weekday, month, date, hour of data collection, frequency of multiple cooling pumps, and frequency of multiple chilled pumps.

[0046] In this way, by screening different data types, the influencing factors related to cooling capacity were identified. There are certain correlations among various types of information, which can complement each other and provide sufficient data information for network training, ensuring the accuracy of prediction results.

[0047] In this context, the feature variables within each set of the same type of feature set have the same data unit.

[0048] The aforementioned intelligent prediction method for cooling capacity of a cold source system constructs a graph structure with feature variables as nodes and the relationships between feature variables as edges. It can perform preliminary segmentation of features through experience and data information, improve data utilization by utilizing the relationships between feature variables, and make real-time and accurate predictions of cooling capacity.

[0049] Optionally, the correlation graph includes nodes and edges, wherein nodes represent feature variables used in cooling quantity prediction, and edges represent correlation coefficients between feature variables.

[0050] In this optional implementation, the correlation graph includes nodes and edges. Nodes represent the feature variables used in cooling capacity prediction, and edges represent the correlation coefficients between features. This allows for simultaneous mathematical representation of the data for different feature variables themselves, as well as the relationships between them, making full use of the collected data.

[0051] Optionally, the association graph satisfies the adjacency matrix A of the graph structure. cor The mathematical expression for satisfies the following formula:

[0052] A CORij =COR(f i ,f j );

[0053] Where COR represents the calculation of the i-th feature variable f i and the j-th feature variable f j The correlation coefficient between them, A CORij Let A be the adjacency matrix. COR The element in the i-th row and j-th column, where the value is the i-th feature variable f. i and the j-th feature variable f j The correlation coefficient between them is a positive real number ranging from 0 to 1. The closer the correlation coefficient is to 1, the stronger the correlation coefficient is. i and f j The stronger the correlation between them. It should be noted that the COR (Correlation Ratio) is calculated using methods including, but not limited to, Pearson correlation coefficient and Spearman rank correlation coefficient.

[0054] Optionally, dividing the set of similar features into homogeneous feature subsets based on the properties of the feature variables includes:

[0055] Based on the data source of the feature variables and the set of similar features to which the feature variables belong, it is determined whether each pair of feature variables is homogeneous. If they are homogeneous, the set of similar features is divided into homogeneous feature subsets. The steps for determining whether each pair of feature variables is homogeneous are as follows:

[0056] The two feature variables belong to the same set of features, and;

[0057] The data for both feature variables were obtained from the same type of equipment, and;

[0058] The two characteristic variables represent the same physical change process, or the two characteristics represent the same operating status index of the same type of equipment.

[0059] In this optional embodiment, the feature variables are divided into several homogeneous feature subsets according to whether the feature variables are homogeneous, and the feature variables contained in each set of the same type of feature subset are homogeneous.

[0060] In one example, the partitioning of homogeneous feature subsets can be represented as follows:

[0061]

[0062]

[0063] Optionally, the homogeneous feature mask matrix A M Satisfy the following formula:

[0064]

[0065] Among them, A MijFor homogeneous feature mask matrix A M The element in the i-th row and j-th column, where the value is the masking weight between the i-th and j-th feature variables, is used when the i-th and j-th feature variables are in the same homogeneous feature subset. Mij Take strong weights; when the i-th feature variable and the j-th feature variable are not in the same homogeneous feature subset, A Mij Use weak weights;

[0066] Both the strong weight and the weak weight are positive real numbers between 0 and 1. The value of the strong weight is always greater than that of the weak weight. The strong weight and the weak weight are used to redistribute the association relationships between the nodes in the association graph.

[0067] Specifically, for a homogeneous feature mask matrix, the strong weights and weak weights can be determined using non-gradient descent and gradient descent methods.

[0068] In this way, by redistributing the relationships between feature variables based on the homogeneity of the data, that is, the relationships between nodes in the relationship graph, we can avoid excessive association of information between non-homogeneous feature variables and prevent excessive convergence of data when performing graph convolution operations.

[0069] In one example, the redistribution of relationships between nodes in the relationship graph is achieved using the following formula:

[0070] A MBij =A CORij ·A Mij ;

[0071] Where, matrix A MB For the mathematical representation of the mask association graph, A MBij The element in the i-th row and j-th column of the mathematical representation of the masked association graph is the value of the redistributed weight between the i-th feature and the j-th feature.

[0072] Optionally, determining the target graph neural network model based on the mask association graph includes:

[0073] The mask association graph is input into the constructed graph spatiotemporal neural network framework for iterative training until a target graph neural network model that meets the preset convergence condition is obtained. The constructed graph spatiotemporal neural network framework includes a graph convolutional network and a time series network.

[0074] In this embodiment, the target graph neural network model can be a graph spatiotemporal neural network framework constructed by combining a graph convolutional network and a time series model. During iterative training, the mask association graph and the feature variable data are input into the constructed graph spatiotemporal neural network framework for training. Through iterative training of the network, the gap between the model's predicted cooling value and the actual cooling value is continuously narrowed. Training stops and the model is saved when a model with the required accuracy is obtained.

[0075] In this way, a spatiotemporal graph neural network is adopted. Through graph convolution operations, feature information can be aggregated according to the correlation between feature variables. Through time series models, the time dimension information of feature variables can be extracted. The fusion of multiple information can further improve the prediction accuracy and generalization ability of the model.

[0076] The trained graph neural network model is run on the system and storage medium. Real-time acquired test data is input into the target graph neural network model to obtain the predicted cooling capacity output by the target graph neural network model. Using the method in this embodiment, the cooling capacity prediction performance of the model trained using the graph neural network is analyzed on a test set. The mean absolute percentage error (MAE) is used as the standard for verifying effectiveness. The energy consumption prediction performance is compared using relatively mature support vector machines and convolutional neural networks. The results are shown below. Figure 2 It can be seen that the accuracy of cold load prediction using the model of this invention is higher than that of the traditional model compared. Figure 3 The prediction effect of this embodiment is shown, and it can be seen that the predicted cooling capacity and the actual cooling capacity using the model of this invention are generally close.

[0077] In summary, the present invention provides an intelligent prediction method, system, and storage medium for cooling capacity in a cold source system. It acquires key data related to cooling capacity from a building cold source management system and its operational dataset, constructing a graph structure with feature variables as nodes and the relationships between these feature variables as edges. Based on the relationships between the collected feature variable data, a relationship graph is constructed between different feature variables. The data types are initially categorized according to the data units of the collected feature variables, constructing a set of similar features. Based on the specific properties of the feature variables, the set of similar features is further divided into homogeneous feature subsets. A homogeneous feature mask matrix is ​​constructed based on these subsets. The relationships between nodes in the constructed relationship graph are redistributed based on the homogeneous feature mask matrix to obtain a masked relationship graph. By utilizing the relationships between feature variables, the data utilization rate is improved, enabling real-time and accurate prediction of cooling capacity. This invention filters different data types, identifies influencing factors related to cooling capacity, and demonstrates the correlation between various information types, allowing them to complement each other and providing sufficient data for network training, thus ensuring the accuracy of prediction results. The correlation between feature variables is redistributed based on data homogeneity, avoiding excessive correlation between non-homogeneous feature variables and preventing excessive convergence of data during graph convolution operations. Strong and weak weights can be flexibly selected based on computer processing power to find the most suitable weight values. A spatiotemporal graph neural network is employed, which aggregates feature information based on the correlation between feature variables through graph convolution operations. The time series model extracts the time dimension information of feature variables, and the fusion of multiple information types further improves the model's prediction accuracy and generalization ability.

[0078] This application also provides an intelligent cold source system cooling capacity prediction system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described method. This intelligent cold source system cooling capacity prediction system can implement various embodiments of the above-described intelligent cold source system cooling capacity prediction method and achieve the same beneficial effects; therefore, further details are omitted here.

[0079] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method steps described above. This readable storage medium can implement various embodiments of the above-described method and achieve the same beneficial effects; further details are omitted here.

[0080] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A cold source system cold quantity intelligent prediction method, characterized in that, include: Obtain the historical operational dataset of the cold source system to be analyzed, identify the factors affecting the cooling capacity in the operational dataset as feature variables, and determine the data unit of the feature variables; Based on the correlation between the feature variables, construct a correlation graph between different feature variables; Based on the data unit and the feature variable, construct a set of similar features, and based on the properties of the feature variable, divide the set of similar features into homogeneous feature subsets; A homogeneous feature mask matrix is ​​constructed based on the homogeneous feature subset, and the association relationships between nodes in the association graph are redistributed based on the homogeneous feature mask matrix to obtain a masked association graph. Based on the mask association graph, the target graph neural network model is determined. The test data of the cold source system to be analyzed, which is acquired in real time, is input into the target graph neural network model to obtain the predicted cooling capacity output by the target graph neural network model. Divide the aforementioned set of similar features into homogeneous feature subsets, including: Based on the data source of the feature variables and the set of similar features to which the feature variables belong, it is determined whether each pair of feature variables is homogeneous. If they are homogeneous, the set of similar features is divided into homogeneous feature subsets. The steps for determining whether each pair of feature variables is homogeneous are as follows: The two feature variables belong to the same set of features, and; The data for both feature variables were obtained from the same type of equipment, and; The two characteristic variables represent the same physical change process or the two characteristics represent the same operating status index of the same type of equipment. The homogeneity feature mask matrix satisfies the following equation: ； wherein, is a homogenous feature mask matrix is an element of the i-th row and j-th column of the homogenous feature mask matrix, the element taking a strong weight when the i-th feature variable and the j-th feature variable are in the same homogenous feature subset, a weak weight when the i-th feature variable and the j-th feature variable are not in the same homogenous feature subset. a weak weight when the i-th feature variable and the j-th feature variable are not in the same homogenous feature subset. Both the strong weight and the weak weight are positive real numbers between 0 and 1. The value of the strong weight is always greater than that of the weak weight. The strong weight and the weak weight are used to redistribute the association relationships between the nodes in the association graph.

2. The intelligent prediction method for cooling capacity of a cold source system according to claim 1, characterized in that, The correlation graph includes nodes and edges, where nodes represent the feature variables used in cooling quantity prediction, and edges represent the correlation coefficients between feature variables.

3. The intelligent prediction method for cooling capacity of a cold source system according to claim 1, characterized in that, The association graph satisfies the adjacency matrix of the graph structure. The mathematical expression for satisfies the following formula: ； in, This indicates the calculation of the i-th feature variable. and the j-th feature variable The correlation coefficient between them Adjacency matrix The element in the i-th row and j-th column, where the value is the i-th feature variable. and the j-th feature variable The correlation coefficient between them is a positive real number ranging from 0 to 1. The closer the correlation coefficient is to 1, the stronger the correlation. and The stronger the correlation between them.

4. The intelligent prediction method for cooling capacity of a cold source system according to claim 1, characterized in that, The redistribution of relationships between nodes in the relationship graph is achieved through the following formula: ； Among them, matrix For the mathematical representation of the mask association graph, The element in the i-th row and j-th column of the mathematical representation of the masked association graph is the value of the redistributed weight between the i-th feature and the j-th feature.

5. The intelligent prediction method for cooling capacity of a cold source system according to claim 1, characterized in that, The neural network model for determining the target graph based on the mask association graph includes: The mask association graph is input into the constructed graph spatiotemporal neural network framework for iterative training until a target graph neural network model that meets the preset convergence condition is obtained. The constructed graph spatiotemporal neural network framework includes a graph convolutional network and a time series network.

6. A cooling capacity intelligent prediction system for a cold source system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-5.

Images