A method, system and storage medium for predicting energy consumption of a single chiller

Abstract

The present invention relates to the technical field of energy consumption prediction for central air-conditioning chillers, and discloses a method, system and storage medium for predicting energy consumption of a single chiller host. The method comprises the following steps: obtaining an operating data set of a building refrigeration station management system to be tested; constructing a global correlation adjacency matrix of features according to the correlation coefficient of the operating data set; dividing the operating data set into a plurality of subsets, and constructing a local correlation adjacency matrix of features according to the correlation coefficient of each subset; constructing a target dynamic correlation adjacency matrix based on the global correlation adjacency matrix and the local correlation adjacency matrix; performing iterative training based on the target dynamic correlation adjacency matrix to obtain a target graph neural network model; inputting the test data obtained in real time into the target graph neural network model to obtain predicted energy consumption, and dynamically representing the correlation between the data in the operating data set of the building refrigeration station management system according to time changes, so as to predict the energy consumption of the single chiller host in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of energy consumption prediction for central air-conditioning chillers, and in particular to a method, system and storage medium for predicting energy consumption of a single chiller. Background Art

[0002] As a crucial component of building energy consumption, HVAC systems require monitoring and forecasting, a crucial prerequisite for further energy conservation. Currently, in large-scale air conditioning systems, the chiller, the core cooling device, possesses a complex system structure and involves numerous physical processes during operation. Therefore, traditional methods for reasonably forecasting air conditioning system energy consumption require considerable theoretical expertise.

[0003] With the increasing computing power of computing devices, machine learning has been widely applied in the industrial sector. Large-scale chiller plants are also equipped with data collection devices, which facilitate data-driven modeling of chiller energy consumption. Since chillers often operate in parallel, there is a strong coupling between them. The interactions between them will have a certain impact on the energy consumption of a single chiller. However, research on chiller energy consumption generally focuses on a single chiller or the entire chiller plant. The interactions between multiple chillers are difficult to explicitly represent, and traditional algorithms also struggle to intuitively characterize the interactions between input variables. Graph neural networks provide an important approach to addressing the degree of correlation between different input variables. This correlation can, to a certain extent, reflect the relationships and interactions between variables. However, in traditional graph neural networks, there is no universal method for constructing correlation graphs composed of nodes and edges. Furthermore, due to the complexity and diversity of the external environment, the operating conditions of chillers may vary at different times, and the correlation between the variables will also change with fluctuations in operating conditions. The time-varying nature of the correlation degree places higher demands on the real-time performance of the characterization method.

[0004] Therefore, how to dynamically characterize the correlation between various collected data according to time changes and make real-time predictions based on this correlation and the currently collected data has become an urgent problem to be solved. Summary of the Invention

[0005] The present invention provides a method, system and storage medium for predicting energy consumption of a single chiller to solve the problems existing in the prior art.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0007] In a first aspect, an embodiment of the present application provides a method for predicting energy consumption of a single chiller, comprising the following steps:

[0008] Obtain the operating data set of the building refrigeration station management system to be tested;

[0009] constructing a global correlation adjacency matrix of features based on the correlation coefficients of the running data set;

[0010] Dividing the running data set into a plurality of subsets, and constructing a local correlation adjacency matrix of features according to the correlation coefficient of each subset;

[0011] Constructing a target dynamic relevance adjacency matrix based on the global relevance adjacency matrix and the local relevance adjacency matrix;

[0012] Iterative training is performed based on the target dynamic association adjacency matrix to obtain a target graph neural network model;

[0013] The data to be measured obtained in real time is input into the target graph neural network model to obtain the predicted energy consumption output by the target graph neural network model.

[0014] Optionally, constructing a global correlation adjacency matrix of features according to the correlation coefficient of the running data set includes:

[0015] Determining characteristic variables of the running data set, and calculating correlation coefficients between the characteristic variables;

[0016] Constructing a global correlation adjacency matrix of features according to the correlation coefficients, wherein the absolute value of the correlation coefficient is used to represent the degree of global correlation between feature variables, and the global correlation adjacency matrix includes the degree of global correlation;

[0017] The global relevance adjacency matrix is ​​a mathematical representation of the global relationship graph.

[0018] Optionally, dividing the operating data set into several subsets includes:

[0019] Continuous sliding window processing is performed on the running data set to obtain a plurality of continuous subsets.

[0020] Optionally, constructing a local correlation adjacency matrix of features according to the correlation coefficient of each subset includes:

[0021] Determine the characteristic variables in each subset, and calculate the correlation coefficients between the characteristic variables of each subset based on the data in each subset;

[0022] Constructing a local correlation adjacency matrix of features according to the correlation coefficients between the feature variables of each subset, wherein the absolute values ​​of the correlation coefficients between the feature variables of each subset are used to represent the degree of local correlation between the feature variables, and the local correlation adjacency matrix includes the degree of local correlation;

[0023] The local relevance adjacency matrix is ​​a mathematical representation of the local relationship graph.

[0024] Optionally, the target dynamic association adjacency matrix satisfies the following relationship:

[0025] A dt =A s [α+(1-α)A vt ], α∈[0,1]

[0026] Among them, A dt A represents the target dynamic association adjacency matrix at time t, s represents the global correlation adjacency matrix, A vt represents the local correlation adjacency matrix at time t, and α represents the stability coefficient of the formula.

[0027] Optionally, the method further comprises: optimizing the stability coefficient in a preset manner;

[0028] The preset method includes at least one of the following:

[0029] Non-gradient descent or gradient descent.

[0030] In the second aspect, an embodiment of the present application also provides a single-machine energy consumption prediction system for a chiller, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method described in the first aspect are implemented.

[0031] In a third aspect, an embodiment of the present application further provides a computer storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0032] The present invention has the following beneficial effects:

[0033] The present invention provides a method for predicting the energy consumption of a single chiller. The method constructs a global correlation adjacency matrix and a local correlation adjacency matrix through an operating data set of a building refrigeration station management system, and constructs a target dynamic correlation adjacency matrix based on the global correlation adjacency matrix and the local correlation adjacency matrix. Iterative training is performed based on the target dynamic correlation adjacency matrix to obtain a target graph neural network model. The test data obtained in real time is input into the target graph neural network model to obtain the predicted energy consumption output by the target graph neural network model. In this way, the correlation between the data in the operating data set of the building refrigeration station management system can be dynamically characterized according to time changes, and the energy consumption of the single chiller can be predicted in real time based on this correlation and the currently collected data. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is one of the flow charts of a method for predicting energy consumption of a single chiller in a preferred embodiment of the present invention;

[0035] Figure 2 This is the second flow chart of a method for predicting energy consumption of a single chiller according to a preferred embodiment of the present invention;

[0036] Figure 3 A schematic diagram of the form of an association matrix provided for a preferred embodiment of the present invention;

[0037] Figure 4 A comparison chart of the prediction results of the energy consumption prediction method provided by the preferred embodiment of the present invention and the existing prediction methods. DETAILED DESCRIPTION

[0038] The following is a clear and complete description of the technical solutions of the present invention. It should be understood that the embodiments described are only a portion of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are intended to fall within the scope of protection of the present invention.

[0039] See Figure 1-Figure 2 The present invention provides a method for predicting energy consumption of a single chiller, comprising the following steps:

[0040] Step 101: Obtain an operating data set of a refrigeration station management system of a building to be tested.

[0041] In this step, the operating data set of the building refrigeration station management system includes, but is not limited to, key data such as chiller operating data, water pump operating data, weather data, and time data. This is for illustrative purposes only and is not intended to be limiting.

[0042] Step 102: Construct a global correlation adjacency matrix of features based on the correlation coefficients of the running data set.

[0043] Step 103: Divide the running data set into several subsets, and construct a local correlation adjacency matrix of the features according to the correlation coefficient of each subset.

[0044] Step 104: Construct a target dynamic correlation adjacency matrix based on the global correlation adjacency matrix and the local correlation adjacency matrix.

[0045] Step 105: Perform iterative training based on the target dynamic association adjacency matrix to obtain the target graph neural network model.

[0046] In this step, the target graph neural network model can be a graph spatiotemporal neural network (GSTNN) formed by combining a graph convolutional network (GCN) and a convolutional neural network (CNN). During iterative training, the target dynamic association adjacency matrix and the collected operational dataset are input into the target GNN model for iterative training. Through repeated training, the gap between the model's fitted energy consumption values ​​and the actual energy consumption values ​​is continuously narrowed. Training is stopped and the model is saved until the required accuracy is achieved.

[0047] Step 106: Input the real-time acquired test data into the target graph neural network model to obtain the predicted energy consumption output by the target graph neural network model.

[0048] The above-mentioned method for predicting the energy consumption of a single chiller is to construct a global correlation adjacency matrix and a local correlation adjacency matrix through the operating data set of the building refrigeration station management system, and to construct a target dynamic correlation adjacency matrix based on the global correlation adjacency matrix and the local correlation adjacency matrix; iterative training is performed based on the target dynamic correlation adjacency matrix to obtain a target graph neural network model; the test data obtained in real time are input into the target graph neural network model to obtain the predicted energy consumption output by the target graph neural network model; in this way, the correlation between the data in the operating data set of the building refrigeration station management system can be dynamically characterized according to time changes, and the energy consumption of the single chiller can be predicted in real time based on this correlation and the currently collected data.

[0049] Optionally, constructing a global correlation adjacency matrix of features according to the correlation coefficient of the running data set includes:

[0050] Determining characteristic variables of the running data set, and calculating correlation coefficients between the characteristic variables;

[0051] Constructing a global correlation adjacency matrix of features according to the correlation coefficients, wherein the absolute value of the correlation coefficient is used to represent the degree of global correlation between feature variables, and the global correlation adjacency matrix includes the degree of global correlation;

[0052] The global relevance adjacency matrix is ​​a mathematical representation of the global relationship graph.

[0053] In this optional embodiment, the characteristic variable of the operating data set may be the host j condenser approach temperature (T cond_j ), host j evaporator approach temperature (T evap_j ), host j condensing pressure (P cond_j ), host j evaporation pressure (P evap_j ), host j power (Power j ), host j chilled water flow (Q j ), host cooling capacity (R j ), host j energy efficiency ratio (COP j), host j chilled water inlet temperature (T chi_j ), host j chilled water outlet temperature (T cho_j ), cooling water inlet temperature of main engine j (T ci_j ), cooling water outlet temperature of main engine j (T co_j ), outdoor wet bulb temperature (T wet ), the week or month date and hour of data collection (Time), the switch status of host j (IO j ), the number of hosts to be started (N). This is just an example and is not a limitation.

[0054] Among them, the global relationship graph includes nodes and edges. The nodes in the global relationship graph represent feature variables, and the edges represent the relationship between feature variables. The degree of correlation between feature variables is regarded as a measure of the distance between nodes. The higher the degree of correlation between two nodes, the smaller the distance is considered, and the higher the degree of fusion of the data between the two nodes when aggregating information.

[0055] All characteristic variables are calculated to obtain a global correlation adjacency matrix symmetrical along the diagonal line, which is formed by sequentially arranging and combining the correlation degrees between each other. The form of the correlation adjacency matrix is ​​as follows: Figure 3 As shown, the global correlation adjacency matrix represents the structure of the global relationship graph and reflects the correlation relationship between each feature variable.

[0056] Specifically, the global correlation adjacency matrix can satisfy the following relationship:

[0057]

[0058] Among them, A mn Represents the value of the mth row and nth column of the correlation adjacency matrix, |correlate(X m , X n )| is feature X m With feature X n The absolute value of the correlation coefficient. In particular, when m=n, the correlation degree of the feature itself is set to 0. σ represents the relationship threshold. When the absolute value of the correlation coefficient between features is less than σ, the correlation value A at the corresponding position is mn Set to 0.

[0059] In this optional implementation, by constructing a global relationship graph, the relationships between data can be fully explored, intuitively reflecting the degree of data association. This can provide sufficient data information for network training to ensure the accuracy of prediction results, and can also provide data information for the construction of the target dynamic association adjacency matrix.

[0060] Optionally, dividing the operating data set into several subsets includes:

[0061] Continuous sliding window processing is performed on the running data set to obtain a plurality of continuous subsets.

[0062] In this optional implementation, each subset of the data set contains the same number of features.

[0063] Optionally, constructing a local correlation adjacency matrix of features according to the correlation coefficient of each subset includes:

[0064] Determine the characteristic variables in each subset, and calculate the correlation coefficients between the characteristic variables of each subset based on the data in each subset;

[0065] Constructing a local correlation adjacency matrix of features according to the correlation coefficients between the feature variables of each subset, wherein the absolute values ​​of the correlation coefficients between the feature variables of each subset are used to represent the degree of local correlation between the feature variables, and the local correlation adjacency matrix includes the degree of local correlation;

[0066] The local relevance adjacency matrix is ​​a mathematical representation of the local relationship graph.

[0067] In this optional embodiment, the running data set is denoted as D, and each subset is denoted as D t , the characteristic variable is recorded as X i Methods used to analyze the relationship between each pair of feature variables in the subset include, but are not limited to, Spearman correlation analysis, Pearson correlation coefficient, cosine similarity, and the inverse of the Euclidean distance of the data. This is for illustrative purposes only and is not intended to be limiting.

[0068] The method of constructing the local relationship graph and the local correlation adjacency matrix is ​​the same as the method of constructing the global one, and will not be described here.

[0069] Optionally, the target dynamic association adjacency matrix satisfies the following relationship:

[0070] A dt =A s [α+(1-α)A vt ], α∈[0,1]

[0071] Among them, A dt A represents the target dynamic association adjacency matrix at time t, s represents the global correlation adjacency matrix, A vt represents the local correlation adjacency matrix at time t, and α represents the stability coefficient of the formula.

[0072] In this optional implementation, data associations are constructed using a target dynamic association adjacency matrix. This reflects the degree of data association between sensors, establishes spatial correlations between sensor data, and enhances the network's ability to extract real-world information. Relationship thresholds can be used to appropriately limit feature associations, eliminating unnecessary data associations and reducing computational complexity.

[0073] Optionally, the method further comprises: optimizing the stability coefficient in a preset manner;

[0074] The preset method includes at least one of the following:

[0075] Non-gradient descent or gradient descent.

[0076] In this optional implementation, by optimizing the stability coefficient, the proportional relationship between the global correlation degree and the local correlation degree can be changed, and the network model can be flexibly adjusted according to the computer processing capability, which can improve the adaptability of the network model.

[0077] In one example, when constructing a global correlation adjacency matrix or a local correlation adjacency matrix, the relationship threshold can be set to 0. During the network training process, the stability coefficient is used as a parameter to be trained. The target dynamic correlation adjacency matrix A dt and the corresponding subset D t The model is trained using the gradient descent method through the spatiotemporal graph neural network to narrow the gap between the model prediction value and the actual value. When a model with the required accuracy is obtained, the training is stopped and the model is saved.

[0078] The trained model is run on the system and storage medium, and the dynamic association adjacency matrix is ​​updated in real time based on historical data and the latest actual data. The energy consumption of the chiller is predicted and the data is saved to the designated storage medium for subsequent use.

[0079] The method in this embodiment is used to analyze the effect of the model trained using the spatiotemporal graph neural network on the test set, and the energy consumption prediction effect is compared using the more mature convolutional neural network and long short-term memory neural network. The results are as follows: Figure 4 As shown, it can be seen that the accuracy of energy consumption prediction when the host is running using the model of the present invention is higher than that of the other two models.

[0080] In summary, based on the spatiotemporal graph neural network model, a dynamic correlation adjacency matrix is ​​constructed using historical data and real-time data. The spatiotemporal graph neural network performs convolution operations on the data in the time domain and the spatial domain respectively. The introduction of the time domain convolution ensures that the model can extract high-order features of the time domain of the data, and the spatial domain convolution ensures the full utilization of the feature correlation information provided by the dynamic correlation adjacency matrix. When predicting the energy consumption of a single chiller host, the method in this embodiment can fully consider the impact of other parallel hosts and other variables on the energy consumption of the target host. In addition, the introduction of the dynamic correlation adjacency matrix makes the correlation information more real-time updateable, which can timely reflect the rich correlation information when multiple chillers in the cold station operate together under different working conditions.

[0081] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for predicting energy consumption of a single chiller, characterized in that: The following steps are involved: Obtain the operating data set of the building refrigeration station management system to be tested; constructing a global correlation adjacency matrix of features based on the correlation coefficients of the running data set; Dividing the running data set into a plurality of subsets, and constructing a local correlation adjacency matrix of features according to the correlation coefficient of each subset; Constructing a target dynamic relevance adjacency matrix based on the global relevance adjacency matrix and the local relevance adjacency matrix; Iterative training is performed based on the target dynamic association adjacency matrix to obtain a target graph neural network model; Inputting the measured data acquired in real time into the target graph neural network model to obtain the predicted energy consumption output by the target graph neural network model; Among them, the global correlation adjacency matrix satisfies the following relationship: Among them, A mn Represents the value of the mth row and nth column of the correlation adjacency matrix, |correlate(X m ,X n )| is feature X m With feature X n The absolute value of the correlation coefficient, when m=n, indicates that the correlation degree of the feature itself is set to 0, σ represents the relationship threshold. When the absolute value of the correlation coefficient between features is less than σ, the correlation value A at the corresponding position mn Set to 0; The operating data set is divided into several subsets, including: Continuous sliding window processing is performed on the running data set to obtain a plurality of continuous subsets.

2. The method for predicting energy consumption of a single chiller according to claim 1, characterized in that: The step of constructing a global correlation adjacency matrix of features according to the correlation coefficient of the running data set includes: Determining characteristic variables of the running data set, and calculating correlation coefficients between the characteristic variables; Constructing a global correlation adjacency matrix of features according to the correlation coefficients, wherein the absolute value of the correlation coefficient is used to represent the degree of global correlation between feature variables, and the global correlation adjacency matrix includes the degree of global correlation; The global relevance adjacency matrix is ​​a mathematical representation of the global relationship graph.

3. The method for predicting energy consumption of a single chiller according to claim 2, characterized in that: The method of constructing a local correlation adjacency matrix of features according to the correlation coefficient of each subset includes: Determine the characteristic variables in each subset, and calculate the correlation coefficients between the characteristic variables of each subset based on the data in each subset; Constructing a local correlation adjacency matrix of features according to the correlation coefficients between the feature variables of each subset, wherein the absolute values ​​of the correlation coefficients between the feature variables of each subset are used to represent the degree of local correlation between the feature variables, and the local correlation adjacency matrix includes the degree of local correlation; The local relevance adjacency matrix is ​​a mathematical representation of the local relationship graph.

4. The method for predicting energy consumption of a single chiller according to claim 3, characterized in that: The target dynamic association adjacency matrix satisfies the following relationship: A dt =A s [α+(1-α)A vt ],α∈[0,1] Among them, A dt A represents the target dynamic association adjacency matrix at time t, s represents the global correlation adjacency matrix, A vt represents the local correlation adjacency matrix at time t, and α represents the stability coefficient of the formula.

5. The method for predicting energy consumption of a single chiller according to claim 4, characterized in that: The method further includes: optimizing the stability coefficient in a preset manner; The preset method includes at least one of the following: Non-gradient descent or gradient descent.

6. A chiller unit energy consumption prediction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When a processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

7. A computer storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images