A method for predicting the operating load of building cooling sources using artificial neural network algorithm

The backpropagation network and parameter fuzzy model are established through artificial neural network algorithms, which solves the complexity and inaccuracy of cold source load prediction in traditional methods, and achieves more accurate and flexible load prediction, optimizes energy management and reduces costs.

CN117744747BActive Publication Date: 2024-12-03GUANGZHOU HAOTE ENERGY SAVING & ENVIRONMENTAL PROTECTION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410099687.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-24
Publication Date
2024-12-03
Estimated Expiration
2044-01-24

AI Technical Summary

Technical Problem

The traditional method of building cold source operation load prediction has computational complexity and inaccuracy, and it is difficult to capture the influence of complex nonlinear relationships and multiple factors.

Method used

Using artificial neural network algorithm, a backpropagation network model and a fuzzy parameter model are established. By training the neural network model, the relationship between cold source load and various factors is learned, the number of hidden layer nodes is optimized, and the appropriate transfer function is selected, and the inverse is fused. Predicting to propagation networks and fuzzy models.

Benefits of technology

Achieve more accurate and adaptable cold source operating load forecasts, suitable for different types of building and cold source equipment configurations, helping to optimize energy management, reduce energy waste and improve energy utilization efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117744747B_ABST
    Figure CN117744747B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for predicting the operating load of a building cold source using an artificial neural network algorithm, and relates to the technical field of building energy management. The specific steps include: establishing a back propagation network model for each cold source device in the building to predict the operating load of the building cold source; establishing a parameter fuzzy model for each cold source device in the building to predict the operating load of the building cold source; establishing an artificial neural network fusion model for the building cold source, and predicting the operating load of the building cold source. The present invention establishes a back propagation network model for each cold source device in the building, and adopts a three-layer neural network; on the other hand, the present invention establishes a parameter fuzzy model for each cold source device in the building; finally, according to the applicable conditions of the two models, a clustering algorithm is used to select, and the model with the best prediction effect is used to obtain the operating load of the building cold source. The technical problems of complex calculation of the operating load of the building cold source and inaccurate calculation results in the prior art are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of building energy management, and in particular relates to a method for predicting the operating load of a building cold source by using an artificial neural network algorithm. Background Art

[0002] In recent years, with the widespread application of artificial intelligence technology in all aspects of life and the rapid development of data collection and processing technology in the construction field, sensors and Internet of Things technologies have enabled a large amount of building operation data to be efficiently collected and recorded.

[0003] On the other hand, traditional building cooling source operation load forecasting methods are usually based on statistical models or rule models, which may have some limitations. For example, statistical models have limited modeling capabilities for nonlinear and non-steady-state characteristics; rule models have difficulty capturing complex correlations. Since building cooling source loads are affected by many factors (such as weather, building characteristics, human activities, etc.), it becomes more difficult to accurately predict load demand.

[0004] Artificial neural network is a method based on machine learning, which has powerful pattern recognition and associative learning capabilities. By training the neural network model using historical data, it can learn the complex nonlinear relationship between the cold source load and various factors. The neural network can adaptively adjust the weights and parameters to improve the prediction accuracy. Therefore, there is great potential for using artificial neural network algorithms to predict the operating load of building cold sources. Summary of the invention

[0005] The purpose of the present invention is to provide a method for predicting the operating load of a building cooling source using an artificial neural network algorithm, so as to solve the technical problems in the prior art of complex calculation of the operating load of a building cooling source and inaccurate calculation results.

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

[0007] The method for predicting the operating load of a building cooling source using an artificial neural network algorithm includes the following specific steps:

[0008] Step S1: Establish a back propagation network model for each cold source equipment in the building;

[0009] Step S2: predicting the operating load of the building cooling source;

[0010] Step S3: Establishing a fuzzy model of parameters of each cold source equipment in the building;

[0011] Step S4: predicting the operating load of the building cooling source;

[0012] Step S5: Establish a building cold source artificial neural network fusion model.

[0013] Preferably, the specific steps of establishing the back propagation network model of each cold source equipment in the building include:

[0014] Step S101: Determine the input and output parameters of the back propagation network;

[0015] Step S102: Determine the back propagation network transfer function;

[0016] Step S103: Determine the back propagation network learning algorithm;

[0017] Step S104: Determine the number of hidden layer nodes of the back propagation network;

[0018] Furthermore, the back propagation network includes an input layer, a hidden layer and an output layer.

[0019] Furthermore, the output parameter of the back propagation network is the cold source operating load; the input parameters of the back propagation network include parameters of each cold source device.

[0020] Furthermore, the cold source equipment includes a chiller, a water pump and a cooling tower; the water pump includes a cold water pump and a frozen water pump; the input parameters of the back propagation network model of the chiller include: cooling capacity , Frozen water inlet temperature , Chilled water temperature difference , cold water inlet temperature Temperature difference between cold water ; Output parameters include chiller operating load ; The input parameters of the back propagation network model of the chilled water pump include: , Frozen water inlet temperature and freezing water temperature difference ; Output parameters include chilled water pump operating load ; Input parameters of the cold water pump back propagation network model include: condensate cooling capacity , cold water inlet temperature Temperature difference between cold water ; Output parameters include chilled water pump operating load ; The input parameters of the cooling tower back propagation network model include: condensation cooling capacity , cold water inlet temperature , cold water temperature difference and wet bulb ambient temperature ; Output parameters include cooling tower operating load .

[0021] Furthermore, the number of hidden layer nodes includes: the chiller has 7 hidden layer nodes, the frozen water pump has 9 hidden layer nodes, the cold water pump has 9 hidden layer nodes, and the cooling tower has 14 hidden layer nodes.

[0022] Preferably, the specific steps of predicting the operating load of the building cooling source in step S2 and step S4 include:

[0023] The cooling capacity , Frozen water inlet temperature , Chilled water temperature difference , cold water inlet temperature Temperature difference between cold water Input the chiller back propagation model and output the chiller operating load ;

[0024] The cooling capacity , Frozen water inlet temperature and freezing water temperature difference Input the back propagation model of the chilled water pump and output the operating load of the chilled water pump ;

[0025] Calculate the condensate cooling capacity and convert it into , cold water inlet temperature Temperature difference between cold water Input the cold water pump back propagation model and output the cold water pump operating load ;

[0026] The condensate cooling capacity calculation formula is: ;

[0027] Condensate cooling , cold water inlet temperature , cold water temperature difference and wet bulb ambient temperature Input cooling tower back propagation model, output cooling tower operating load ;

[0028] By formula Predict the operating load of building cooling sources.

[0029] Preferably, the specific steps of establishing the fuzzy model of the parameters of each cold source equipment in the building in step S3 include:

[0030] Mathematically modeling each cold source device, and establishing a mathematical model for output parameters according to different parameters and coefficients, including unknown coefficients; the output parameter is the operating load of the cold source device;

[0031] Fuzzy identification of unknown coefficients in mathematical models;

[0032] The chiller is modeled to analyze the effects of cooling capacity, chilled water inlet temperature and cold water inlet temperature on the operating efficiency of the chiller, which is expressed by a mathematical model as follows: ,in Indicates the operating efficiency of the chiller. Indicates the cooling capacity, Indicates the freezing water inlet temperature. Indicates the cold water inlet temperature. represents the unknown coefficient; the operating load of the chiller is: ;

[0033] The chilled water pump is modeled. The chilled water pump operating load is related to the chilled water pump flow, condensate cooling capacity, chilled water inlet temperature, and chilled water temperature difference, which can be expressed by the mathematical model as follows:

[0034] ;

[0035] ;

[0036] ;

[0037] in, is the operating load of the cooling water pump, is the cooling water pump flow rate, is the cold water temperature difference, is the condensate cooling capacity, is the water density, is the unknown coefficient;

[0038] The cooling tower is modeled. The operating load of the cooling tower is related to the cooling air volume, which can be expressed by the mathematical model as follows:

[0039] ,in, is the cooling tower operating load, is the cooling air volume, is the unknown coefficient.

[0040] Furthermore, the specific steps of the fuzzy identification method include: using the nonlinear solver Octave to perform fuzzy solution on the unknown coefficients of the chiller model using the dichotomy method; the Octave command is: ,in, is a set of unknown coefficients, bin indicates that the algorithm used is dichotomy, C indicates the residual, and N indicates the input data set of the model.

[0041] Preferably, the specific steps of establishing the building cold source artificial neural network fusion model in step S5 include:

[0042] Step S501: data preparation;

[0043] Step S502: cluster analysis;

[0044] Step S503: Model prediction and selection.

[0045] Furthermore, the specific steps of data preparation include: determining a set of input parameters as feature data; and obtaining training data of the back propagation network model, including input parameters and output parameters.

[0046] Furthermore, the steps of cluster analysis include: initializing cluster centers and selecting multiple data points as initial cluster centers; matching the number of data points with the number of features of feature data; for each data point in the training data, finding the cluster center closest to it and assigning the data point to this cluster, wherein the distance between the data point and the cluster center is calculated by Euclidean distance; based on the existing cluster assignments, recalculating the cluster center by taking the average of all feature data assigned to the cluster; finally, after obtaining the new cluster center, repeatedly assigning each data point in the training set to the side of the latest cluster center, and then recalculating the cluster center again, and repeating the above steps until the cluster center no longer changes significantly or a preset number of iterations is reached.

[0047] Furthermore, the specific steps of model prediction and selection include: the new input parameters are , for each cluster center ,in , calculated by the formula The Euclidean distance from the center is given by ,in, is the number of input parameters; after that, the proximity threshold is determined According to the above calculation, we get Minimum distance to all cluster centers ,if , then it is believed that Close to the cluster, that is, close to the training data, then Input the back propagation network model for prediction; if , then Input parameters to the fuzzy model for prediction.

[0048] Furthermore, the proximity threshold The calculation method includes: calculating the average distance from the center of all clusters to the center of other clusters, and taking the average value as the proximity threshold.

[0049] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0050] 1. The present invention can establish a highly accurate and adaptable back propagation network model by integrating the back propagation network model and the parameter fuzzy model, accurately defining the input and output parameters, selecting a suitable transfer function, determining an effective learning algorithm, and optimizing the number of hidden layer nodes; taking into account the different types and characteristics of cold source equipment, the model can adapt to different buildings and cold source equipment configurations, and the method can more accurately predict the operating load of building cold source equipment.

[0051] 2. The present invention can adapt to various types of buildings and different cold source equipment configurations. By independently modeling different equipment such as chillers, water pumps and cooling towers, and considering various input parameters such as cooling capacity and water temperature, it can flexibly adapt to the specific needs and conditions of various buildings.

[0052] 3. The present invention helps to more effectively manage and allocate energy resources through accurate load forecasting, especially in building cooling systems. It helps building managers optimize equipment operation, reduce energy waste, and reduce operating costs, while also helping to improve energy efficiency and reduce environmental impact. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] 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.

[0054] Figure 1 The flow chart of the method for predicting the operating load of a building cooling source using an artificial neural network algorithm of the present invention is shown;

[0055] Figure 2 The back propagation network model input and output parameter diagram of the building cooling source operation load prediction method using the artificial neural network algorithm of the present invention is shown. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not 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. Example 1

[0057] Figure 1A method for predicting the operating load of a building cooling source using an artificial neural network algorithm provided by an embodiment of the present invention is shown. This embodiment can be applied to data. This embodiment provides a method for predicting the operating load of a building cooling source using an artificial neural network algorithm, specifically comprising:

[0058] Step S1: Establish a back propagation network model for each cold source equipment in the building.

[0059] Step S2: Predict the operating load of the building cooling source.

[0060] Step S3: Establish a fuzzy model of the parameters of each cold source equipment in the building.

[0061] Step S4: predicting the operating load of the building cooling source.

[0062] Step S5: Establish a building cold source artificial neural network fusion model.

[0063] The beneficial effects of this embodiment are as follows: this embodiment combines the back propagation network model and the parameter fuzzy model to improve the prediction accuracy of the building cooling source operating load from different dimensions and angles; through accurate prediction, the operation of the building cooling source equipment can be effectively managed and optimized, energy consumption can be reduced, and the overall energy utilization efficiency can be improved; it can be applicable to buildings of different types and sizes, and can flexibly handle various cooling source equipment, with strong adaptability and universality. Example 2

[0064] Based on Example 1, this example provides specific steps for establishing a back propagation network model for each cold source device in a building. The specific steps include:

[0065] Step S101: Determine the input and output parameters of the back propagation network.

[0066] The output parameter of the back propagation network is the cold source operating load; the input parameters of the back propagation network include the parameters of each cold source equipment.

[0067] Among them, the cold source equipment refers to equipment used to generate and adjust cold capacity, which reduces the temperature of air or objects by absorbing heat; in an embodiment of the present invention, the cold source equipment includes a chiller, a water pump and a cooling tower; the water pump includes a cold water pump and a frozen water pump.

[0068] As a feasible implementation method, which is also convenient for parameter collection and measurement, refer to Figure 2 As shown, the input parameters of the chiller back propagation network model include: , Frozen water inlet temperature , Chilled water temperature difference , cold water inlet temperature Temperature difference between cold water ; Output parameters include chiller operating load .

[0069] The input parameters of the back propagation network model of the chilled water pump include: , Frozen water inlet temperature and freezing water temperature difference ; Output parameters include chilled water pump operating load .

[0070] The input parameters of the cold water pump back propagation network model include: condensate cooling capacity , cold water inlet temperature Temperature difference between cold water ; Output parameters include chilled water pump operating load .

[0071] The input parameters of the cooling tower back propagation network model include: condensate cooling capacity , cold water inlet temperature , cold water temperature difference and wet bulb ambient temperature ; Output parameters include cooling tower operating load .

[0072] Among them, the main purpose of the building cold source is to meet the building's demand for cooling. Cooling refers to the heat that needs to be removed by the cold source system to maintain the desired temperature and humidity in a specific area; the wet-bulb ambient temperature refers to the ambient temperature when the air reaches a saturated state under a certain atmospheric pressure, which has a greater impact on the cooling load; the chilled water inlet temperature refers to the temperature of the chilled water sent to the cold source equipment, which affects the heat exchange efficiency and operating load of the cold source system; the cold water inlet temperature refers to the temperature of the water sent to the cooling tower; the chilled water temperature difference refers to the temperature difference between the chilled water inlet and outlet, which reflects the heat absorption efficiency of the chilled water and is used to evaluate the system performance; the cold water temperature difference refers to the temperature difference between the cooling water inlet and outlet, which is used to measure the cooling effect and system energy efficiency.

[0073] Step S102: Determine the back propagation network transfer function.

[0074] In artificial neural networks, in particular, in the back propagation network of the embodiment of the present invention, the transfer function, i.e., the activation function, plays an important role, including: introducing nonlinearity. The transfer function in the neural network is usually nonlinear. The nonlinearity enables the network to learn and approximate complex functions and process data that are nonlinearly separable; helping gradient propagation. In the back propagation network, the derivative of the activation function determines the propagation of the error signal. The gradient is crucial for adjusting the weights in the network to reduce the difference between the predicted output and the actual output.

[0075] The embodiment of the present invention does not limit the selection of the transfer function, and the transfer function may be a log-sigmoid function, a tan-sigmoid function, and a purelin function.

[0076] Step S103: Determine the back propagation network learning algorithm.

[0077] The back propagation network learning algorithm is used to train multi-layer feedforward neural networks under supervised learning. Its functions include: weight update. The back propagation algorithm updates the weights in the neural network by calculating the gradient of the loss function (such as mean square error) relative to the network weights; error back propagation. The core of the algorithm is the concept of error back propagation. The algorithm first propagates the input data forward in the network, calculates the output error, and then backpropagates this error back to the network to update the weights layer by layer.

[0078] Network learning algorithms include traingd algorithm, traingda algorithm, traindm algorithm, trainlm algorithm and trainoss algorithm.

[0079] Step S104: Determine the number of hidden layer nodes of the back propagation network.

[0080] In the embodiment of the present invention, the back propagation network includes an input layer, a hidden layer and an output layer. When the sizes of the input layer and the output layer are determined, the prediction error of the back propagation network is analyzed through simulation calculation under different numbers of hidden layer nodes. When the number of hidden layer nodes increases from a small number, the connection between the input and the output is enhanced due to the increase in the number of hidden layer nodes, and the potential mapping relationship therein can be discovered, and the prediction error decreases; the number of nodes is increased again until a certain number, and the prediction error tends to be stable; if the number of nodes is increased again, due to the excessive number of hidden layer nodes, overfitting occurs, and the error data of the input parameters fluctuates, which may be detected by the hidden layer as a potential influence on the output parameters, and the prediction error increases instead.

[0081] As a preferred embodiment, the chiller has 7 hidden layer nodes, the frozen water pump has 9 hidden layer nodes, the cold water pump has 9 hidden layer nodes, and the cooling tower has 14 hidden layer nodes.

[0082] The above embodiments illustrate the specific structure of the back propagation network, the number of network layers and nodes, etc. After the back propagation network is constructed, it is necessary to train it through training data, give input parameters, and give output parameters, compare the input and output parameters with the output of the back propagation network, and adjust the neuron weights through the back propagation network learning algorithm so that after the given input, the output of the back propagation network is close to the output corresponding to the training data.

[0083] The beneficial effects of this embodiment are as follows: this embodiment can establish a highly accurate and adaptable back propagation network model by accurately defining input and output parameters, selecting appropriate transfer functions, determining effective learning algorithms, and optimizing the number of hidden layer nodes; it takes into account the different types and characteristics of cold source equipment, such as chillers, water pumps and cooling towers, and their specific operating parameters, so that the model can adapt to different buildings and cold source equipment configurations and provide customized prediction solutions; through more accurate load forecasting, the cold source equipment of the building can be managed more effectively and its operating status can be optimized, thereby reducing energy consumption and improving overall energy efficiency. Example 3

[0084] This embodiment provides a method for predicting the operating load of a building cooling source based on the first embodiment, which specifically includes:

[0085] By transferring the cold , Frozen water inlet temperature , Chilled water temperature difference , cold water inlet temperature Temperature difference between cold water Input the chiller back propagation model and output the chiller operating load .

[0086] By transferring the cold , Frozen water inlet temperature and freezing water temperature difference Input the back propagation model of the chilled water pump and output the operating load of the chilled water pump .

[0087] The condensate cooling capacity is calculated by the formula, and the condensate cooling capacity is , cold water inlet temperature Temperature difference between cold water Input the cold water pump back propagation model and output the cold water pump operating load .

[0088] The calculation formula of condensate cooling capacity is: .

[0089] By converting the condensate cooling , cold water inlet temperature , cold water temperature difference and wet bulb ambient temperature Input cooling tower back propagation model, output cooling tower operating load .

[0090] By formula Predict the operating load of building cooling sources.

[0091] The beneficial effects of this embodiment are as follows: by combining various parameters (such as cooling capacity, water inlet temperature, temperature difference, etc.) and inputting them into a special back propagation model, this method can provide more accurate operating load prediction for each cold source equipment (such as chillers, refrigerated water pumps, cold water pumps and cooling towers). Example 4

[0092] This embodiment provides a method for establishing a fuzzy model of parameters of each cold source equipment in a building on the basis of embodiment 1, and the specific steps include:

[0093] First, mathematical modeling is performed on each cold source device, and a mathematical model for output parameters is established according to different parameters and coefficients, including unknown coefficients; in this embodiment, the output parameter is the operating load of the cold source device.

[0094] The embodiments of the present invention do not limit the mathematical modeling method. The mathematical modeling method may be a theoretical model based on physical laws, a purely data-driven empirical model, or an empirical model summarized from industry work experience.

[0095] Secondly, unknown coefficients in the mathematical model are fuzzily identified; the embodiment of the present invention does not limit the fuzzy identification method of the unknown coefficients.

[0096] As a preferred embodiment, a chiller is modeled to analyze the effects of cooling capacity, chilled water inlet temperature, and chilled water inlet temperature on the operating efficiency of the chiller, which is expressed by a mathematical model as follows: ,in Indicates the operating efficiency of the chiller. Indicates the cooling capacity, Indicates the freezing water inlet temperature. Indicates the cold water inlet temperature. Represents the unknown coefficient.

[0097] In this way, the operating load of the chiller is: .

[0098] In the embodiment of the present invention, the nonlinear solver Octave is used to perform fuzzy solution on the unknown coefficients of the chiller model using the dichotomy method; as a preferred embodiment, the Octave instruction is: ,in, is a set of unknown coefficients, bin indicates that the algorithm used is dichotomy, C indicates the residual, and N indicates the input data set of the model.

[0099] As a preferred embodiment, a cold water pump is modeled. The cold water pump operating load is related to the cold water pump flow, condensate cooling capacity, cold water inlet temperature and cold water temperature difference, which can be expressed by a mathematical model as follows:

[0100] ;

[0101] ;

[0102] ;

[0103] in, is the operating load of the cooling water pump, is the cooling water pump flow rate, is the cold water temperature difference, is the condensate cooling capacity, is the water density, is the unknown coefficient.

[0104] Likewise, the unknown coefficients are fuzzily solved by a nonlinear solver.

[0105] As a preferred embodiment, a cooling tower is modeled, and the operating load of the cooling tower is related to the cooling air volume, which is expressed by a mathematical model as follows:

[0106] ,in, is the cooling tower operating load, is the cooling air volume, are the unknown coefficients; then, the unknown coefficients are fuzzily solved through the nonlinear solver.

[0107] The beneficial effects of this embodiment are as follows: this embodiment can more accurately adapt to various equipment and operating conditions by mathematically modeling different cold source equipment and fuzzy identifying the unknown coefficients in the model; the use of mathematical models combined with fuzzy identification methods can more accurately reflect the working status and operating efficiency of the actual equipment, thereby providing more reliable prediction results. Example 5

[0108] This embodiment provides a method for establishing an artificial neural network fusion model of a building cooling source based on Embodiment 1. In steps S1 to S4 of the embodiment of the present invention, the operating load of the building cooling source is predicted respectively by a back propagation network model and a parameter fuzzy model. Due to different parameter conditions, the prediction accuracy of the two models is different.

[0109] Since the back propagation network model is an artificial neural network model, the network is trained by training data. When the input parameters are close to the parameters of the training data, the output of the back propagation neural network has higher credibility and accuracy. Under this condition, the output of the back propagation network model is selected as the predicted building cooling source operating load; when the input parameters are far from the parameters of the training data, the output of the parameter fuzzy model is selected as the predicted building cooling source operating load.

[0110] As a feasible implementation method, the building cold source artificial neural network fusion model is implemented through a clustering algorithm. The specific steps include:

[0111] Step S501: Data preparation.

[0112] A set of input parameters is determined as feature data; on the other hand, training data of the back propagation network model is obtained, which includes input parameters and output parameters.

[0113] Step S502: cluster analysis.

[0114] The specific steps of cluster analysis include: using the K clustering algorithm to cluster the training data to form several clusters; calculating the center of each cluster.

[0115] Specifically, first, initialize the cluster center and select multiple data points as the initial cluster center; the number of data points matches the number of features in the feature data; then, for each data point in the training data, find the cluster center closest to it and assign the data point to this cluster, where the distance between the data point and the cluster center is calculated by Euclidean distance; then, based on the existing cluster assignment, recalculate the cluster center by taking the average of all feature data assigned to the cluster; finally, after obtaining the new cluster center, repeatedly assign each data point in the training set to the side of the latest cluster center, then recalculate the cluster center again, and repeat the above steps until the cluster center no longer changes significantly or the preset number of iterations is reached.

[0116] Step S503: Model prediction and selection.

[0117] The strategy for model selection includes: the distance between the new input parameter and the center of each cluster needs to be calculated, and then a threshold is set to determine whether the input parameter is close to the cluster of the training data, which is used to determine which model to select for predicting the operating load of the building cooling source.

[0118] Specifically, the new input parameters are , for each cluster center ,in , calculated by the formula The Euclidean distance from the center is given by ,in, is the number of input parameters; after that, the proximity threshold is determined According to the above calculation, we get Minimum distance to all cluster centers ,if , then it is believed that Close to the cluster, that is, close to the training data, then Input the back propagation network model for prediction; if , then Input parameters to the fuzzy model for prediction.

[0119] As a possible implementation, the proximity threshold The calculation of can be done by calculating the average distance from the center of all clusters to the center of other clusters, and taking the average value as the proximity threshold.

[0120] The beneficial effects of this embodiment are as follows: by fusing the back propagation network model and the parameter fuzzy model, this embodiment can select the most appropriate model for prediction under different conditions, thereby improving the overall prediction accuracy and adaptability to different parameter conditions; by using the clustering algorithm to determine whether the input parameters are close to the training data, the appropriate model can be selected under more accurate conditions, thereby enhancing the reliability of the prediction results; accurate load prediction helps to more effectively manage and adjust the operation of building cooling equipment, and provide more scientific decision-making support for energy management.

[0121] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0122] 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 predicting the operating load of a building cooling source using an artificial neural network algorithm, characterized in that: The specific steps include: Step S1: Establish a back propagation network model for each cold source equipment in the building; Step S2: predicting the operating load of the building's cooling source through the back propagation network model of each cooling source device in the building; Step S3: Establishing a fuzzy model of parameters of each cold source equipment in the building; Step S4: predicting the operating load of the building's cooling source through the fuzzy model of the building's cooling source equipment parameters; Step S5: establishing a building cooling source artificial neural network fusion model, and selecting a model with the best prediction effect according to a clustering algorithm to perform building cooling source operation load prediction; The specific steps of establishing the fuzzy model of the parameters of each cold source equipment in the building in step S3 include: Mathematically modeling each cold source device, and establishing a mathematical model for output parameters according to different parameters and coefficients, including unknown coefficients; the output parameter is the operating load of the cold source device; Fuzzy identification of unknown coefficients in mathematical models; The chiller is modeled to analyze the effects of cooling capacity, chilled water inlet temperature and cold water inlet temperature on the operating efficiency of the chiller, which is expressed by a mathematical model as follows: Where W represents the operating efficiency of the chiller, Q b Indicates cooling capacity, T dsr Indicates the freezing water inlet temperature, T lsr represents the cold water inlet temperature, β1~β8 represent unknown coefficients; the operating load of the chiller is: The chilled water pump is modeled. The chilled water pump operating load is related to the chilled water pump flow, condensate cooling capacity, chilled water inlet temperature, and chilled water temperature difference, which can be expressed by the mathematical model as follows: Q nsb =Q b +P lsj ; Among them, P lsb is the operating load of the cooling water pump, G lsb is the cooling water pump flow rate, ΔT lsc is the cold water temperature difference, Q nsb is the condensate cooling capacity, ρ w is the water density, h1~h3 are unknown coefficients; The cooling tower is modeled. The operating load of the cooling tower is related to the cooling air volume, which can be expressed by the mathematical model as follows: Among them, P lqt is the operating load of the cooling tower, m is the cooling air volume, j1~j3 are unknown coefficients; The specific steps of the fuzzy identification method include: using a nonlinear solver Octave to perform fuzzy solving on the unknown coefficients of the chiller model using a dichotomy method; the Octave command is: [x,bin,C]=regre(N,P lsj ), where x is the set of unknown coefficients, bin indicates that the algorithm used is dichotomy, C indicates the residual, and N indicates the input data set of the model; The specific steps of step S5 to establish the building cold source artificial neural network fusion model include: Step S501: data preparation; Step S502: cluster analysis; Step S503: model prediction and selection; The specific steps of data preparation include: determining a set of input parameters as feature data; obtaining training data of a back propagation network model, including input parameters and output parameters; The steps of cluster analysis include: initializing cluster centers, selecting multiple data points as initial cluster centers; matching the number of data points with the number of features of feature data; for each data point in the training data, finding the closest cluster center, and assigning the data point to this cluster, wherein the distance between the data point and the cluster center is calculated by Euclidean distance; recalculating the cluster center based on the existing cluster assignment, and obtaining it by taking the average value of all feature data assigned to the cluster; finally, after obtaining the new cluster center, repeatedly assigning each data point in the training set to the side of the latest cluster center, and then recalculating the cluster center again, and repeating the above steps until the cluster center no longer changes significantly, or a preset number of iterations is reached; The specific steps of model prediction and selection include: the new input parameter is x, for each cluster center u i , where i = 1, 2, 3…, K, the Euclidean distance between x and the center is calculated by the formula: Where m is the number of input parameters; then, determine the proximity threshold θ, and according to the above calculation, get the minimum distance d between x and the center of all clusters min =min(d(x,u i )), if d min <θ, then x is considered to be close to the cluster, that is, close to the training data. At this time, x is input into the back propagation network model for prediction; if d min >θ, then x is input into the parameter fuzzy model for prediction; The calculation of the proximity threshold θ is performed by calculating the average value of the distances from the centers of all clusters to the centers of other clusters, and taking the average value as the proximity threshold.

2. The method for predicting the operating load of a building cooling source using an artificial neural network algorithm according to claim 1 is characterized in that: The specific steps of establishing the reverse propagation network model of each cold source equipment in the building in step S1 include: Step S101: Determine the input and output parameters of the back propagation network; Step S102: Determine the back propagation network transfer function; Step S103: Determine the back propagation network learning algorithm; Step S104: Determine the number of hidden layer nodes of the back propagation network; The back propagation network includes an input layer, a hidden layer and an output layer; The output parameter of the reverse propagation network is the cold source operation load; the input parameters of the reverse propagation network include the parameters of each cold source equipment; The cold source equipment includes a chiller, a water pump and a cooling tower; the water pump includes a cold water pump and a frozen water pump; the input parameters of the back propagation network model of the chiller include: cooling capacity Q b , Frozen water inlet temperature T dsr , Chilled water temperature difference ΔT dsc , cold water inlet temperature T lsr and cold water temperature difference ΔT lsc ; Output parameters include chiller operating load P lsj ; The input parameters of the back propagation network model of the chilled water pump include: cooling capacity Q b , Frozen water inlet temperature T dsr and the temperature difference of frozen water ΔT dsc ; Output parameters include chilled water pump operating load P dsb ; The input parameters of the cold water pump back propagation network model include: condensate cooling capacity Q nsb , cold water inlet temperature T lsr and cold water temperature difference ΔT lsc ; Output parameters include the cooling water pump operating load P lsb ; The input parameters of the cooling tower back propagation network model include: condensation cooling capacity Q nsb , cold water inlet temperature T lsr , cold water temperature difference ΔT lsc and wet bulb ambient temperature T hj ; Output parameters include cooling tower operating load P lqt .

3. The method for predicting the operating load of a building cooling source using an artificial neural network algorithm according to claim 2 is characterized in that: The chiller has 7 hidden layer nodes, the frozen water pump has 9 hidden layer nodes, the cold water pump has 9 hidden layer nodes, and the cooling tower has 14 hidden layer nodes.

4. The method for predicting the operating load of a building cooling source using an artificial neural network algorithm according to claim 1, characterized in that: The specific steps of predicting the operating load of the building cooling source in step S2 and step S4 include: The cooling capacity Q b , Frozen water inlet temperature T dsr , Chilled water temperature difference ΔT dsc , cold water inlet temperature T lsr and cold water temperature difference ΔT lsc Input the chiller back propagation model and output the chiller operating load P lsj ; The cooling capacity Q b , Frozen water inlet temperature T dsr and the temperature difference of frozen water ΔT dsc Input the back propagation model of the chilled water pump and output the chilled water pump operating load P dsb ; Calculate the condensate cooling capacity and convert the condensate cooling capacity Q nsb , cold water inlet temperature T lsr and cold water temperature difference ΔT lsc Input the cold water pump back propagation model and output the cold water pump operating load P lsb ; The condensate cooling capacity calculation formula is: Q nsb =Q b +P lsj ; The condensate cooling capacity Q nsb , cold water inlet temperature T lsr , cold water temperature difference ΔT lsc and wet bulb ambient temperature T hj Input the cooling tower back propagation model and output the cooling tower operating load P lqt ; By formula P all =P lsj +P dsb +P lsb +P lqt Predict the operating load of building cooling sources.