A method for HVAC system parameter identification using grid uniform sampling

Abstract

The present invention discloses a method for identifying parameters of a HVAC system using grid uniform sampling, comprising the following steps: (1) acquiring historical operating data of the HVAC system and dividing it into a training set and a validation set; performing correlation analysis on the training set data to determine characteristic variables reflecting the operating conditions of the HVAC system; (2) re-expressing the training set data in the dimension of characteristic variables according to the determined characteristic variables, dividing the corresponding data space into multiple grids and sampling, extracting representative data points, and forming a representative historical operating condition training set; (3) using the representative historical operating condition training set, using an optimization algorithm to perform parameter identification on a simulation model of the HVAC system. The present invention improves the identification accuracy and computational efficiency, and provides guidance for achieving more efficient and accurate model parameter identification with minimal data.

Description

Technical Field

[0001] The present invention belongs to the field of HVAC system parameter identification, and in particular relates to a HVAC system parameter identification method using grid uniform sampling. Background Art

[0002] The HVAC system is large in scale and complex in structure, and its optimized operation has always been a challenging task. In order to achieve efficient operation and maintenance of the system, one of the important ways is to establish an accurate and reliable digital twin model. In the process of building a digital twin model, the key is to perform parameter identification, ensure the accuracy and reliability of the system model, and narrow the gap between the actual measurement data and the model output.

[0003] For example, a Chinese patent document with publication number CN102981410A discloses a method for identifying parameters of an air-conditioning energy consumption model based on an artificial bee colony. The method uses a mechanism analysis method to establish an energy consumption model function of the entire air-conditioning system. The energy consumption model function is then used as the objective function and optimized using an artificial bee colony algorithm to determine the optimal parameter combination under a certain load state.

[0004] A Chinese patent document with publication number CN104238368A discloses a method for identifying parameters of an air conditioning energy consumption model based on simulated annealing particle swarm. The energy consumption model function of the entire air conditioning system established by the mechanism analysis method is used as the objective function, and an algorithm combining simulated annealing and particle swarm is used to optimize it to determine the optimal parameter combination under a certain load.

[0005] The operation time of HVAC systems is usually relatively long, which may last for several months, and a large amount of historical data has been accumulated. Due to the long duration of certain working conditions, there is a certain redundancy in the operating data. The traditional method of not processing the data has too high a computational cost for parameter identification. Traditional clustering methods may generate working condition data that has never appeared in practice. When the training data is limited, the model may overfit such data, which has a negative impact on parameter identification. Therefore, it is of great significance to develop a high-computational efficiency and accurate parameter identification method for the field of digital twin modeling of HVAC systems. Summary of the invention

[0006] The present invention provides a HVAC system parameter identification method using grid uniform sampling, which improves the parameter identification efficiency while ensuring that the model can accurately capture the real system characteristics.

[0007] A method for identifying parameters of a HVAC system using grid uniform sampling comprises the following steps:

[0008] (1) Obtain historical operating data of the HVAC system and divide it into a training set and a validation set; perform correlation analysis on the training set data to determine characteristic variables that reflect the operating conditions of the HVAC system;

[0009] (2) According to the determined characteristic variables, the training set data is re-represented in the dimension of the characteristic variables, the corresponding data space is divided into multiple grids and sampled, representative data points are extracted, and a representative historical working condition training set is formed;

[0010] (3) Using a representative historical operating condition training set, an optimization algorithm is used to identify the parameters of the simulation model of the HVAC system.

[0011] Furthermore, in step (1), the historical operating data of the HVAC system includes power, cooling load, chilled water outlet temperature, chilled water supply temperature, chilled water return temperature, chilled water inlet temperature, cooling water outlet temperature, cooling water inlet temperature, chilled water flow rate and cooling water flow rate.

[0012] In step (1), correlation analysis is performed on the training set data. The specific process is as follows:

[0013] Perform correlation analysis on the training set data to obtain the correlation coefficient r between each variable, quantify the degree of association between different variables, and the calculation formula of the correlation coefficient is:

[0014]

[0015] Where X and Y are the two variables to be analyzed, cov(X,Y) is the covariance between X and Y, s(X) and s(Y) are the standard deviations of X and Y respectively;

[0016] The variables with higher correlation coefficients are replaced, and a group of representative features are selected from the highly correlated variables to obtain the key characteristic variables that reflect the operating conditions of the HVAC system.

[0017] Preferably, substitution is performed between a group of variables with a correlation coefficient higher than 0.75. For example, for two highly correlated variables X and Y, if their correlation coefficient is 0.8, only one of them, such as X, can be retained, and Y can be replaced by X.

[0018] In step (2), the method of dividing the corresponding data space into multiple grids and sampling is specifically as follows:

[0019] (2-1) Determine the data space range to be sampled;

[0020] (2-2) Set the sampling rate and calculate the number of data points that need to be sampled based on the sampling rate and the size of the training set;

[0021] (2-3) Divide each dimension of the data space into equal sub-regions to form a 100×100 grid structure;

[0022] (2-4) Determine the number of samples for each grid. If the number of data points in the grid exceeds the number of samples, randomly select the number of points in the grid. If the number of data points in the grid is less than or equal to the number of samples, ignore the grid.

[0023] (2-5) If the total number of samples does not meet the requirements, multiple adjustments can be made until the set total number of data points is reached.

[0024] The specific process of step (3) is as follows:

[0025] (3-1) Using the representative historical operating condition training set as the input of the simulation model;

[0026] (3-2) The optimization algorithm is used for parameter identification. The optimization variable is the parameter to be identified of the model, and the objective function is:

[0027]

[0028] In the formula, is the model output value, y i is the true value in the training set, and n is the number of moments of model simulation;

[0029] (3-3) The optimization algorithm first randomly generates a group of individuals and initializes the set of parameters to be identified; the generated parameters to be identified are passed into the simulation model of the HVAC system for simulation, and the output of the simulation model is used to calculate the objective function; the performance of each individual is evaluated using the objective function, thereby calculating the fitness of each individual; after the selection, crossover and mutation processes, a new population is formed; this process is iterated until the stopping criterion is reached, thereby obtaining the identification parameters of the model;

[0030] (3-4) Assign the obtained identification parameters to the model to obtain the identified model, and verify the model on the verification set;

[0031] (3-5) If the performance of the model on the validation set meets the predetermined criteria, the parameter identification is considered complete; otherwise, steps (3-2) to (3-5) are iterated until the final model identification result is obtained.

[0032] Preferably, in step (3), the input of the HVAC system simulation model is the cooling load; the output of the simulation model includes the model power and the outlet water temperature, which participate in the calculation of the objective function.

[0033] Optionally, the type of the HVAC system includes but is not limited to a chiller, a refrigeration water pump, a cooling water pump or a cooling tower.

[0034] The optimization algorithm includes but is not limited to genetic algorithm GA, ant colony algorithm ACO or particle swarm algorithm PSO.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. Compared with the traditional method of not processing the operating data, the present invention reduces the redundancy of the operating data by screening the characteristic variable dimensions of the operating data and extracting representative operating data points, thereby improving the efficiency of parameter identification.

[0037] 2. In traditional clustering methods, the operating conditions of the data in the same cluster may be significantly different, and synthetic data may be generated that are not actually present. In the case of limited training data, the model may overfit the synthetic data, which has a negative impact on parameter identification. The data points selected by the present invention are the operating conditions that exist in actual operation, ensuring that the model can accurately capture the real system characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a method for identifying parameters of a HVAC system using grid uniform sampling according to the present invention.

[0039] Figure 2 It is the model diagram of the cold source system TRNSYS in the embodiment.

[0040] Figure 3 Schematic diagram of correlation coefficients of running data variables in the embodiment.

[0041] Figure 4 Schematic diagram of uniform grid sampling results in the feature variable space in the embodiment.

[0042] Figure 5 This is the model power fitting curve of the inventive method in the embodiment on the test set.

[0043] Figure 6 This is a model power fitting curve of the clustering-based method in the embodiment on the test set.

[0044] Figure 7 This is the model power fitting curve on the test set for the method without processing data in the embodiment. DETAILED DESCRIPTION

[0045] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0046] like Figure 1As shown, a parameter identification method for a HVAC system using grid uniform sampling mainly includes two steps: characteristic condition extraction based on grid uniform sampling and model parameter identification based on representative data.

[0047] Feature condition extraction based on grid uniform sampling refers to obtaining representative historical data conditions so that less data can be used in parameter identification to achieve more efficient parameter identification. This step includes obtaining training data, correlation analysis, feature variable dimension selection, grid division, and grid uniform sampling.

[0048] Model parameter identification based on representative data refers to taking representative data as the input data of the model to perform parameter identification and optimization. This step includes obtaining representative data, optimizing the parameters to be identified, and finally obtaining the identification parameters.

[0049] The following describes in detail the specific implementation of the two steps of characteristic condition extraction based on grid uniform sampling and model parameter identification based on representative data in a preferred embodiment of the present invention.

[0050] Step S1 is to extract characteristic working conditions based on grid uniform sampling, and the specific steps are as follows:

[0051] S11, obtaining historical load data of the HVAC system from the SCADA system, wherein the historical operation data of the HVAC system includes power, cooling load, chilled water outlet temperature, chilled water supply temperature, chilled water return temperature, chilled water inlet temperature, cooling water outlet temperature, cooling water inlet temperature, chilled water flow rate and cooling water flow rate, etc.

[0052] S12, the HVAC system historical operation data set is divided into a training set and a validation set at a ratio of 1:20. A shorter training data set is used for training, and the effect of model identification is evaluated based on the mean absolute percentage error, standardized mean deviation error, coefficient of variation of the root mean square error, determination coefficient and goodness of fit of the simulation value and the actual value of the validation set in the longer operation period, which is in line with the actual scenario of parameter identification.

[0053] S13, perform correlation analysis on the training set data to obtain the correlation coefficient r between each variable, and quantify the degree of association between different variables. The calculation formula of the correlation coefficient is:

[0054]

[0055] Where X and Y are the two variables to be analyzed, cov(X,Y) is the covariance between X and Y, and s(X) and s(Y) are the standard deviations of X and Y, respectively.

[0056] S14, replace a group of variables with correlation coefficients higher than 0.75. For example, for two highly correlated variables X and Y, if their correlation coefficient is 0.8, only one of them, such as X, can be retained, and Y can be replaced by X. Thus, a group of representative features are selected from the highly correlated variables to obtain the characteristic variables of the HVAC system operation conditions.

[0057] S15, reorganizing the training set data according to the selected characteristic variable as the dimension, obtaining a re-representation of the training set data under the selected characteristic variable dimension, and the dimension of the formed data space is the dimension of the selected variable.

[0058] S16, dividing the new data space into multiple grids according to the variable dimension, setting the boundary of the grid according to the range of the data space, selecting a suitable grid division granularity, dividing the data space into small blocks, and forming a grid.

[0059] S17, uniform sampling is performed in each grid, one or more data points are randomly selected as representative data points in the grid, and a representative subset is extracted from the overall data. When the data has a high density or large variation in certain areas, the overall characteristics of the data can be better captured.

[0060] S18, extract representative data points from each grid to form a representative historical operating condition training set that is smaller in size than the original data set but still retains the main features of the data.

[0061] Step S2 is model parameter identification based on representative data, and the specific steps are as follows:

[0062] S21, using the representative historical operating condition training set described in steps 1-8 as input of the simulation model;

[0063] S22, using optimization algorithm to identify parameters, the optimization variable is the model parameters to be identified, the objective function is:

[0064]

[0065] In the formula, is the model output value, y i is the true value in the training set, and n is the number of moments of model simulation;

[0066] S23, the optimization algorithm first randomly generates a group of individuals and initializes the set of parameters to be identified. The generated parameters to be identified are passed into the HVAC model for simulation, and the output of the model is used to calculate the objective function. The performance of each individual is evaluated using the objective function, thereby calculating the fitness of each individual. After the selection, crossover and mutation processes, a new population is formed. This process is iterated until the stopping criteria is reached, such as reaching the maximum number of iterations, thereby obtaining the identification parameters of the model;

[0067] S24, assigning the identification parameters to the model to obtain an identified model, verifying the model on a verification set, and using an evaluation index to evaluate the performance of the model under real working conditions;

[0068] S25, if the performance of the model on the validation set meets the predetermined standard, the parameter identification is considered to be completed. Otherwise, S22 to S25 can be iterated until a satisfactory model identification result is obtained;

[0069] In order to further demonstrate the advantages of the present invention, it is applied to a specific scenario example below to demonstrate its technical effects.

[0070] This example selects a cold source system of a factory in Wuhan as the data source for example verification, and uses the operating data of the system's 1# chiller. In addition, local weather information is also selected, including dry-bulb temperature and wet-bulb temperature. The training set and test set are divided into a ratio of 1:20. The training set is used to identify and calibrate the model, and the test set is used to predict the identified model. The cold source system is modeled using TRNSYS software, and the established system model is as follows: Figure 2 shown.

[0071] In order to study the performance of the present invention, two methods were selected for comparison. One method uses the complete original data set as the training set, and the other method uses clustering, and the center point of the cluster obtained by clustering is used as the training set data. The size of the training data set based on the clustering method is consistent with the present invention.

[0072] This embodiment performs correlation analysis on the training set data, and the correlation coefficient matrix heat map is as follows Figure 3As shown. The Pearson correlation coefficient measures the linear relationship between two parameters, and its value ranges from -1 to 1. 1 indicates a perfect positive correlation, 0 indicates no linear correlation, and -1 indicates a perfect negative linear correlation. The correlation coefficients between power-cooling load (r=0.99), power-chilled water return temperature (r=0.91), power-chilled water inlet temperature (r=0.91), and power-cooling water outlet temperature (r=0.95) are very high, which means that the power variable can be effectively substituted for other variables. The correlation coefficients between instantaneous flow rate of chilled water-chilled water outlet temperature (r=0.78) and instantaneous flow rate of chilled water-chilled water supply temperature (r=0.78) indicate that the instantaneous flow rate of chilled water can be substituted for the chilled water outlet temperature and the chilled water supply temperature. The instantaneous flow rate of cooling water remains stable, so the correlation coefficient with other variables is low, and it is excluded when selecting characteristic variables. Based on the above analysis, chiller power and instantaneous flow rate of chilled water are selected as characteristic variables.

[0073] In this embodiment, the data space is divided into 100×100 grids according to the range of power and flow, and the data is dispersed in 30 grids, such as Figure 4 As shown in the figure, it can be seen that the data set is concentrated in certain specific values ​​or ranges, indicating that many data points reflect the same operating conditions and there is data redundancy. Uniform sampling is performed in each grid. If there are more than 5 data points, 5 of them are randomly selected. The number of samples can be adjusted according to the size of the original data set. The representative data points obtained by uniform grid sampling total 55, as shown in Figure 4 The sampled representative dataset will be used as the training set for model calibration.

[0074] The optimization algorithm used in this embodiment is a genetic algorithm (GA), which is implemented using the MATLAB global optimization toolbox, and parameter identification is performed by joint simulation of MATLAB and TRNSYS software. The population size and the maximum number of iterations are 30 and 10, respectively. The model parameters corresponding to the minimum value of the objective function are considered to be the optimal identification parameters. Table 1 lists the results of parameter identification using the method of the present invention and the other two methods.

[0075] Table 1 Parameter identification results

[0076]

[0077] Comparison of the simulation results and actual values ​​of the chiller power of the three methods on the test data set Figure 5 to Figure 7As shown. The calibration accuracy of the above three methods was evaluated on the test data set, as shown in Table 2. Compared with other methods, the present invention has higher accuracy. To achieve the same accuracy, the clustering-based method requires 640 iterations of the genetic algorithm, which takes 3.25 hours to converge. In contrast, the present invention only took less than half the time, that is, 1.5 hours to achieve convergence. At the same time, the method that does not process data requires 1040 iterations and takes 5.5 hours to converge.

[0078] Therefore, compared with traditional methods that do not process data and clustering-based methods, the present invention can achieve a significant improvement in computing efficiency while ensuring accuracy, which is of great significance for improving the efficiency of digital twin modeling.

[0079] Table 2 Comparison of the accuracy of the present invention and the traditional method

[0080]

[0081] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying parameters of a HVAC system using grid uniform sampling, characterized in that: The following steps are involved: (1) Obtain historical operating data of the HVAC system and divide it into a training set and a validation set; perform correlation analysis on the training set data to determine characteristic variables that reflect the operating conditions of the HVAC system; (2) According to the determined characteristic variables, the training set data is re-represented in the dimension of the characteristic variables, the corresponding data space is divided into multiple grids and sampled, representative data points are extracted, and a representative historical working condition training set is formed; (3) Using a representative historical operating condition training set, an optimization algorithm is used to identify the parameters of the simulation model of the HVAC system.

2. The HVAC system parameter identification method using grid uniform sampling according to claim 1, characterized in that: In step (1), the historical operating data of the HVAC system includes power, cooling load, chilled water outlet temperature, chilled water supply temperature, chilled water return temperature, chilled water inlet temperature, cooling water outlet temperature, cooling water inlet temperature, chilled water flow rate and cooling water flow rate.

3. The HVAC system parameter identification method using grid uniform sampling according to claim 1, characterized in that: In step (1), correlation analysis is performed on the training set data. The specific process is as follows: Perform correlation analysis on the training set data to obtain the correlation coefficient r between each variable, quantify the degree of association between different variables, and the calculation formula of the correlation coefficient is: Where X and Y are the two variables to be analyzed, cov(X,Y) is the covariance between X and Y, s(X) and s(Y) are the standard deviations of X and Y respectively; The variables with higher correlation coefficients are replaced, and a group of representative features are selected from the highly correlated variables to obtain the key characteristic variables that reflect the operating conditions of the HVAC system.

4. The HVAC system parameter identification method using grid uniform sampling according to claim 3 is characterized in that: Substitution was performed between a group of variables with correlation coefficients higher than 0.

75.

5. The HVAC system parameter identification method using grid uniform sampling according to claim 1, characterized in that: In step (2), the method of dividing the corresponding data space into multiple grids and sampling is specifically as follows: (2-1) Determine the data space range to be sampled; (2-2) Set the sampling rate and calculate the number of data points that need to be sampled based on the sampling rate and the size of the training set; (2-3) Divide each dimension of the data space into equal sub-regions to form a 100×100 grid structure; (2-4) Determine the number of samples for each grid. If the number of data points in the grid exceeds the number of samples, randomly select the number of points in the grid. If the number of data points in the grid is less than or equal to the number of samples, ignore the grid. (2-5) If the total number of samples does not meet the requirements, multiple adjustments are made until the set total number of data points is reached.

6. The method for identifying parameters of a HVAC system using grid uniform sampling according to claim 1, characterized in that: The specific process of step (3) is as follows: (3-1) Using the representative historical operating condition training set as the input of the simulation model; (3-2) The optimization algorithm is used for parameter identification. The optimization variable is the parameter to be identified of the model, and the objective function is: In the formula, is the model output value, y i is the true value in the training set, and n is the number of moments of model simulation; (3-3) The optimization algorithm first randomly generates a group of individuals and initializes the set of parameters to be identified; the generated parameters to be identified are transmitted to the simulation model of the HVAC system for simulation, and the output of the simulation model is used to calculate the objective function; the performance of each individual is evaluated using the objective function, thereby calculating the fitness of each individual; after the selection, crossover and mutation processes, a new population is formed; This process is iterated until the stopping criterion is reached, thereby obtaining the identification parameters of the model; (3-4) Assign the obtained identification parameters to the model to obtain the identified model, and verify the model on the verification set; (3-5) If the performance of the model on the validation set meets the predetermined criteria, the parameter identification is considered complete; Otherwise, steps (3-2) to (3-5) are iterated until the final model identification result is obtained.

7. The HVAC system parameter identification method using grid uniform sampling according to claim 1, characterized in that: In step (3), the types of the HVAC system include a chiller, a refrigeration water pump, a cooling water pump or a cooling tower.

8. The method for identifying parameters of a HVAC system using grid uniform sampling according to claim 1, characterized in that: In step (3), the optimization algorithm is a genetic algorithm GA, an ant colony algorithm ACO or a particle swarm algorithm PSO.

Images