Parameter identification method of HVAC model based on GPS-PSO algorithm

Abstract

The invention relates to the technical field of HVAC system modeling and simulation, and discloses a HVAC model parameter identification method based on a GPS-PSO algorithm. The method comprises the following steps: determining key parameters that affect the accuracy of a HVAC model based on expert knowledge and association rules; performing sensitivity analysis on the key parameters, and sorting them from high to low according to the sensitivity, to construct a parameter set of the HVAC model to be identified; collecting actual operation data related to the parameter set of the HVAC model to be identified, and performing preprocessing; dividing the preprocessed actual operation data into a calibration data set and a verification data set according to a preset ratio; based on the calibration data set, taking the normalized mean error as the optimization target, and performing parameter identification using a GPS-PSO two-stage optimization algorithm; and using the verification data set to verify the effectiveness of parameter identification, thereby improving the efficiency and accuracy of HVAC model parameter identification.

Description

Technical Field

[0001] The present invention relates to the technical field of HVAC system modeling and simulation, and in particular to a HVAC model parameter identification method based on a GPS-PSO algorithm. Background Art

[0002] With the increase in energy demand in the construction industry, building energy consumption accounts for about 30% of the world's total energy consumption, of which air conditioning equipment accounts for more than 50% or even more. Traditional building energy management relies on manual experience, and there are problems such as operational errors, time waste and inaccurate decision-making, making it difficult to achieve energy conservation and low-carbon goals. Emerging technologies such as the Internet of Things, artificial intelligence and big data have promoted the digital transformation of the construction industry. In the future, the construction of energy models based on advanced optimization algorithms and information technology will help to effectively simulate, predict, analyze and optimize building energy systems. However, the complexity and nonlinearity of heating, ventilation and air conditioning (HVAC) systems and the influence of dynamic environments make model parameter identification face many technical challenges.

[0003] The existing methods for identifying parameters of heating, ventilation and air conditioning (HVAC) systems have certain limitations. The calibration method based on the least squares method requires known performance curves and clear mathematical expressions, which is suitable for systems with known input-output relationships. However, it is difficult to fully capture the dynamic behavior of complex nonlinear, time-varying and multivariable coupled HVAC systems. Another type of method is based on machine learning algorithms, such as neural networks and support vector machines, which can learn and predict through data-driven learning, but they are highly dependent on data quality, and the identification effect will decrease when there is noise or poor data. In addition, the existing technology is complex in calculation and it is difficult to meet the needs of fast, accurate and real-time online parameter adjustment.

[0004] Therefore, there is an urgent need for a HVAC model parameter identification method based on the GPS-PSO algorithm to improve the efficiency and accuracy of HVAC model parameter identification. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a HVAC model parameter identification method based on the GPS-PSO algorithm, which can improve the efficiency and accuracy of HVAC model parameter identification.

[0006] The present invention provides a HVAC model parameter identification method based on GPS-PSO algorithm, comprising the following steps:

[0007] S1. Determine the key parameters that affect the accuracy of the HVAC model based on expert knowledge and association rules;

[0008] S2. Perform sensitivity analysis on the key parameters and sort them from high to low according to sensitivity to construct a set of parameters to be identified for the HVAC model;

[0009] S3, collecting actual operation data according to the parameter set to be identified of the HVAC model, and performing preprocessing;

[0010] S4, dividing the preprocessed actual operation data into a calibration data set and a verification data set according to a preset ratio;

[0011] S5. Based on the calibration data set, the normalized mean error is used as the optimization target, and the GPS-PSO two-stage optimization algorithm is used for parameter identification;

[0012] S6. Use the validation data set to verify the effectiveness of parameter identification.

[0013] Furthermore, in S2, the key parameters are subjected to sensitivity analysis and are sorted from high to low according to sensitivity, including:

[0014] S21. Perform a single parameter change experiment, change the set value of a key parameter alone, fix other parameters, run the HVAC model and record the output results;

[0015] S22, calculating the standard deviation and mean square error of the output results caused by the change of a single parameter;

[0016] S23. Sort the sensitivity of key parameters from high to low according to standard deviation and mean square error.

[0017] Furthermore, in S3, collecting actual operation data according to the set of parameters to be identified of the HVAC model includes:

[0018] S31, arranging the minimum data set required for the operation of the HVAC model according to the parameter set to be identified of the HVAC model;

[0019] S32. Classify the data in the data set into construction boundary data and characterization accuracy data according to different uses of the data, wherein the construction boundary data is used to define the limit conditions of the model working range, and the characterization accuracy data is used to improve the prediction accuracy of the model;

[0020] S33. Determine the cycle, frequency and scope of data collection according to the collection requirements of different data types, and collect actual operation data.

[0021] Furthermore, in S3, the actual operation data is preprocessed, including:

[0022] Use the Laida criterion and the quartile criterion to identify and delete outliers in the actual operating data;

[0023] Check for missing values ​​in actual operational data;

[0024] Missing values ​​are filled using the mean method, moving average method, or similarity algorithm.

[0025] Furthermore, the S5 specifically includes:

[0026] S51, according to the order of the sensitivity of the parameters to be identified of the HVAC model from high to low, the parameters to be identified are set as optimization variables in sequence, and the optimization range of the optimization variables is determined according to the nameplate parameters and the calibration data;

[0027] S52, using a particle swarm optimization (PSO) algorithm to encode and optimize discrete variables, with the goal of minimizing the normalized mean error, searching in the solution space, and obtaining preliminary model configuration parameter setting values;

[0028] S53, after obtaining the initial model configuration parameter setting values ​​in the PSO stage, enter the generalized mode search GPS algorithm stage to further optimize the accuracy of the solution until the normalized mean error of the calibration target reaches the minimum output value, and end the iterative process;

[0029] S54. Output the identification parameters that minimize the normalized mean error.

[0030] Furthermore, in S52, the particle swarm optimization PSO algorithm includes:

[0031] S521, initializing the particle swarm, and randomly assigning a position and a velocity to each particle;

[0032] S522, calculating the fitness value of each particle;

[0033] S523, updating the individual extreme value of each particle;

[0034] S524, updating the global extreme value;

[0035] S525, updating the speed and position of each particle according to the speed update formula and the position update formula;

[0036] S526. Repeat steps S522-S525 until the maximum number of iterations is reached.

[0037] Furthermore, the speed update formula is:

[0038] ;

[0039] In the formula, is the velocity of the i-th particle at the k+1th iteration; is the velocity of the i-th particle at the k-th iteration; is the inertia weight, which controls the influence of the previous iteration of the particle velocity; and are learning factors, which control the response degree of particles to individual extreme values ​​and global extreme values ​​respectively; and is a random number in the range [0,1]; is the optimal solution found by the i-th particle so far; The optimal solution found by the entire particle swarm so far;

[0040] The position update formula is:

[0041] ;

[0042] In the formula, x i (k+1) is the position of the i-th particle at the k+1th iteration; x i (k) is the position of the i-th particle at the k-th iteration.

[0043] Furthermore, in S53, the generalized mode search GPS algorithm specifically includes:

[0044] S531, selecting an initial design point and an initial step size;

[0045] S532, according to the initial design point and the initial step length, calculate the value of the objective function at the current position and in the preset neighborhood, and determine the moving direction for the next step;

[0046] S533, adjusting the step size and updating the design point according to the change of the objective function;

[0047] S534, check whether the convergence condition is met, if so, stop the algorithm; otherwise, return to step S532.

[0048] Furthermore, the S6 specifically includes:

[0049] Using Python to run the HVAC model and collect simulation result data under the constructed boundary data in the validation dataset;

[0050] The root mean square error between the characterization accuracy data and the simulation result data in the validation data set was calculated, and the validity of the parameter identification was judged by combining the root mean square error coefficient of variation and the coefficient of determination.

[0051] The embodiments of the present invention have the following technical effects:

[0052] 1. Automated parameter identification is achieved by applying the GPS-PSO two-stage optimization algorithm, which reduces human intervention and improves accuracy and objectivity. The particle swarm algorithm is used to encode and optimize discrete variables. The particle swarm algorithm can quickly conduct a global search in the solution space and gradually approach the global optimal solution through information sharing and collaboration between particles. When a better solution is obtained in the PSO stage, the GPS algorithm conducts a detailed search of the neighborhood of the solution and gradually adjusts the accuracy of the solution. The combination of the two in the optimization process not only improves the convergence speed, but also reduces the risk of local optimality, making the final model parameter identification more accurate.

[0053] 2. Through expert knowledge and association rule analysis, key parameters are determined, and single parameter variation experiments and statistical analysis are used to optimize parameter calibration and improve model prediction accuracy. By building a minimum data set and clarifying classification, the time and cost of data collection and processing are reduced, and the identification efficiency is improved. Multiple indicators such as root mean square error (RMSE), root mean square error coefficient of variation (CVRMSE) and determination coefficient (R²) are used to comprehensively evaluate the model calibration effect. Based on Python implementation, it is easy to program and deploy, which improves the convenience and efficiency of the model parameter identification process. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Figure 1 is a flow chart of a HVAC model parameter identification method based on a GPS-PSO algorithm provided in an embodiment of the present invention;

[0056] Figure 2 is a flow chart of a PSO algorithm provided by an embodiment of the present invention;

[0057] Figure 3 is a flow chart of a GPS algorithm provided by an embodiment of the present invention;

[0058] Figure 4 A chiller performance curve fitting diagram provided by an embodiment of the present invention;

[0059] Figure 5 A comparison chart of the simulation and measured data results of the cold water supply temperature provided by the embodiment of the present invention;

[0060] Figure 6 A comparison chart of the cooling water return temperature simulation and measured data results provided by an embodiment of the present invention;

[0061] Figure 7 A comparison diagram of power simulation and measured data results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. 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 belong to the scope of protection of the present invention.

[0063] Figure 1 is a flow chart of a HVAC model parameter identification method based on GPS-PSO algorithm provided by an embodiment of the present invention. Figure 1 The present invention provides a HVAC model parameter identification method based on GPS-PSO algorithm, comprising the following steps:

[0064] S1. Determine the key parameters that affect the accuracy of the HVAC model based on expert knowledge and association rules;

[0065] In some embodiments, expert knowledge can be obtained through interviews, literature surveys, experience summaries, etc. Under the guidance of experts, determine which parameters may have a significant impact on the output results of the model. By analyzing the relationship between different parameters, identify the key parameters that affect the accuracy of the model. Commonly used association rule mining algorithms, such as the Apriori algorithm or the FP-growth algorithm. Perform association rule mining on a large amount of historical data to find out which input variables (such as the COP of the chiller) have a strong correlation with the output variables (cooling water return temperature, cold water supply temperature). These variables are the model parameters to be identified that we need to focus on.

[0066] S2. Perform sensitivity analysis on the key parameters and sort them from high to low according to sensitivity to construct a set of parameters to be identified for the HVAC model, including:

[0067] S21. Perform a single parameter change experiment, change the set value of a key parameter alone, fix other parameters, run the HVAC model and record the output results;

[0068] S22, calculating the standard deviation and mean square error of the output results caused by the change of a single parameter;

[0069] S23. Sort the sensitivity of key parameters from high to low according to standard deviation and mean square error.

[0070] In some embodiments, a single parameter change experiment is performed, by changing the set value of each parameter to be identified, keeping other parameters unchanged, and observing how the system output changes. Only change the value of one parameter at a time, ensuring that other parameters remain unchanged, so as to quantify the independent impact of the parameter on the model output. Most of the parameters in the model are fixed, and one parameter is selected to change (for example, the rated COP of the chiller, or the rated electrical power, etc.). Change the set value of the parameter (for example, increase or decrease a certain amplitude), run the model and record the output results of the model. Repeat this process, perform similar experiments on all key parameters to be identified, and obtain the model response under each parameter change. After each experiment, record the output of the model (such as the cold water supply temperature of the chiller, the cooling water return temperature, etc.) and its changes. Then, calculate the standard deviation and mean square error of the output change caused by the parameter change. For each parameter, calculate the standard deviation and mean square error of the output result caused by its range of change.

[0071] For example, set the range of a parameter change to ΔP, measure the corresponding output change, and calculate the standard deviation (σ) or mean square error (MSE). By comparing these values ​​of different parameters, we can determine which parameters have a greater impact on the model output.

[0072] Based on the results of standard deviation and mean square error, all parameters to be identified are sorted by their influence. Parameters with greater influence should be calibrated and optimized first. Through this sorting, we can effectively focus on the parameters that have the greatest impact on model accuracy, thereby improving identification accuracy and efficiency.

[0073] S3, collecting actual operation data related to the parameter set to be identified of the HVAC model, and performing preprocessing;

[0074] Furthermore, in S3, the actual operation data collected according to the parameter set to be identified of the HVAC model includes:

[0075] S31. Arrange the minimum data set required for the operation of the HVAC model according to the parameter set to be identified of the HVAC model; with the model data identification as the goal, arrange the minimum data set required for the operation of the HVAC model to reduce the time required for data collection, cleaning and classification, avoid the complexity caused by redundant data, and improve the efficiency of model identification.

[0076] S32. According to the different uses of data, the data set is classified into construction boundary data and characterization accuracy data. Construction boundary data is used to define the extreme conditions of the model's working range. For example, the operating data of the HVAC system in extreme environments can help determine the performance boundary of the system. Characterization accuracy data is used to improve the prediction accuracy of the model; it usually represents detailed data under normal working conditions, which can help adjust the internal parameters of the model to enable it to have better prediction capabilities for normal operating conditions.

[0077] S33. According to the collection requirements of different data types, determine the cycle, frequency and scope of data collection, and collect actual operation data. Different data types require different data collection plans. According to the existing data collection system, formulate data testing cycles and plans to ensure the comprehensiveness and accuracy of the data.

[0078] Furthermore, in S3, the actual operation data is preprocessed, including:

[0079] Use the Laida criterion and the quartile criterion to identify and delete outliers in the actual operating data;

[0080] Check for missing values ​​in actual operational data;

[0081] Missing values ​​are filled using the mean method, moving average method, or similarity algorithm.

[0082] S4, dividing the preprocessed actual operation data into a calibration data set and a verification data set according to a preset ratio;

[0083] In some embodiments, the actual operation data after preprocessing is divided into a calibration data set and a verification data set according to a preset ratio. In the implementation process, the preset ratio can be determined according to the actual situation. For example, when the data is sufficient, the data can be divided according to a ratio of 70% / 30% or 80% / 20%. When more data is needed for verification, the data can be divided according to a ratio of 60% / 40%.

[0084] S5. Based on the calibration data set, the normalized mean error is used as the optimization target, and the GPS-PSO two-stage optimization algorithm is used for parameter identification, which specifically includes:

[0085] S51, according to the order of the sensitivity of the parameters to be identified of the HVAC model from high to low, the parameters to be identified are set as optimization variables in sequence, and the optimization range of the optimization variables is determined according to the nameplate parameters and the calibration data;

[0086] In step S2, sensitivity analysis is performed on the parameters to be identified, and the parameters to be identified are sorted according to sensitivity, and can be optimized in sequence according to the sensitivity sorting. A specific optimization range is set for each optimization variable according to the nameplate data and actual conditions to ensure that physical and technical constraints are not exceeded during the optimization process.

[0087] S52. Use the particle swarm optimization (PSO) algorithm to encode and optimize discrete variables, with the goal of minimizing the normalized mean error, search in the solution space, and obtain preliminary model configuration parameter setting values.

[0088] Furthermore, the goal of minimizing the normalized mean error is as follows:

[0089] ;

[0090] Where: min(NMBE) represents the minimization of the mean error; s(t) represents the simulation value at time t; m(t) represents the measured value at time t; t0 represents the starting time; Indicates the calibration period.

[0091] Furthermore, in S52, the particle swarm optimization PSO algorithm is used, such as Figure 2 As shown, including:

[0092] S521, initializing the particle swarm, and randomly assigning a position and a velocity to each particle;

[0093] S522, calculating the fitness value of each particle;

[0094] S523, updating the individual extreme value of each particle;

[0095] S524, updating the global extreme value;

[0096] S525, updating the speed and position of each particle according to the speed update formula and the position update formula;

[0097] The speed update formula is:

[0098] ;

[0099] In the formula, is the velocity of the i-th particle at the k+1th iteration; is the velocity of the i-th particle at the k-th iteration; is the inertia weight, which controls the influence of the previous iteration of the particle velocity; and are learning factors, which control the response degree of particles to individual extreme values ​​and global extreme values ​​respectively; and is a random number in the range [0,1]; is the optimal solution found by the i-th particle so far; The optimal solution found by the entire particle swarm so far;

[0100] The position update formula is:

[0101] ;

[0102] In the formula, x i (k+1) is the position of the i-th particle at the k+1th iteration; x i (k) is the position of the i-th particle at the k-th iteration.

[0103] S526. Repeat steps S522-S525 until the maximum number of iterations is reached.

[0104] Through the particle swarm optimization PSO algorithm, a solution close to the global optimal solution is quickly searched in the solution space, and a set of better solution vectors or better model configuration parameter settings are obtained. Then the generalized pattern search GPS algorithm stage is entered to further optimize the accuracy of the solution by searching and gradually adjusting the neighborhood of the solution.

[0105] S53, after obtaining the initial model configuration parameter setting values ​​in the PSO stage, enter the generalized mode search GPS algorithm stage to further optimize the accuracy of the solution until the normalized mean error of the calibration target reaches the minimum output value, and the iterative process ends.

[0106] Further, in S53, the generalized mode searches the GPS algorithm, such as Figure 3 As shown, specifically including:

[0107] S531, selecting an initial design point and an initial step size;

[0108] S532, according to the initial design point and the initial step length, calculate the value of the objective function at the current position and in the preset neighborhood, and determine the moving direction for the next step;

[0109] S533, adjusting the step size and updating the design point according to the change of the objective function;

[0110] S534, check whether the convergence condition is met, if so, stop the algorithm; otherwise, return to step S532.

[0111] S54, outputting the identification parameters that minimize the normalized mean error. When the GPS algorithm ends, the output parameters are the optimal parameters after the two-stage optimization. These parameters are the parameter set that can minimize the normalized mean error and represent the optimal HVAC system model.

[0112] S6. Use the validation data set to verify the effectiveness of parameter identification, including:

[0113] Using Python to run the HVAC model and collect simulation result data under the constructed boundary data in the validation dataset;

[0114] Calculate the root mean square error between the characterization accuracy data and the simulation result data in the validation data set, and combine the root mean square error coefficient of variation and determination coefficient to determine the effectiveness of parameter identification;

[0115] The specific calculation formula of the root mean square error (RMSE) is as follows:

[0116] ;

[0117] Where: s i represents the simulation value; m i represents the measured value; n represents the total number of sample points.

[0118] The specific calculation formula of the coefficient of variation of the root mean square error (CVRMSE) is as follows:

[0119] ;

[0120] Where: Indicates the average value of the measured value; s i represents the simulation value; m i represents the measured value; n represents the total number of sample points.

[0121] Coefficient of determination R 2 The specific calculation formula is as follows:

[0122] ;

[0123] Where: s i represents the simulation value; m i represents the measured value; n represents the total number of sample points; Indicates the average of the measured values; Represents the average of the simulation values.

[0124] The root mean square error (RMSE) reflects the difference between the actual temperature and the simulated temperature. A lower RMSE indicates that the difference between the simulation result and the actual result is smaller. Generally, the smaller the RMSE value, the better the prediction effect of the model. The root mean square error coefficient of variation (CVRMSE) reflects the difference between the actual power and the simulated power. The smaller the CVRMSE, the lower the proportion of the error to the actual value, and the stronger the prediction ability of the model. Generally, if the CVRMSE is less than 10%, the model is considered to have a good fit. 2 As an indicator for evaluating the similarity between the simulation result curve and the actual result curve, R 2A value close to 1 indicates a strong linear relationship between the simulation results and the actual results. Usually, R 2 A value above 0.8 indicates that the model has a good fit.

[0125] The Particle Swarm Optimization (PSO) algorithm gradually guides the particle swarm to the global optimal solution by iteratively updating the speed and position of the particles. The algorithm relies on information sharing and collaboration between particles to achieve effective search and optimization in the solution space. In addition, the PSO algorithm has strong parallel computing characteristics because the updates of particles are independent of each other and can be performed simultaneously, which has certain advantages when solving large-scale problems. However, the PSO algorithm may face the problem of premature convergence, that is, it stagnates near the local optimal solution and it is difficult to further find the global optimal solution, resulting in poor convergence performance.

[0126] The generalized pattern search algorithm (GPS) is a direct search method that does not require information about the derivative of the objective function. The algorithm gradually approaches the optimal solution by searching for a predefined pattern (i.e., step size), and evaluates only based on the size of the objective function value to determine the fastest descent direction. Since it does not rely on the derivative of the objective function, the GPS algorithm is particularly suitable for situations where the objective function is not differentiable or difficult to differentiate. It has high computational efficiency and strong convergence, and can find a better solution in a limited time. Therefore, the GPS algorithm shows significant advantages in time-consuming optimization problems that require reducing the number of simulations.

[0127] The GPS-PSO algorithm that combines the GPS algorithm with the PSO algorithm in this application is a two-stage optimization method that comprehensively utilizes the advantages of the two algorithms. Its core idea is to solve the discrete variable optimization problem with the help of the global search capability of PSO, while improving the quality of the solution through the fine adjustment capability and convergence of GPS. Specifically: PSO stage: First, use the particle swarm algorithm to encode and optimize discrete variables, and use information sharing and collaboration between particles to quickly search in the solution space to approach the global optimal solution. GPS stage: After obtaining the better solution generated in the PSO stage, enter the GPS stage. This stage further optimizes the accuracy of the solution by searching and gradually adjusting the neighborhood of the solution, enhancing the convergence of the algorithm and approaching the optimal solution.

[0128] Through this staged optimization strategy, the GPS-PSO algorithm can not only effectively solve the discrete variable optimization problem, but also significantly improve the quality and convergence of the final solution.

[0129] For example, taking a chiller as an example, a chiller model based on the Carnot cycle calculates the actual COP of the chiller under a certain working condition by calculating the partial load rate correction coefficient and the nominal COP value, and calculates the heat exchange between the evaporator and the condenser based on the actual COP and the electric power, thereby calculating the outlet temperature of the medium. Its main mechanism is described as follows:

[0130] The calculation of heat transfer is:

[0131] Q Con =PQ Eva ;Q Eva = -P × COP;

[0132] Where: Q Con represents the heat exchange of the condenser, W; P represents the electric power of the compressor, W; Q Eva Indicates the evaporator heat exchange rate, W; COP indicates the actual COP.

[0133] Compressor part load factor calculation:

[0134] ;

[0135] Where: y represents the partial load rate of the compressor; P represents the electrical power of the compressor, W; P nominal Indicates the nominal electrical power of the compressor, W.

[0136] The actual COP of the chiller is calculated as: ;

[0137] Where: represents the nominal COP; Indicates the partial load correction factor.

[0138] The calculation of the part load correction factor is: ;

[0139] Where: y PL Indicates the part load rate; Represents the modified polynomial coefficients.

[0140] Through step S1 and step S2, the model parameters to be identified in this model are determined as follows: , and P nominal , first based on the least squares fitting , then identify and P nominal order.

[0141] Through step S3, the actual operation data finally collected is determined, as shown in Table 1.

[0142] Table 1 Actual operation data of chiller

[0143]

[0144] And the data is preprocessed in the following ways;

[0145] The quartile criterion is used to detect outliers. The definitions of the lower and upper quartile limits are as follows:

[0146] LT=Q1-1.5×(Q3-Q1);UT=Q3+1.5×(Q3-Q1);

[0147] Where: LT represents the lower limit of the interquartile range; UT represents the upper limit of the interquartile range; Q1 represents the first quartile; Q3 represents the third quartile.

[0148] Use a simple moving average method to fill missing values, as shown below:

[0149] ;

[0150] Where: m represents the window size; SMA t represents the moving average at time t.

[0151] Then, based on the actual operation data after preprocessing, the GPS-PSO two-stage optimization algorithm is used to identify the model parameters. , and determined according to nameplate parameters and sample data The optimization range is [4,8], P nominal The optimization range is [100,250]kW.

[0152] Set the initial values ​​of the model parameters to be identified, use the measured data as the boundary input, use Python to run the model, and simulate the simulation results under this boundary; use the measured values ​​and simulation results to calculate the normalized mean error NMBE; the optimization algorithm updates the model parameters with the goal of reducing the normalized mean error NMBE, and outputs the new set values. Repeat the above steps, and continuously simulate and calculate with the normalized mean error of the calibration target as the optimization goal, until the normalized mean error of the calibration target reaches the minimum output value, end the iterative process, and output the identification parameter setting value that minimizes the normalized mean error.

[0153] The effectiveness of parameter identification is verified by verifying the data set.

[0154] like Figure 4 As shown in the figure, the performance curve fitting results of the chiller are shown, and the relationship between the COP and PLR of the chiller is shown. By fitting the performance curve of the chiller, it can be seen that the curve of the simulation data (simulation results) and the actual data fits well, proving the accuracy of the model after parameter identification. Figure 5As shown in the figure, the comparison between the simulation results and the measured data of the cold water supply temperature is shown. The simulated temperature is close to the measured temperature, which shows that the model can better reflect the operation of the actual system. Figure 6 As shown in the figure, the cooling water return temperature refers to the temperature of the cooling water returning to the chiller after passing through the air conditioning cooling system. The difference between the simulated temperature and the measured data is very small, which verifies the reliability of the model. Figure 7 As shown, the comparison between the simulation results of system power consumption and the actual measured data is shown. The simulated power is highly consistent with the actual power, which shows that the model can effectively reflect the actual energy consumption characteristics of the chiller.

[0155] This application uses the PSO-GPS two-stage optimization algorithm to efficiently identify HVAC system parameters and minimize the normalized mean error. The PSO stage is responsible for quickly searching for preliminary solutions, while the GPS stage further improves the accuracy of the solution through refinement and optimization. The optimal parameter set outputted in the end can accurately describe the behavior of the system and provide a reliable system model.

[0156] It should be noted that the terms used in the present invention are only for describing specific embodiments, rather than limiting the scope of the present application. As shown in the present specification, unless the context clearly indicates an exception, the words "one", "a", "a kind of" and / or "the" do not specifically refer to the singular, but may also include the plural. The terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of more restrictions, the elements defined by the sentence "include one..." do not exclude the presence of other identical elements in the process, method or device including the elements.

[0157] It should also be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. Unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", etc. should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A HVAC model parameter identification method based on GPS-PSO algorithm, characterized in that: The steps include: S1. Determine the key parameters that affect the accuracy of the HVAC model based on expert knowledge and association rules; S2. Perform sensitivity analysis on the key parameters and sort them from high to low according to sensitivity to construct a set of parameters to be identified for the HVAC model; S3, collecting actual operation data according to the parameter set to be identified of the HVAC model, and performing preprocessing; S4, dividing the preprocessed actual operation data into a calibration data set and a verification data set according to a preset ratio; S5. Based on the calibration data set, the normalized mean error is used as the optimization target, and the GPS-PSO two-stage optimization algorithm is used for parameter identification; Specifically include: S51, according to the order of the sensitivity of the parameters to be identified of the HVAC model from high to low, the parameters to be identified are set as optimization variables in sequence, and the optimization range of the optimization variables is determined according to the nameplate parameters and the calibration data; S52, using a particle swarm optimization (PSO) algorithm to encode and optimize discrete variables, with the goal of minimizing the normalized mean error, searching in the solution space, and obtaining preliminary model configuration parameter setting values; S53, after obtaining the initial model configuration parameter setting values ​​in the PSO stage, enter the generalized mode search GPS algorithm stage to further optimize the accuracy of the solution until the normalized mean error of the calibration target reaches the minimum output value, and end the iterative process; S54, outputting identification parameters that minimize the normalized mean error; S6. Use the validation data set to verify the effectiveness of parameter identification.

2. The HVAC model parameter identification method based on GPS-PSO algorithm according to claim 1 is characterized in that: In S2, the key parameters are subjected to sensitivity analysis and ranked from high to low according to sensitivity, including: S21. Perform a single parameter change experiment, change the set value of a key parameter alone, fix other parameters, run the HVAC model and record the output results; S22, calculating the standard deviation and mean square error of the output results caused by the change of a single parameter; S23. Sort the sensitivity of key parameters from high to low according to standard deviation and mean square error.

3. The HVAC model parameter identification method based on GPS-PSO algorithm according to claim 1 is characterized in that: In S3, according to the set of parameters to be identified of the HVAC model, actual operation data is collected, including: S31, arranging the minimum data set required for the operation of the HVAC model according to the parameter set to be identified of the HVAC model; S32. Classify the data in the data set into construction boundary data and characterization accuracy data according to different uses of the data, wherein the construction boundary data is used to define the limit conditions of the model working range, and the characterization accuracy data is used to improve the prediction accuracy of the model; S33. Determine the cycle, frequency and scope of data collection according to the collection requirements of different data types, and collect actual operation data.

4. The HVAC model parameter identification method based on GPS-PSO algorithm according to claim 1 is characterized in that: In S3, the actual operation data is preprocessed, including: Use the Laida criterion and the quartile criterion to identify and delete outliers in the actual operating data; Check for missing values ​​in actual operational data; Missing values ​​are filled using the mean method, moving average method, or similarity algorithm.

5. The HVAC model parameter identification method based on GPS-PSO algorithm according to claim 1 is characterized in that: In S52, the particle swarm optimization PSO algorithm includes: S521, initializing the particle swarm, and randomly assigning a position and a velocity to each particle; S522, calculating the fitness value of each particle; S523, updating the individual extreme value of each particle; S524, updating the global extreme value; S525, updating the speed and position of each particle according to the speed update formula and the position update formula; S526. Repeat steps S522-S525 until the maximum number of iterations is reached.

6. The HVAC model parameter identification method based on GPS-PSO algorithm according to claim 5 is characterized in that: The speed update formula is: ； In the formula, is the velocity of the i-th particle at the k+1th iteration; is the velocity of the i-th particle at the k-th iteration; is the inertia weight, which controls the influence of the previous iteration of the particle velocity; and are learning factors, which control the response degree of particles to individual extreme values ​​and global extreme values ​​respectively; and is a random number in the range [0,1]; is the optimal solution found by the i-th particle so far; The optimal solution found by the entire particle swarm so far; The position update formula is: ； In the formula, x i (k+1) is the position of the i-th particle at the k+1th iteration; x i (k) is the position of the i-th particle at the k-th iteration.

7. The HVAC model parameter identification method based on GPS-PSO algorithm according to claim 1 is characterized in that: In S53, the generalized mode search GPS algorithm specifically includes: S531, selecting an initial design point and an initial step size; S532, according to the initial design point and the initial step length, calculate the value of the objective function at the current position and in the preset neighborhood, and determine the moving direction for the next step; S533, adjusting the step size and updating the design point according to the change of the objective function; S534, check whether the convergence condition is met, if so, stop the algorithm; otherwise, return to step S532.

8. The HVAC model parameter identification method based on GPS-PSO algorithm according to claim 1 is characterized in that: The S6 specifically includes: Using Python to run the HVAC model and collect simulation result data under the constructed boundary data in the validation dataset; The root mean square error between the characterization accuracy data and the simulation result data in the validation data set was calculated, and the validity of the parameter identification was judged by combining the root mean square error coefficient of variation and the coefficient of determination.

Images