A method for determining dynamic benchmark of heating / cooling capacity of HVAC system

Abstract

The present invention relates to a method for determining a dynamic benchmark for heat / cooling consumption of a heating, ventilation, and air-conditioning system. The method comprises the following steps: establishing a heat (cooling) consumption classification model for processed operating data of the heating, ventilation, and air-conditioning system; performing data processing before association rule mining using the K-means clustering discretization method and the equal-width discretization method; mining association rules using the Apriori algorithm and processing the mined rules; and calculating the dynamic benchmark for heat (cooling) consumption of each energy station or thermal power station based on the mined association rules. Compared with existing methods, the present invention takes the energy station or thermal power station of the heating, ventilation, and air-conditioning system as the object, mines and integrates multi-factor features to establish a reasonable dynamic energy consumption benchmark for energy stations or thermal power stations with different conditions and levels. Compared with traditional mechanism models, the calculation is simpler, the time and cost are lower, and it provides a quantitative assessment basis and energy-saving direction for energy conservation and carbon reduction of the heating, ventilation, and air-conditioning system.

Description

Technical Field

[0001] This invention relates to a method for determining the dynamic baseline of heat / cooling capacity of HVAC systems based on decision tree classification and association rule mining. Background Technology

[0002] Heating, ventilation, and air conditioning (HVAC) systems are critical energy-consuming systems in buildings, directly impacting energy consumption and carbon emissions. The heat (cooling) consumption of HVAC systems is influenced by various factors, including outdoor air temperature, indoor load demand, equipment operating status, and system control strategies, and changes dynamically with these factors. Therefore, mining existing HVAC system operating data and establishing a reasonable dynamic heat (cooling) consumption benchmark is a crucial foundation for reducing building energy consumption and achieving energy-efficient operation.

[0003] Currently, there are numerous methods for determining the benchmark for heat (cooling) consumption of HVAC systems, but these methods have certain problems and shortcomings: First, most existing methods are based on mechanistic models, which are highly dependent on the input characteristics of the models. The uncertainty of the input data affects the reliability of the simulation results, and the modeling process is complex, time-consuming, and costly. Second, most existing research focuses on building energy consumption benchmarks, with little research on the heat (cooling) consumption benchmarks of HVAC system energy stations or heating stations. Third, currently recognized energy consumption benchmarks for HVAC systems in standards and specifications are mostly regional static benchmarks, which cannot effectively guide the operation and control of heating stations.

[0004] The present invention provides a method for determining the dynamic benchmark of heat (cooling) consumption in HVAC systems based on decision tree classification and association rule mining. It screens characteristic factors that affect the energy consumption of energy stations or heating stations in HVAC systems, establishes a heat (cooling) consumption classification model, and establishes a dynamic benchmark model of heat (cooling) consumption in HVAC systems based on association rule mining according to the classification results. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the technical problem to be solved by this invention is to provide a method for determining the dynamic benchmark of heat (cooling) consumption in a heating, ventilation, and air conditioning (HVAC) system. This method uses historical operating data of the HVAC system to determine the dynamic benchmark of heat (cooling) consumption of the energy station or heating station of the HVAC system based on decision tree classification and association rule mining, providing a reliable reference for the energy-saving and emission-reduction operation of the HVAC system.

[0006] The technical solution adopted in this invention is as follows:

[0007] A method for determining the dynamic baseline of heat / cooling capacity in a heating, ventilation, and air conditioning system, comprising the following steps:

[0008] Step 1: Feature Selection

[0009] Data on HVAC systems under different operating conditions were obtained. After removing outlier data, Person correlation coefficients were calculated between each energy consumption feature and heat (cooling) consumption. Features with Person correlation coefficients greater than or equal to 0.24 were retained. At the same time, the ReliefF algorithm was used to calculate the importance index of heat (cooling) consumption, and features with importance indices greater than or equal to 0.003 were selected.

[0010] After merging the Person correlation coefficient and the features selected by the ReliefF algorithm, the random forest embedding method was used to select features with a random forest importance index greater than 0.3 as the main features affecting energy consumption.

[0011] Step 2: Construct a heat (cold) consumption classification model:

[0012] The heat (cold) consumption classification model is based on a decision tree model. The nodes of the decision tree include a root node, leaf nodes, and decision nodes. The root node of the decision tree is the set of data from all heating stations. The decision nodes correspond to the energy consumption feature division rules. The decision nodes divide the data at the root node into several subsets according to the energy consumption feature rules. The leaf nodes are the final energy consumption classification nodes. The heating stations within the leaf nodes are classified into the same category to achieve the heat consumption classification of the heating stations.

[0013] The input features of the decision tree model are the main energy consumption features and corresponding heat consumption determined in step 1. Bayesian optimization is used to find the optimal parameters of the decision tree model. At the same time, the wrapping method is used to fully search the features to determine the best combination of decision tree regression features with a score of over 75%.

[0014] Next, the decision tree model is pruned using the cost complexity post-pruning method to remove unnecessary nodes, thereby obtaining the heat (cold) consumption classification model and the final input features of the model.

[0015] Step 3: Using the heat (cooling) consumption classification model constructed in Step 2, and taking the finally determined input features as input, classify the HVAC system data;

[0016] Step 4: For each type of HVAC system data, recalculate the Person correlation coefficient between each energy consumption feature and heat (cooling) consumption. Use all energy consumption features with a Person correlation coefficient greater than 0.7 as the rule front term for association rule mining, and use heat (cooling) consumption as the rule back term.

[0017] Step 5: Discretize the HVAC system data. After discretization, use the Apriori algorithm to mine association rules according to the rule antecedent and rule consequent determined in Step 4 to obtain the association rules for all classes.

[0018] Step 6: Process the association rules of all classes obtained in Step 5 to obtain usable rules, and use the usable rules to determine the dynamic baseline of the heat (cooling) consumption of the HVAC system.

[0019] Furthermore, the rule processing procedure is as follows:

[0020] The outdoor temperature is discretized with equal width to obtain different outdoor temperature ranges. All the class association rules obtained in step 5 are grouped according to whether they are in the same outdoor temperature range, and the confidence level of the rules in each group is determined.

[0021] Set a minimum confidence level, and extract all rules with a confidence level greater than the minimum confidence level for expert analysis;

[0022] By analyzing the rules using knowledge from the heating sector, unreasonable rules are replaced with reasonable rules that have a lower confidence level than their neighbors.

[0023] Furthermore, the process of determining the dynamic benchmark for dynamic heat (cooling) consumption is as follows: statistically analyze the operating conditions and available rules for heat consumption under all outdoor temperature ranges, and calculate the dynamic heat (cooling) consumption benchmark 1 and benchmark 2 of the heating station based on the upper and lower bounds of the heat (cooling) consumption range after the rule.

[0024] Furthermore, in step 5, the discretization process is as follows: the flow rate, pressure, outdoor air temperature, and average temperature parameters of the supply and return water in the primary network of the HVAC system are discretized: K-means clustering is used to discretize the operating data and energy consumption data, and equal-width discretization is used to discretize the indoor and outdoor air temperatures.

[0025] Furthermore, in the post-pruning method using cost complexity to prune the decision tree model, random 5-fold cross-training is used for model validation, and the mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²) are calculated for 5 training and test sets. 2 The mean of the explained variance Evar, when the mean squared error (MSE) and mean absolute error (MAE) of the model are less than 0.05, and the coefficient of determination R0 is... 2 If the explained variance Evar is greater than 75%, the model is considered to have passed the test and can be used for heat (cold) consumption classification.

[0026] Furthermore, the HVAC system is a heat station or energy station.

[0027] Furthermore, in step 4, the preceding terms of the association rule mining are the primary flow rate of the energy station or heating station, the outdoor temperature, the average temperature of the secondary supply and return water, and the indoor temperature of a typical user.

[0028] Furthermore, in the Apriori algorithm, the minimum support is set to 0.002, the minimum confidence is 60%, and the minimum lift is 1.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] The most prominent and essential feature of this invention is:

[0031] 1. This invention relies on historical data to determine the dynamic benchmark for heat (cooling) consumption of energy stations or heating stations in HVAC systems, without depending on complex mechanistic models, thus reducing time and cost. It uses a decision tree classification-association rule mining method to establish the dynamic benchmark for heat (cooling) consumption of HVAC systems. This invention can mine and integrate multiple factors to establish reasonable dynamic energy consumption benchmarks for energy stations or heating stations under different conditions and levels in numerous HVAC systems.

[0032] 2. Based on the feature selection results, this invention uses Bayesian optimization and cost complexity reduction followed by pruning to refine the model and establish a heat (cooling) consumption classification model for HVAC systems, providing a foundation for the establishment of dynamic energy consumption benchmark models for various types of energy stations or heating stations.

[0033] 3. This invention utilizes association rule mining algorithms to mine dynamic benchmarks for the heat (cooling) consumption of HVAC systems in various energy stations or heating stations. Association rule mining is a commonly used descriptive statistical method that can extract hidden rules from large amounts of data. It overcomes the shortcomings of traditional mechanistic models, which are affected by historical data quality, operating modes, abnormal operating conditions, and human experience adjustments. It has strong interpretability and can be used for HVAC system control optimization. Effective association rules (usable rules) established through association rule mining can reasonably determine the dynamic benchmarks for the heat (cooling) consumption of energy stations or heating stations, providing a reliable foundation for optimizing system operation and achieving energy conservation and carbon reduction.

[0034] The method for determining the dynamic benchmark of heat (cooling) consumption in this invention provides a quantitative assessment basis and energy-saving operation direction for energy stations or heating stations of various types in HVAC systems, and provides a prerequisite for calculating the carbon reduction rate and assessing the carbon reduction potential of the system.

[0035] Based on the selected energy consumption characteristics, a classification model for the heat (cooling) consumption of energy stations or heating stations in HVAC systems is established. Association rules for HVAC systems are then mined, and dynamic benchmarks for heat (cooling) consumption are calculated based on these rules. This method uses the flow rate and pressure of the primary network, outdoor air temperature, average supply and return water temperatures of the secondary network, and user indoor temperatures as independent variables to determine the dynamic benchmarks for heat (cooling) consumption. Attached Figure Description

[0036] Figure 1This is a flowchart of a method for determining the dynamic benchmark of heat consumption in HVAC systems based on decision tree classification and association rule mining, according to the present invention.

[0037] Figure 2 This is a flowchart illustrating the feature selection process for the present invention.

[0038] Figure 3 The graph shows the results of the Person correlation coefficient calculation for the heat consumption characteristics of this invention.

[0039] Figure 4 This is a graph showing the calculation results of the importance index of the heat consumption feature ReliefF algorithm in this invention.

[0040] Figure 5 The result is the calculation result of the importance index of the random forest feature selection for the heat consumption feature involved in this invention.

[0041] Figure 6 This is an example diagram of the decision tree structure of this invention.

[0042] Figure 7 This is a schematic diagram of the Bayesian optimization process of the present invention.

[0043] Figure 8 This is a schematic diagram of the cost complexity pruning process of the present invention.

[0044] Figure 9 This is a schematic diagram illustrating the principle of the 5-fold cross-validation of this invention.

[0045] Figure 10 This is a flowchart of the post-processing of the decision tree regression heat consumption classification model of this invention.

[0046] Figure 11-13 This is a box plot showing the heat consumption, flow rate, and building comprehensive physical property coefficient of various types of heating stations according to the present invention.

[0047] Figure 14 , Figure 15 These figures represent the dynamic heat consumption benchmarks for floor radiant heating and radiator heating stations according to embodiments of the present invention. The black curve in the figure represents the outdoor temperature during the heating season, while the other curves of different colors represent the dynamic heat consumption benchmarks for each type of floor radiant heating station throughout the entire heating season. Detailed Implementation

[0048] The present invention will be further explained below with reference to the embodiments and accompanying drawings. The accompanying drawings are simplified schematic diagrams, but are not intended to limit the scope of protection of this application.

[0049] The present invention provides a method for determining the dynamic baseline of heat (cooling) consumption in a heating, ventilation, and air conditioning system (see [reference]). Figure 1 The method includes the following steps:

[0050] Step S1: Use filtering and embedding methods to initially select the main energy consumption characteristics that affect the amount of heat (cooling) consumption from the HVAC system data;

[0051] Data on HVAC systems under different operating conditions were obtained. After removing outlier data, Person correlation coefficients were calculated between each energy consumption feature and heat (cooling) consumption. Features with Person correlation coefficients greater than or equal to 0.24 were retained. At the same time, the ReliefF algorithm was used to calculate the importance index of heat (cooling) consumption, and features with importance indexes greater than or equal to 0.003 were selected.

[0052] After merging the Person correlation coefficient and the features selected by the ReliefF algorithm, the random forest embedding method was used to select features with a random forest importance index greater than 0.3 as the main features affecting energy consumption.

[0053] Step S2: Construct a heat (cold) consumption classification model:

[0054] The heat (cold) consumption classification model is based on a decision tree model. The nodes of the decision tree include a root node, leaf nodes, and decision nodes. The root node of the decision tree is the set of data from all heating stations. The decision nodes correspond to the energy consumption feature division rules. The decision nodes divide the data at the root node into several subsets according to the energy consumption feature rules. The leaf nodes are the final energy consumption classification nodes. The heating stations within the leaf nodes are classified into the same category to achieve the heat consumption classification of the heating stations.

[0055] The input features of the decision tree model are the main features that affect energy consumption and the corresponding heat consumption. Bayesian optimization is used to find the optimal parameters of the decision tree model. At the same time, the wrapping method is used to fully search the features to determine the best combination of decision tree regression features with a score of over 75%.

[0056] Next, the decision tree model is pruned using the cost complexity post-pruning method to remove unnecessary nodes, thereby obtaining the heat (cold) consumption classification model and the final input features of the model.

[0057] By employing Bayesian optimization and cost-complexity post-pruning, the generalization ability of the model can be improved, resulting in a simpler and more efficient decision tree. In cost-complexity post-pruning, when the mean squared error (MSE) and mean absolute error (MAE) of the regression model are less than 0.05, the coefficient of determination R0 is... 2 If the explained variance Evar is greater than 75%, the model is considered to have passed the test.

[0058] Step S3: Using the heat (cooling) consumption classification model constructed in step S2, and taking the finally determined input features as input, classify the HVAC system data;

[0059] Step S4: For each type of HVAC system data, recalculate the Person correlation coefficient between each energy consumption feature and heat (cooling) consumption. Use all energy consumption features with a Person correlation coefficient greater than 0.7 as the rule front term for association rule mining, and use heat (cooling) consumption as the rule back term.

[0060] Step S5: Discretize the HVAC system data. After discretization, use the Apriori algorithm to mine association rules according to the rule antecedent and rule consequent determined in Step 4 to obtain the association rules for all classes.

[0061] When determining the antecedent and consequent of association rules for each type of energy station or heating station, the data used is the original data after removing outliers. The data used for rule mining is the discretized data.

[0062] Step S6: Process the association rules of all classes obtained in Step S5 to obtain usable rules, and use the usable rules to determine the dynamic baseline of heat (cooling) consumption of the HVAC system.

[0063] In step S5, the discretization process involves discretizing the primary network flow rate, pressure, outdoor air temperature, and secondary network supply and return water temperature parameters of the HVAC system. K-means clustering is used for discretization of the operational and energy consumption data, while equal-width discretization is used for indoor and outdoor air temperatures. This process obtains the discrete attribute data needed for association rule mining, which serves as the antecedent and consequent of the rules.

[0064] The rule processing procedure is as follows:

[0065] The outdoor temperature is discretized with equal width to obtain different outdoor temperature ranges. All class association rules obtained in step S5 are grouped according to whether they are in the same outdoor temperature range, and the confidence level of the rules in each group is determined.

[0066] Set a minimum confidence level, and extract all rules with a confidence level greater than the minimum confidence level for expert analysis;

[0067] By analyzing the rules using knowledge from the heating sector, unreasonable rules are replaced with reasonable rules that have a lower confidence level than their neighbors.

[0068] In practice, after determining the baseline, the range of outdoor temperature of the heating station is used to determine the baseline range of its heat (cooling) consumption.

[0069] Preferably, the minimum confidence level is set to 60%.

[0070] The Apriori algorithm is used to discover frequent itemsets. The algorithm sets thresholds for minimum support, minimum confidence, and minimum lift, and then uses these thresholds to discover mining rules.

[0071] Example 1

[0072] This embodiment uses the operational status of a typical HVAC system in a cold region as the identification object. This HVAC system comprises 169 heating stations, with an actual heating area of ​​approximately 8.49 million square meters. Historical heat source operation data of the system was collected, including meteorological parameters, unit information, operational data, and indoor data of typical users under the units. Of the 169 heating units, 63 units serve buildings with floor radiant heating, and 106 units serve buildings with radiator heating.

[0073] This embodiment of the method for determining the dynamic baseline of heat consumption in a centralized HVAC system based on decision tree classification and association rule mining includes the following steps:

[0074] Step S1: Use filtering and embedding methods to select energy consumption characteristics related to the dynamic benchmark of heat consumption of the heating station, and screen out the main characteristics that affect energy consumption.

[0075] This invention combines filtering and embedding methods, which can quickly screen and accurately optimize features through complementary advantages, improve the accuracy and efficiency of model prediction, reduce the risk of overfitting, and thus provide a more robust feature selection scheme for establishing dynamic benchmarks for heat consumption at heating stations.

[0076] See feature selection process Figure 2 First, a filtering method is used: 1) the Pearson correlation coefficient between each energy consumption characteristic and heat consumption is calculated; 2) simultaneously, the ReliefF algorithm is used to calculate the importance index of heat (cold) consumption to evaluate the nonlinear correlation of each energy consumption characteristic. This invention implements the ReliefF algorithm using the "relieff" function in MATLAB R2021. The "regression" parameter is set to solve the regression problem, and the maximum number of iterations is set to 1000, with an error threshold of 10. -6 With a learning rate of 0.01, feature selection using the ReliefF algorithm was performed. The energy consumption characteristic symbols representing the factors influencing heat consumption in the heating station are shown in Table 1.

[0077] Table 1. Factors affecting heat consumption and energy consumption characteristics

[0078]

[0079] Figure 3 This is a graph showing the calculated Person correlation coefficient between energy consumption characteristics and heat consumption in this embodiment. Figure 3 It can be seen that the flow rate G and the rate of change of the secondary supply and return water temperature std t2The correlation coefficient between heat consumption and energy consumption is 0.34, which is the strongest among all energy consumption characteristics. The absolute values ​​of the correlation coefficients between indoor temperature and the rate of change of room temperature and heat consumption are around 0.1, which is the weakest among all characteristics. This embodiment selects characteristics with a Person correlation coefficient greater than or equal to 0.24, namely water consumption W, flow rate G, and the rate of change of secondary supply and return water temperature std. t2 The average temperature t2 of the secondary supply and return water and the temperature difference Δt2 between the secondary supply and return water are used as candidate features.

[0080] Figure 4 This is a graph showing the calculation results of the importance index of the ReliefF algorithm in this embodiment. (Source: [Insert graph here]) Figure 4 It can be seen that the importance index of the flow rate G is 0.06, which is much higher than the other characteristics, indicating that the flow rate index has a significant impact on the heat consumption index. The characteristics selected with an importance index greater than or equal to 0.003 are water consumption W, flow rate G, heating area A, building comprehensive physical property coefficient Bc, and secondary supply and return water temperature change rate std. t2 As candidate features.

[0081] Based on the combined results of Person coefficient and ReliefF feature filtering, the following parameters were initially filtered out: water consumption W, heating area A, building comprehensive physical property coefficient Bc, and secondary supply and return water temperature change rate std. t2 The system uses seven features: flow rate G, average temperature t2 of secondary supply and return water, and temperature difference Δt2 between secondary supply and return water. Then, it employs a random forest embedding method to select features from these seven features.

[0082] Figure 5 This is a graph showing the calculation results of the importance of energy consumption characteristics using the random forest embedding method in this embodiment. Figure 5 As shown, flow rate characteristics have the greatest impact on heat consumption, with five features having an importance index exceeding 0.3 in the random forest. After screening, the following factors were ultimately selected: heating area A, building composite physical property coefficient Bc, flow rate index G, and secondary supply and return water temperature change rate std. t2 The five characteristics of the average temperature of the secondary supply and return water (t2) are the main factors affecting energy consumption.

[0083] Step 2: Establish a heat consumption classification model based on a decision tree according to the five main energy consumption characteristics selected:

[0084] Decision trees are a common machine learning algorithm primarily used for classification and regression tasks. They work by segmenting data using a series of decision rules to ultimately classify or predict the data.

[0085] Figure 6Taking the heat consumption classification model of a heating station as an example, the decision tree diagram is shown. The nodes of the decision tree include a root node, leaf nodes, and decision nodes. For the heat consumption classification problem of the heating station in this invention, the root node of the decision tree is the set of all heating station data; in the diagram, the root node includes four data points D1-D4. The decision nodes correspond to the energy consumption feature partitioning rules, dividing the data at the root node into several subsets according to the energy consumption feature rules; for example, decision node G>1 divides D1 and D2 into one subset, and D3 and D4 into another subset. The leaf nodes are the final energy consumption classification nodes. The heating stations within the leaf nodes are classified into the same category, thus achieving the heat consumption classification of the heating stations; the example in the diagram divides the heating station data into three leaf nodes, classifying the heating stations into three categories.

[0086] Bayesian optimization and cost-complexity post-pruning are two different optimization methods. Using only Bayesian optimization may not effectively reduce the complexity of the constructed model, while using only cost-complexity post-pruning may lack precise searching for optimal parameters, resulting in suboptimal model performance. This invention establishes a decision tree-based heat consumption classification model based on a two-step optimization approach using Bayesian optimization and cost-complexity post-pruning, and selects a typical heat station for subsequent benchmark establishment. The two-step optimization method can accurately select parameters, enhance the model's generalization ability, significantly reduce manual intervention, and improve overall training efficiency.

[0087] Figure 7 This is a schematic diagram of the Bayesian optimization process involved in this invention.

[0088] by Figure 7 For example, Bayesian optimization mainly includes 5 steps:

[0089] Step 1. Define the objective function. This invention selects the decision tree regression accuracy score as the objective function. In the example figure, the vertical axis of the objective function represents the decision tree regression accuracy score.

[0090] Step 2. Define the parameter space. For decision tree regression, the parameter space includes the splitting criterion, maximum tree depth, minimum number of samples for internal node re-splits, minimum number of samples for leaf nodes, splitting strategy, and weight parameters. The example in the figure uses the splitting criterion and the maximum tree depth of decision tree regression as the parameter space. The red and black curves represent the mean absolute error and mean squared error of the splitting criterion, respectively; the horizontal axis represents the maximum tree depth, and the curves represent the accuracy scores obtained for different splitting criteria and different maximum tree depths. The curves in the figure are black-box functions; the maximum tree depth and splitting criterion corresponding to the highest score need to be determined through model optimization.

[0091] Step 3. Using a Gaussian mixture model as a surrogate model can effectively fit complex data distributions and perform probability density estimation and soft clustering.

[0092] Step 4. Explore the optimal parameter points and evaluate the objective function. The pink squares in the example diagram represent the initial sampling points. After sampling, calculate the accuracy score of the sampling points and compare the score with the target score; if the target score cannot be achieved, resample according to the surrogate function.

[0093] Step 5. Obtain the optimal solution. Using the new parameter values ​​and corresponding accuracy scores, update the surrogate model, iterating until the preset convergence condition is met to obtain the parameter combination with the best accuracy. In the example shown in the figure, the target score is 80%, and after four iterations, the preset target is met, as indicated by the red pentagram.

[0094] Figure 8 This is a schematic diagram of the cost complexity pruning process involved in this invention.

[0095] by Figure 8 For example, post-pruning with varying cost complexity involves 5 steps:

[0096] Step 1. Model Building. The model is a decision tree regression model. In the example shown in the figure, the input features are flow rate G and the average temperature t2 of the secondary supply and return water, and the objective function is the heat consumption Q. After model building, the blue decision tree diagram in the figure is generated. The tree diagram includes 3 decision nodes and 4 leaf nodes.

[0097] Step 2. Cost Complexity Calculation. Calculate the cost complexity of each internal node according to formulas (1) and (2). The example in the figure shows the cost complexity index of the three internal nodes after pruning.

[0098]

[0099] In the formula, α represents the cost complexity index; MSE represents the mean squared error; MSE(a) represents the increase in mean squared error of the current node after pruning; MSE(A) represents the sum of the increase in mean squared errors of all nodes under the current node after pruning; N(A) represents the number of leaf nodes pruned from the current node. n is the number of nodes, y i For the i-th actual observation, Let be the i-th predicted value.

[0100] Step 3. Prune the tree diagram, selecting trees with lower cost complexity. The example in the figure shows a new tree diagram in green after comparative pruning.

[0101] Step 4. Repeat steps 2 and 3 until the cost function no longer decreases. The example in the diagram recalculates the cost complexity of the pruned tree, and it is found that the cost complexity no longer decreases after the recalculation.

[0102] Step 5. Obtain the optimal pruning decision tree. The example in the figure, after post-pruning, results in the green tree diagram, retaining two decision nodes and three leaf nodes.

[0103] Figure 9 This is a schematic diagram of the 5-fold cross-validation principle involved in this invention.

[0104] by Figure 9 For example, the reliability assessment of a 5-fold cross-validation model consists of four steps:

[0105] Step 1. Dataset partitioning. Randomly divide the dataset into 5 equal parts. The example dataset in the figure has five data points, which are then divided into 5 parts, with one data point in each part.

[0106] Step 2. Model Training. Select one subset as the validation set and the remaining four subsets as the training sets for model training. The example in the figure shows five training and test sets obtained after partitioning, resulting in five decision tree models.

[0107] Step 3. Model Validation. Calculate the mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²) for the test and training sets. 2 Explained variance (Evar) is used for model validation. The figure shows five example regression models that have been validated, resulting in five model scores. The formulas for explained variance and expected value are shown in equations (3) and (4).

[0108] Step 4. Calculate the average performance metrics. Calculate the average of the five model performance metrics to evaluate the model's performance.

[0109]

[0110] In the formula, Evar represents the explainable variance; E x It represents the mathematical expectation of a certain independent variable (or feature) x.

[0111] This invention performs the steps of full feature selection using the wrapping method and finding the optimal parameters of the decision tree regression model simultaneously using Bayesian optimization. For model optimization, the DecisionTreeRegressor module from the "scikit-learn" library in Python 3 is used to build the decision tree regression model. The BayesSearchCV module is used for Bayesian optimization (the input training set data at this point has been filtered and embedded). Set the node splitting criterion to Friedman mean squared error (friedman_mse), mean squared error (mse), or mean absolute error (mae); the maximum tree depth (max_depth) is 3–9; the minimum number of samples required for internal node splitting (min_samples_split) is 2–8; the minimum number of samples required for leaf nodes (min_samples_leaf) is 1–4; the maximum number of classification features (max_features) is: all features considered (None), the square root of the total number of features is the maximum number of features (sqrt), or the maximum number of features is the logarithm of the total number of features (log2); the weight parameter (class_weigh) is: all classes have the same weight (None) or the weight parameter is inversely proportional to the number of samples in each class (balanced); the splitting strategy (splitter) is: best split (best) or random split (random).

[0112] For full search feature selection, the "cross_val_score" module in Python 3 is used to train and perform 5-fold cross-validation for each feature combination. A "for" loop is used to traverse all possible decision tree regression feature combinations, and the best cross-validation score for all feature combinations and parameter combinations is calculated to achieve full search feature selection and Bayesian optimization parameter tuning.

[0113] After calculation, six feature groups achieved optimal scores exceeding 75%, meeting the model accuracy requirements. These are {G,t2} and {G,B}. C}、{G,B C std t2}、{G,B C ,t2}、{G,B C ,t2 std t2 A CHART decision tree regression model was established for the six feature combinations {G,t2,A}. The model input parameters were set according to the model optimization results. The model was pruned after cost complexity was assessed, and the model was validated using the five-fold cross-validation method.

[0114] In the post-cost complexity pruning part, a grid search method was used to specify the search range of the cost complexity (ccp_alpha) parameter as 0 to 1, with a search step size of 0.01. The "fit" function was called to fit the feature factors and the objective function to find the optimal cost complexity parameter to control the degree of pruning. In the model validation part, the "cross_val_score" module was used to perform random 5-fold cross-training, and the mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²) were calculated for 5 training and test sets. 2 The mean of the explained variance Evar is used to validate model performance. When the mean squared error (MSE) and mean absolute error (MAE) of the regression model are less than 0.05, the coefficient of determination R0 is considered valid. 2 If the explained variance Evar is greater than 75%, the model is considered to have passed the test and can be used for heat consumption classification calculation.

[0115] After model training, the feature combination {G,t2} and the feature combination {G,B} are obtained. C Both models (which can be used as leaf nodes for subsequent classification of heat stations) passed model validation, demonstrating strong interpretability. The model validation results are shown in Table 2.

[0116] Table 2 Performance validation results of the decision tree regression model for heat consumption at the heating station

[0117]

[0118] Step 3: Validation and post-processing of the decision tree heat consumption classification model:

[0119] The obtained decision tree regression results cannot be directly used for classification benchmark calculation. Post-processing is required to remove unreasonable results. The regression tree results are analyzed based on knowledge of the heating industry, and heating stations are reasonably classified according to the results. Typical heating stations representing the operational characteristics of each type are selected for subsequent benchmark model establishment. This invention establishes an 8-step post-processing workflow for the decision tree heat consumption classification model, as follows: Figure 10 As shown.

[0120] Figure 10 This is a flowchart of the post-processing of the decision tree regression heat consumption classification model involved in this invention.

[0121] The mean absolute percentage error (MAPE) of each thermal station in the figure is calculated according to formula (5) for the prediction results of each regression model.

[0122]

[0123] It has been verified that the feature combination {G,t2} and the feature combination {G,B} are... CThe prediction results of the decision tree regression model can effectively reflect the heat consumption of the heating station.

[0124] Based on a decision tree regression model for the heat consumption of heating stations, the heat consumption of heating stations is classified through three steps: removing outlier samples, initial classification of leaf nodes, and reclassification by comparing heat consumption with the model. The heating stations corresponding to the leaf nodes of the regression tree after removing unreasonable samples are counted, and the heating stations are initially classified according to the leaf nodes of each model. Then, the categories with fewer heating stations in the initial classification results are counted, and these heating stations are assigned to the leaf node categories that are more similar to their features and more adjacent to them. First, heating stations that are classified into a single category are merged, then categories with two heating stations are merged, and so on, gradually merging categories with fewer heating stations to reduce the number of categories for classifying heat consumption.

[0125] Calculations determined that all heating stations were ultimately categorized into 13 classes, with categories 1 to 6 being floor radiant heating stations and categories 7 to 13 being radiator heating stations. A decision tree-based heat consumption classification model was established, utilizing selected features (feature combination {G,t2} or feature combination {G,B}). C The heat consumption of heating stations is classified.

[0126] For each type of heating station, the Person correlation coefficient between each energy consumption characteristic and the daily average heat consumption was calculated again. Factors with a correlation of 0.7 or higher were selected. The factors that were significantly correlated with the real-time heat consumption of the heating station were: primary flow rate, outdoor temperature, and secondary supply and return water average temperature.

[0127] Figure 11-13 Box plots showing the heat consumption, flow rate, and building property coefficients of various types of heating stations involved in this invention.

[0128] from Figures 11 to 13 It can be seen that the flow rate, heat consumption, and average temperature of secondary supply and return water for heating stations of categories 1 to 13 generally show an increasing trend, while the building comprehensive physical property coefficient shows a decreasing trend. That is, the heat consumption, flow rate, and average temperature of secondary supply and return water for radiator heating stations are generally higher than those for floor radiant heating, while the building comprehensive physical property coefficient is generally lower for floor radiant heating. According to the definition, B... C The value is the ratio of the product of the heat dissipation area and heat transfer coefficient of the heating system under the heat station to the product of the building envelope area and heat transfer coefficient of the building under the heat station. Therefore, the higher the building comprehensive physical property coefficient, the better the thermal insulation performance of the buildings under the heat station, the smaller the heat demand of the heat station, and the lower the corresponding average temperature, flow rate and heat consumption of the secondary supply and return water. This is consistent with the actual heating law. The classification results can effectively reflect the heat consumption characteristics of various types of heat stations, indicating that the constructed heat (cold) consumption classification model has a good classification effect. This model will be used for heat station classification in the future.

[0129] Further analysis of heat consumption data for various types of heating stations was conducted to verify the rationality of the classification results. Daily operating data for each type of heating station after classification were statistically analyzed, and Person correlation coefficients were calculated between heat consumption and parameters such as outdoor temperature and average temperature of secondary supply and return water. Analysis revealed significant correlations between heat consumption and primary flow rate, outdoor temperature, and average temperature of secondary supply and return water.

[0130] Step 4: Determine the antecedent and consequent of the rule:

[0131] Based on the analysis results in step 3 (which indicate that heat consumption is strongly correlated with primary flow rate, outdoor temperature, and average temperature of secondary supply and return water, and can be used as the antecedent of the rule), the primary flow rate of the heating station, outdoor temperature, average temperature of secondary supply and return water, and indoor temperature of typical users are selected as the antecedents of the association rule mining, and heat consumption is selected as the consequent of the rule.

[0132] Association rule mining requires discrete attribute data as the antecedents and consequents of rules. This invention uses the K-means clustering algorithm to discretize the primary flow rate, heat consumption, and secondary supply and return water average temperature data, with the number of clusters determined by the silhouette coefficient. An equal-width discretization method is used to discretize the indoor and outdoor temperature data, setting a discretization interval every 2°C. Calculations show that the optimal number of clusters for the primary flow rate, heat consumption, and secondary supply and return water average temperature are 13, 12, and 13, respectively. K-means clustering is then performed on these three data types according to the optimal number of clusters. Simultaneously, equal-width discretization is applied to the indoor and outdoor temperature data, ultimately discretizing the indoor temperature into 7 intervals and the outdoor temperature into 23 intervals. The discretization results are shown in Tables 3 and 4.

[0133] Table 3. Mining Discrete Intervals Based on Indoor Temperature Correlation Rules at Heating Stations

[0134]

[0135] Table 4. Mining Discrete Intervals Based on the Correlation Rules of Outdoor Temperature at Heating Stations

[0136]

[0137] Step 5: Use the Apriori algorithm to mine association rules for the discretized data of each type of heating station to determine the heat consumption benchmarks for the 13 types of heating stations.

[0138] The Apriori algorithm was used to discover frequent itemsets. The minimum support was set to 0.002, the minimum confidence to 60%, and the minimum lift to 1. The support, confidence, and lift were calculated according to formulas (6), (7), and (8) to generate association rules. Finally, a total of 17,167 original rules were generated for typical heat stations. The number of rules generated for each type of typical heat station is shown in Table 5.

[0139]

[0140] Where X and Y represent the preceding and following terms of the association rule mining, respectively; Support(X→Y) represents the support of the association rule; P(Y∪X) is the probability that X and Y overlap in the dataset; Count(XY) represents the number of times X and Y appear simultaneously; D is the size of the total dataset; Confidence(X→Y) represents the confidence of the association rule; Lift(X→Y) represents the lift of the association rule; and Support(X) represents the confidence of the preceding term of the rule.

[0141] Table 5. Original rules generated from the correlation rules of heat consumption at heating stations.

[0142]

[0143] Step 6: Rule Post-processing:

[0144] The original association rules discovered cannot be directly used for dynamic benchmark calculation of heat consumption and require post-processing, including rule grouping, rule comparison, and expert analysis. First, the association rules corresponding to all typical heating stations are grouped, with rules sharing the same outdoor temperature range grouped together. Then, the rules are compared, and rules with higher confidence (a confidence level greater than 60% of the minimum confidence level) are extracted. Finally, the rules are analyzed using heating domain knowledge, replacing unreasonable rules with rules that have slightly lower confidence (meaning lower than the current unreasonable rule's confidence level) but are reasonable (consistent with heating domain knowledge). The first rule that meets both the minimum confidence level and heating domain knowledge is selected to replace the unreasonable rule. Ultimately, each heating station generates 23 rules, corresponding to 23 outdoor temperature ranges. Taking Category 3 underfloor heating station 168 and Category 10 radiator heating station 74 as examples, the intervals of the antecedent and consequent of the rules for the two heating stations, along with their support, confidence, and lift indices, are shown in Tables 6 and 7, respectively.

[0145] Table 6. Generating Rules for Correlation Rules of Heat Consumption at Type 3 Heating Stations

[0146]

[0147]

[0148] Table 7. Rules for Generating Correlation Rules of Heat Consumption at Type 10 Heating Stations

[0149]

[0150]

[0151]

[0152] Step 7: Dynamic baseline for heat consumption:

[0153] The dynamic heat consumption benchmarks for each type of heat station were calculated using the association rules generated by 13 typical heat stations. The operating conditions and heat consumption rules of each heat station under each outdoor temperature range were statistically analyzed. The dynamic heat consumption benchmarks 1 and 2 of the heat station were calculated using the upper and lower bounds of the heat consumption range after the rule.

[0154] Figure 14-15 This is a heat consumption baseline chart, all selected based on outdoor temperature.

[0155] Figure 14 , Figure 15 These figures represent the dynamic heat consumption benchmarks for floor radiant heating and radiator heating stations according to embodiments of the present invention. The black curve in the figure represents the outdoor temperature during the heating season, while the other curves of different colors represent the dynamic heat consumption benchmarks for each type of floor radiant heating station throughout the entire heating season.

[0156] Depend on Figure 14 It can be seen that there is a significant negative correlation between the heat consumption benchmark and the outdoor temperature. Type 1 and Type 2 heating stations have the lowest heat consumption benchmarks, while Type 3 and Type 4 heating stations have moderate benchmarks. Type 5 heating stations have the highest heat consumption benchmark among the six types of underfloor heating stations, and Type 6 heating stations have the highest heat consumption benchmark. Figure 15 It can be seen that there is a significant negative correlation between the heat consumption benchmark and the outdoor temperature. The heat consumption benchmark of type 8 heating station is the lowest, followed by type 11 heating station; the heat consumption benchmarks of type 7, type 9, type 10 and type 12 heating stations are relatively high; and the heat consumption benchmark of type 13 heating station is the highest.

[0157] Based on the above description, relevant personnel can make various changes and modifications without departing from the technical concept of this invention.

[0158] Any aspects not covered in this invention are subject to existing technology.

Claims

1. A method for determining the dynamic baseline of heat / cooling capacity in a heating, ventilation, and air conditioning system, characterized in that, The method includes the following steps: Step 1: Feature Selection Data on HVAC systems under different operating conditions were obtained. After removing outliers, Person correlation coefficients were calculated between each energy consumption characteristic and heat / cooling load. Characteristics with Person correlation coefficients greater than or equal to 0.24 were retained. At the same time, the ReliefF algorithm was used to calculate the importance index of heat / cooling load, and characteristics with importance indexes greater than or equal to 0.003 were selected. After merging the Person correlation coefficient and the features selected by the ReliefF algorithm, the random forest embedding method was used to select features with a random forest importance index greater than 0.3 as the main features affecting energy consumption. Step 2: Construct a heat / cold energy classification model: The heat / cold energy classification model is based on a decision tree model. The nodes of the decision tree include a root node, leaf nodes, and decision nodes. The root node of the decision tree is the set of data from all heating stations. The decision nodes correspond to the energy consumption feature division rules. The decision nodes divide the data at the root node into several subsets according to the energy consumption feature rules. The leaf nodes are the final energy consumption classification nodes. The heating stations within the leaf nodes are classified into the same category to achieve the classification of heat consumption of the heating stations. The input features of the decision tree model are the main energy consumption features and corresponding heat consumption determined in step 1. Bayesian optimization is used to find the optimal parameters of the decision tree model. At the same time, the wrapping method is used to fully search the features to determine the best combination of decision tree regression features with a score of over 75%. Next, the decision tree model is pruned using a cost-complexity post-pruning method to remove unnecessary nodes, thereby obtaining the heat / cold energy classification model and the final input features of the model. Step 3: Using the heat / cooling capacity classification model constructed in Step 2 and the finally determined input features as input, classify the HVAC system data; Step 4: For each type of HVAC system data, recalculate the Person correlation coefficient between each energy consumption feature and heat / cooling load. Use all energy consumption features with a Person correlation coefficient greater than 0.7 as the antecedent of the association rule mining, and use heat / cooling load as the consequent of the rule. Step 5: Discretize the HVAC system data. After discretization, use the Apriori algorithm to mine association rules according to the rule antecedent and rule consequent determined in Step 4 to obtain the association rules for all classes. Step 6: Process the association rules of all classes obtained in Step 5 to obtain usable rules, and use the usable rules to determine the dynamic baseline of heat / cooling capacity of the HVAC system.

2. The method according to claim 1, characterized in that, The rule processing procedure is as follows: The outdoor temperature is discretized with equal width to obtain different outdoor temperature ranges. All the class association rules obtained in step 5 are grouped according to whether they are in the same outdoor temperature range, and the confidence level of the rules in each group is determined. Set a minimum confidence level, and extract all rules with a confidence level greater than the minimum confidence level for expert analysis; By analyzing the rules using knowledge from the heating sector, unreasonable rules are replaced with reasonable rules that have a lower confidence level than their neighbors.

3. The method according to claim 1, characterized in that, The process of determining the dynamic heat / cooling capacity dynamic baseline is as follows: statistically analyze the operating conditions and available heat consumption rules under all outdoor temperature ranges, and calculate the dynamic heat / cooling capacity baseline 1 and baseline 2 of the heating station based on the upper and lower bounds of the heat / cooling capacity range after the rule.

4. The method according to claim 1, characterized in that, In step 5, the discretization process is as follows: the flow rate, pressure, outdoor air temperature, and average temperature parameters of the supply and return water in the primary network of the HVAC system are discretized: K-means clustering is used to discretize the operating data and energy consumption data, and equal-width discretization is used to discretize the indoor and outdoor air temperatures.

5. The method according to claim 1, characterized in that, In the post-pruning method for decision tree models using cost complexity reduction, random 5-fold cross-training is used for model validation, and the mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²) are calculated for 5 training and test sets. 2 The mean of the explained variance Evar, when the mean squared error (MSE) and mean absolute error (MAE) of the model are less than 0.05, and the coefficient of determination R0 is... 2 If the explained variance Evar is greater than 75%, the model is considered to have passed the test and can be used for heat / cold energy classification.

6. The method according to claim 1, characterized in that, The heating, ventilation and air conditioning system is a heat station or energy station.

7. The method according to claim 1, characterized in that, In step 4, the preceding terms of the association rule mining rules are the primary flow rate of the energy station or heating station, the outdoor temperature, the average temperature of the secondary supply and return water, and the indoor temperature of typical users.

8. The method according to claim 1, characterized in that, In the Apriori algorithm, the minimum support is set to 0.002, the minimum confidence is 60%, and the minimum lift is 1.

Images