Training method, control method and device of air conditioner cluster control model

Abstract

The invention provides a training method and device and a control method and device for an air conditioner cluster control model. The training method comprises the steps that historical data of an air conditioner cluster are obtained, feature extraction is conducted on the historical data, and a feature training set is constructed; constructing an initial air conditioner cluster control model; based on the feature training set, utilizing a reward and punishment mechanism to train an initial air conditioner cluster control model to obtain an air conditioner cluster control model; the reward and punishment mechanism is used for optimizing an influence parameter matrix of a prediction temperature mapping relation according to the accuracy degree of the prediction temperature mapping relation from the pre-regulation environment temperature matrix to the target prediction environment temperature matrix; the prediction temperature mapping relation is used for obtaining a target prediction environment temperature matrix according to the influence degree of other devices in the air conditioner cluster on the environment temperature of the installation position of any device and the current environment temperature of the installation position of any device. Multiple influences existing in an industrial production site can be eliminated, automatic multi-dimensional temperature regulation and control are achieved, and the temperature regulation and control precision is improved.

Description

Technical Field

[0001] The present application relates to the technical field of air conditioning cluster control, and in particular to a training method, a control method and a device for an air conditioning cluster control model. Background Art

[0002] In industrial production sites, the temperature of the entire industrial production site is usually controlled by a central air-conditioning system, wherein the industrial production site can use multiple air-conditioning equipment for temperature control. However, in the actual application of temperature control, there are many factors in the industrial production site that have multiple influences on the temperature control results. Under the above multiple influences, it is difficult to achieve precise temperature control of the entire industrial production site. Summary of the Invention

[0003] The embodiments of the present application provide a training method, a control method, and a device for an air-conditioning cluster control model. By using the air-conditioning cluster control model trained by the training method to perform temperature control, multiple influences existing in the industrial production site can be eliminated, thereby realizing automatic multi-dimensional temperature control and improving the temperature control accuracy.

[0004] In a first aspect, an embodiment of the present application provides a training method for an air-conditioning cluster control model, comprising: obtaining historical data of an air-conditioning cluster, performing feature extraction on the historical data, and constructing a feature training set; constructing an initial air-conditioning cluster control model; wherein the air-conditioning cluster control model comprises an input layer and an output layer; the input layer is used to receive the device state parameter matrix of the air-conditioning cluster under a target control cycle, as well as the pre-control ambient temperature matrix and the target ambient temperature matrix at the installation location of each device in the air-conditioning cluster; the output layer is used to output the predicted device action parameter matrix of the air-conditioning cluster; based on the feature training set, the initial air-conditioning cluster control model is trained using a reward and punishment mechanism to obtain the air-conditioning cluster control model; wherein the reward and punishment mechanism is used to optimize the influence parameter matrix of the predicted temperature mapping relationship according to the accuracy of the predicted temperature mapping relationship from the pre-control ambient temperature matrix to the target predicted ambient temperature matrix; the predicted temperature mapping relationship is used to obtain the target predicted ambient temperature matrix according to the degree of influence of each device in the air-conditioning cluster on the ambient temperature of each installation location of the device and the current ambient temperature at the installation location of each device.

[0005] In a possible implementation, the method of training the initial air conditioning cluster control model based on the feature training set and utilizing a reward and punishment mechanism to obtain the air conditioning cluster control model includes: constructing the predicted temperature mapping relationship; wherein the mathematical expression of the predicted temperature mapping relationship includes: X(t+1)=A·X(t)+∑ i [B i Qi (t)]; wherein X(t+1) is the target predicted ambient temperature matrix of the t-th control cycle, X(t) is the ambient temperature matrix before control of the t-th control cycle, A is the temperature influence matrix containing the intrinsic influence weights between the devices in the air-conditioning cluster, Q i (t) is the state parameter matrix of the equipment of type i in the t-th control cycle, B i is an i-th type of equipment state parameter influence matrix containing the intrinsic influence weights between the equipment in the air-conditioning cluster; based on the feature training set, constructing a linear regression equation group corresponding to the predicted temperature mapping relationship;

[0006] Solve the linear regression equations based on the least squares method to obtain the temperature influence matrix A and the i-th type equipment state parameter influence matrix B. i The initial value of .

[0007] In a possible implementation, the linear regression equations are solved based on the least squares method to obtain the temperature influence matrix A and the i-th type equipment state parameter influence matrix B. i After the initial value of the training set is obtained, the method further includes: step 11, constructing a predicted temperature mapping equation for this round of training based on the predicted temperature mapping relationship and the feature training set; step 12, solving the predicted temperature mapping equation based on the Kalman filter algorithm to obtain the target predicted ambient temperature matrix for this round of training; step 13, judging whether the difference between the target predicted ambient temperature matrix for this round of training and the target ambient temperature matrix for this round of training is less than a second set threshold; step 14, if not less than, re-solving the linear regression equation group based on the least squares method, and updating the temperature influence matrix A and the i-th type equipment state parameter influence matrix B. i , and return to step 11; step 15, if it is less than, the model obtained in this round of training is used as the air-conditioning cluster control model, and the training is ended.

[0008] In one possible implementation, the types of the equipment state parameter matrix include one or more of an air-conditioning specific power matrix, a workshop production equipment capacity matrix, a workshop production product type matrix, an air-conditioning refrigeration water temperature matrix, an air-conditioning refrigeration water valve opening matrix, an air-conditioning fan frequency matrix, an air-conditioning fan outlet temperature matrix, a season type matrix, a weather type matrix, a factory external temperature matrix, and a fresh air system valve opening matrix.

[0009] In one possible implementation, the historical data of the air-conditioning cluster is obtained, feature extraction is performed on the historical data, and a feature training set is constructed, including: cleaning and / or filling the historical data of the air-conditioning cluster to obtain preprocessed historical data; feature extraction is performed on the preprocessed historical data to obtain original feature data; derivative processing is performed on the original feature data to obtain expanded feature data; feature screening is performed on the expanded feature data to obtain final feature data, and the feature training set is constructed.

[0010] In one possible implementation, the feature screening of the expanded feature data to obtain final feature data and construct the feature training set includes: performing variance value screening on the expanded feature data to obtain first screened feature data; performing multicollinearity screening on the first screened feature data to obtain second screened feature data; performing parameter combination on the feature types in the second screened feature data based on a hyperparameter optimization algorithm, and screening the second screened feature data according to the results of cross-validation to obtain third screened feature data; and using an XGBoost model to screen out one or more types of feature data that are most sensitive to the target predicted ambient temperature matrix from the third screened feature data to obtain the final feature data.

[0011] In the second aspect, an embodiment of the present application also provides an air-conditioning cluster control method, including: obtaining an equipment status parameter matrix of the air-conditioning cluster; inputting the target ambient temperature matrix and the equipment status parameter matrix into the air-conditioning cluster control model obtained by the training method provided in the first aspect to obtain a predicted equipment action parameter matrix; according to the predicted equipment action parameter matrix, controlling the air-conditioning cluster to adjust the ambient temperature at the installation location of each device in the air-conditioning cluster to the temperature corresponding to the target ambient temperature matrix.

[0012] In a third aspect, an embodiment of the present application also provides a training device for an air-conditioning cluster control model, comprising: a data acquisition module for acquiring historical data of the air-conditioning cluster, performing feature extraction on the historical data, and constructing a feature training set; a model construction module for constructing an initial air-conditioning cluster control model; wherein the air-conditioning cluster control model comprises an input layer and an output layer; the input layer is used to receive the device state parameter matrix of the air-conditioning cluster under the target control cycle, as well as the pre-control ambient temperature matrix and the target ambient temperature matrix at the installation location of each device in the air-conditioning cluster; the output layer is used to output the predicted device action parameter matrix of the air-conditioning cluster; a training module is used to train the initial air-conditioning cluster control model based on the feature training set using a reward and punishment mechanism to obtain the air-conditioning cluster control model; wherein the reward and punishment mechanism is used to optimize the influence parameter matrix of the predicted temperature mapping relationship according to the accuracy of the predicted temperature mapping relationship from the pre-control ambient temperature matrix to the target predicted ambient temperature matrix; the predicted temperature mapping relationship is used to obtain the target predicted ambient temperature matrix based on the degree of influence of each device in the air-conditioning cluster on the ambient temperature of each installation location of the device and the current ambient temperature at the installation location of each device.

[0013] In the fourth aspect, an embodiment of the present application also provides an air-conditioning cluster control device, including: a parameter acquisition module for obtaining the equipment status parameter matrix of the air-conditioning cluster; an input module for inputting the target ambient temperature matrix and the equipment status parameter matrix into the air-conditioning cluster control model obtained by the training method provided in the first aspect to obtain the predicted equipment action parameter matrix; a control module for controlling the air-conditioning cluster according to the predicted equipment action parameter matrix to adjust the ambient temperature at the installation location of each device in the air-conditioning cluster to the temperature corresponding to the target ambient temperature matrix.

[0014] In a fifth aspect, an embodiment of the present application further provides an electronic device, comprising a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the method provided in the first aspect and / or the second aspect.

[0015] Through the above technical solution, the air conditioning cluster control model can be trained by the training method. During the training process, the area covered by the air conditioning cluster can be decoupled to obtain an influence parameter matrix for eliminating the influence of temperature, and then the trained air conditioning cluster control model is obtained after optimizing the predicted temperature mapping relationship of the model through the influence parameter matrix. In the actual temperature control application process, the predicted device action parameter matrix can be obtained according to the trained air conditioning cluster control model; and then, according to the predicted device action parameter matrix, the air conditioning cluster can be controlled to adjust the ambient temperature at the installation location of each device in the air conditioning cluster to the temperature corresponding to the target ambient temperature matrix. Based on the implementation of this solution, the automatic and efficient energy-saving operation of the air conditioning system can be realized, and energy consumption can be saved on the basis that the actual temperature is close to the target temperature. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0017] Figure 1 A schematic diagram of an air conditioning system application scenario provided by one embodiment of the present application;

[0018] Figure 2 A flowchart of a training method for an air conditioning cluster control model provided in one embodiment of the present application;

[0019] Figure 3 This is a schematic diagram of the input and output of an air conditioning cluster control model provided in one embodiment of the present application;

[0020] Figure 4 A flow chart of an air conditioning cluster control method provided in one embodiment of the present application. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0022] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0023] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0024] Figure 1 A schematic diagram of an air conditioning system application scenario provided for one embodiment of the present application.

[0025] Reference Figure 1 As shown, the scene may be a factory building at an industrial production site, which may include i production equipment Di running production, and the ambient temperature in the factory building can be controlled by the air-conditioning system to make the ambient temperature reach the target ambient temperature.

[0026] In some embodiments, the air conditioning system may be a temperature control system based on an air conditioning cluster. The air conditioning devices of the air conditioning cluster are dispersedly installed in different locations of the factory, and each air conditioning device controls the temperature of the area it is responsible for.

[0027] In order to reduce or eliminate the multiple influences of various factors on the temperature control results in the industrial production site, the embodiment of the present application obtains an air-conditioning cluster control model through model training. When performing temperature control applications, the equipment state parameter matrix and the target ambient temperature matrix of the air-conditioning cluster can be input into the air-conditioning cluster control model to obtain the predicted equipment action parameter matrix; according to the predicted equipment action parameter matrix, the air-conditioning cluster is controlled to adjust the ambient temperature at the installation location of each device in the air-conditioning cluster to the temperature corresponding to the target ambient temperature matrix.

[0028] In order to obtain the air-conditioning cluster control model, the present application also provides a training method for the air-conditioning cluster control model, so that the air-conditioning cluster control model can be obtained through the training method.

[0029] The following describes in detail the training method of the air conditioning cluster control model provided in the embodiment of the present application with reference to the accompanying drawings.

[0030] Figure 2 A flowchart of a training method for an air conditioning cluster control model provided in one embodiment of the present application.

[0031] Reference Figure 2As shown, the training method may include the following steps:

[0032] S201: Obtain historical data of the air conditioning cluster, extract features from the historical data, and construct a feature training set.

[0033] In some embodiments, a specific implementation method of obtaining historical data of an air conditioning cluster, performing feature extraction on the historical data, and constructing a feature training set may include the following steps:

[0034] S201a: Clean and / or fill the historical data of the air-conditioning cluster to obtain pre-processed historical data.

[0035] In some embodiments, the specific method of cleaning the historical data may include deleting data with more than 90% blank rows and deleting data with more than 90% blank columns in the historical data to complete data cleaning.

[0036] In some embodiments, the specific method of filling the historical data may include maintaining the integrity of the data structure by filling in when there are discontinuous null values in the data sequence.

[0037] It should be noted that when corresponding problems exist in the historical data (such as more than 90% empty rows, more than 90% empty columns, and intermittent empty values), pre-processed historical data can be obtained by performing the above-mentioned cleaning and / or filling processing; if there are no response problems in the historical data, the acquired historical data can be directly used as pre-processed historical data.

[0038] S201b: Perform feature extraction processing on the pre-processed historical data to obtain original feature data.

[0039] In some embodiments, feature extraction processing of pre-processed historical data may include extracting one or more of the following feature types: air conditioner specific power, workshop production equipment capacity, workshop production product types, indoor temperature, air conditioner outlet temperature, air conditioner seasonal mode (air conditioner seasonal type), fan operating status, air conditioner cooling water temperature, air conditioner heating water temperature, weather type, factory external temperature, factory external humidity, air conditioner cooling water valve opening, fan frequency, fan valve opening (fresh air system valve opening).

[0040] S201c: Perform derivative processing on the original feature data to obtain expanded feature data.

[0041] In some embodiments, a specific implementation of the derivation processing of the original feature data may include: deriving lag term features and deriving time dimension features.

[0042] The following example illustrates the meaning of features in the original feature data. For example, a piece of original feature data obtained is "ZZ1_D01_3F_AHU08__RM_TT_lag_10", where "3F" represents the name of the air-conditioning equipment (AHU name), "RM_TT" represents the indoor temperature, and "lag_10" represents 10 periods. If the data collection frequency is 1 minute, the "lag_10" feature in the data can be used to determine that the data is the indoor temperature 10 minutes ago.

[0043] Derived lag features can be understood as feature parameters collected before the current round of parameter collection (which can be the previous round, the previous previous round, or a certain amount of time ago). Derived time dimension features can be understood as expanding features from different time dimensions, such as adding corresponding features in dimensions such as minutes, hours, days, months, quarters, and years.

[0044] It should be noted that this step is used to expand the types of features to facilitate the formation of more comprehensive parameter features for subsequent screening and machine learning training.

[0045] S201d: Perform feature screening on the expanded feature data, obtain final feature data, and construct a feature training set.

[0046] In some embodiments, performing feature screening on the expanded feature data, obtaining final feature data, and constructing a feature training set includes the following steps:

[0047] Sd1: Perform variance value screening on the expanded feature data to obtain the first screened feature data.

[0048] In some embodiments, performing variance screening on the extended feature data may specifically include deleting features with a variance value of 0 from the extended feature data. For example, if the feature value of the "factory external temperature" feature in the extended feature data is all "26°C," and its variance value is 0, it can be considered that the feature has little or no impact on air conditioning temperature control and can be deleted.

[0049] Sd2: Perform multicollinearity screening on the first screening feature data to obtain the second screening feature data.

[0050] In some embodiments, multicollinearity screening of the first screening feature data can specifically be performed by deleting corresponding features with multicollinearity. For example, if variables a, b, and c satisfy a=b+c, it is considered that multicollinearity exists between a, b, and c, and feature a can be deleted.

[0051] Sd3: Based on the hyperparameter optimization algorithm, the parameters of the feature types in the second screening feature data are combined, and according to the results of cross-validation, the second screening feature data are screened to obtain the third screening feature data.

[0052] In some embodiments, after obtaining the second screening feature data, discrete intervals of the XGBoost parameters "tree depth" and "learning rate" can be given respectively based on the hyperparameter optimization algorithm, and then different parameter combinations can be tried using grid search, and the optimal parameter combination can be selected based on the results of cross-validation.

[0053] Sd4: Using the XGBoost model, one or more types of feature data that are most sensitive to the target predicted ambient temperature matrix are screened out from the third filtered feature data to obtain the final feature data.

[0054] In some embodiments, after obtaining the third screening feature data, a sensitivity analysis can be performed. Specifically, the sensitivity analysis can be performed to screen out corresponding features according to the degree of influence of the input quantity change on the model output result. Specifically, one or more types of feature data that are most sensitive to the target predicted ambient temperature matrix can be screened out from the third screening feature data to obtain the final feature data. Exemplarily, the third screening feature data includes data 1 and data 2, wherein data 1 is a target ambient temperature table, the independent variables in the table are parameter features other than indoor temperature, and the dependent variable is the factory ambient temperature of the next period, and the XGBoost model can be used; data 2 is a reward and punishment table, the independent variables in the table are parameter features other than reward and punishment values, and the dependent variable is the reward and punishment value. The XGBoost model can obtain multiple features with the highest ranking in data 1 (for example, top 8 features), and filter out the most sensitive feature in data 1 through sensitivity analysis to obtain the first sensitive feature. Similarly, the XGBoost model can obtain multiple features with the highest ranking in data 2 (for example, top 8 features), and filter out the most sensitive feature in data 2 through sensitivity analysis to obtain the second sensitive feature. The first sensitive feature and the second sensitive feature can then be combined to obtain the final feature data.

[0055] By executing the above steps Sd1 to Sd4, it is possible to perform feature screening on the expanded feature data, obtain final feature data, and construct a feature training set.

[0056] In some embodiments, after constructing the feature training set, it is further possible to determine whether the feature training set is suitable for model training. For example, a Markov property verification can be performed on the feature training set to verify whether the feature training set is suitable for model training, thereby obtaining a feature training set suitable for model training.

[0057] S202: Construct an initial air conditioning cluster control model.

[0058] Figure 3 This is a schematic diagram of the input and output of an air conditioning cluster control model provided in one embodiment of the present application.

[0059] Reference Figure 3 As shown, the air conditioning cluster control model may include an input layer and an output layer.

[0060] The input layer is used to receive the device state parameter matrix of the air conditioning cluster under the target control cycle, as well as the pre-control ambient temperature matrix and the target ambient temperature matrix at the installation location of each device in the air conditioning cluster (air conditioning device j among n air conditioning devices). The ambient temperature matrix can be a collection of ambient temperature values for multiple areas within an industrial production site.

[0061] For example, refer to Figure 1 The scenario diagram shown can divide the industrial production site into area A1, area A2, area A3, area A4, area A5 and area A6 based on the temperature control area of the air-conditioning cluster, wherein each area is responsible for temperature control by different air-conditioning equipment, and one area can be temperature-controlled by one or more air-conditioning equipment, and this application does not limit this. Under this condition, the ambient temperature matrix can include the temperatures within multiple areas of area A1, area A2, area A3, area A4, area A5, and area A6, that is, the ambient temperature matrix can be a 6×1 column matrix or a 1×6 row matrix, which can be flexibly selected based on actual calculations. It can be further determined that the ambient temperature matrix before control may include the temperatures of area A1, area A2, area A3, area A4, area A5 and area A6 before control, and the target ambient temperature matrix may include the target ambient temperatures of area A1, area A2, area A3, area A4, area A5 and area A6 after control, that is, the target ambient temperature matrix can be a 6×1 column matrix or a 1×6 row matrix, which can be flexibly selected based on actual calculations.

[0062] Among them, it should be noted that in the process of temperature control of industrial production sites (i.e. workshops or factory buildings) by air-conditioning clusters, there are many factors that affect the actual room temperature of the factory workshop. Specifically, when the temperature of the air-conditioning cluster is adjusted to the target ambient temperature without considering the influence of the above-mentioned factors, there is a difference between the actual room temperature of the factory workshop and the target ambient temperature, which results in the temperature control in the factory workshop failing to meet the preset temperature conditions.

[0063] In order to overcome the influence of the above-mentioned factors on temperature control, the above-mentioned factors are included in the reference range of temperature control. Specifically, the equipment state parameter matrix can be used as input data for model training during the model training process, so that after temperature control is performed based on the output value of the trained model, the actual room temperature of the factory workshop is closer to the target ambient temperature.

[0064] In some embodiments, the above-mentioned multiple factors may include one or more of the specific power of the air conditioner, the production capacity of the workshop production equipment, the types of products produced in the workshop, the air conditioner cooling water temperature, the opening of the air conditioner cooling water valve, the air conditioner fan frequency, the air conditioner fan outlet temperature, the season type, the weather type, the factory external temperature, and the opening of the fresh air system valve.

[0065] Among them, the specific power of the air conditioner determines the cooling capacity of the air conditioner, which directly affects the actual room temperature of the factory workshop.

[0066] The production capacity of workshop production equipment comprehensively reflects the heat generated by the production equipment. That is, when the production capacity of workshop production equipment increases, the heat generated by the production equipment will increase accordingly. This heat generated will affect the temperature around the production equipment, thereby affecting the actual room temperature of the factory workshop.

[0067] When production equipment produces different types of products, the heat emitted by the production equipment is different. Therefore, the type of products produced in the workshop affects the heat generated by the production equipment. This heat will affect the temperature around the production equipment, thereby affecting the actual room temperature of the factory workshop.

[0068] The air conditioning cooling water temperature will change with the operating time of the air conditioning equipment, which will cause the cooling temperature of the air conditioning equipment to change accordingly, thus affecting the actual room temperature of the factory workshop.

[0069] The opening degree of the air conditioning cooling water valve is directly related to the heat exchange efficiency inside the air conditioning equipment, which will affect the cooling temperature of the air conditioning equipment, and in turn will affect the actual room temperature of the factory workshop.

[0070] When the frequency of the air conditioner fan changes, the cooling capacity output from the air outlet of the air conditioner equipment will be affected, thereby affecting the actual room temperature of the factory workshop.

[0071] When the air outlet temperature of the air conditioner fan changes, the cooling temperature of the air outlet of the air conditioning equipment will change, thereby affecting the actual room temperature of the factory workshop.

[0072] In different seasons, different types of seasons will have different impacts on the actual room temperature of the factory workshop.

[0073] Different weather types will have different effects on the actual room temperature in the factory workshop.

[0074] The temperature outside the factory will exchange heat with the factory workshop, and changes in the temperature outside the factory will affect the actual room temperature in the factory workshop.

[0075] The opening degree of the fresh air system valve will affect the exhaust volume of the fresh air system outlet, and the exhaust volume will affect the actual room temperature of the factory workshop.

[0076] Taking into account the impact of the above-mentioned factors on the actual room temperature of the factory workshop, the equipment state parameter matrix can be added to the training method provided in the embodiment of the present application to improve the accuracy of the model calculation results after training and its adaptability to actual scenarios.

[0077] In some embodiments, the types of equipment status parameter matrices include one or more of an air conditioning specific power matrix, a workshop production equipment capacity matrix, a workshop production product type matrix, an air conditioning refrigeration water temperature matrix, an air conditioning refrigeration water valve opening matrix, an air conditioning fan frequency matrix, an air conditioning fan outlet temperature matrix, a season type matrix, a weather type matrix, a factory external temperature matrix, and a fresh air system valve opening matrix.

[0078] The output layer is used to output the predicted equipment action parameter matrix of the air conditioning cluster.

[0079] S203: Based on the feature training set, an initial air conditioning cluster control model is trained using a reward and punishment mechanism to obtain the air conditioning cluster control model.

[0080] In some embodiments, based on the feature training set, an initial air conditioning cluster control model is trained using a reward and punishment mechanism, and obtaining the air conditioning cluster control model includes the following steps:

[0081] S203a: Constructing a predicted temperature mapping relationship;

[0082] The mathematical expressions for predicting the temperature mapping relationship include:

[0083] X(t+1)=A·X(t)+∑ i [B i Q i (t)];

[0084] Among them, X(t+1) is the target predicted ambient temperature matrix of the t-th control cycle, X(t) is the ambient temperature matrix before control of the t-th control cycle, A is the temperature influence matrix containing the intrinsic influence weights between each device in the air conditioning cluster, Q i (t) is the state parameter matrix of the i-th type of equipment in the t-th control cycle, B i is the state parameter influence matrix of the i-th type of equipment, which includes the intrinsic influence weights among the equipment in the air-conditioning cluster.

[0085] Specifically, the predicted temperature mapping relationship is used to obtain a target predicted ambient temperature matrix based on the degree of influence of each device in the air conditioning cluster on the ambient temperature of the installation location of each device and the current ambient temperature at the installation location of each device.

[0086] Here, the state parameter influence matrix B of the i-th type equipment is i Give examples to illustrate the extent to which other equipment affects the ambient temperature at the installation location of any equipment.

[0087]

[0088] The influence matrix B of the state parameters of the i-th type of equipment i In, b mn is the weight factor of the impact of the equipment in area Am on the ambient temperature of the equipment in area An under the state parameters of the i-th type of equipment. Specifically, b 11 is the weight factor of the impact of the equipment in area A1 on the ambient temperature of the equipment in area A1 under the state parameters of the i-th type of equipment, b 12 is the weight factor of the impact of the ambient temperature of the equipment in area A1 on the equipment in area A2 under the state parameters of the i-th type of equipment, b 23 is the weight factor of the impact of the equipment in area A2 on the ambient temperature of the equipment in area A3 under the i-th type of equipment status parameters, and so on.

[0089] In some embodiments, the ambient temperature matrix X(t) before the regulation of the t-th regulation cycle and the state parameter matrix Q of the i-th type of equipment in the t-th regulation cycle are obtained. i (t), and X(t) and Q i (t) is brought into the predicted temperature mapping relationship to obtain the target predicted ambient temperature matrix X(t+1) for the t-th control cycle.

[0090] In some embodiments, the state parameter matrix Q of the i-th type of equipment in the t-th control cycle is i The type of (t) may include the air conditioning specific power matrix Q P (t), workshop production equipment capacity matrix Q Q (t), workshop production product category matrix Q pc (t), air conditioning cooling water temperature matrix Q tc (t), air conditioning cooling water valve opening matrix Q voc (t), air conditioning fan frequency matrix Q FP (t), air conditioning fan outlet temperature matrix Q FT (t), seasonal category matrix Q S (t), weather type matrix Q W (t), factory external temperature matrix Q TO (t) and fresh air system valve opening matrix Qvof (t) one or more.

[0091] For example, when the state parameter matrix Q of the i-th type of equipment in the t-th control cycle is i When the types of (t) include all the above types, the acquired device state parameter matrix of all types can be brought into the predicted temperature mapping relationship, so that the mathematical expression of the predicted temperature mapping relationship is as follows:

[0092] X(t+1)=A·X(t)+B P Q P (t)+B Q Q Q (t)+B pc Q pc (t)+B tc Q tc (t)+B voc Q voc (t)+B FP Q FP (t)+B FT Q FT (t)+B S Q S (t)+B W Q W (t)+B TO Q TO (t)+B vof Q vof (t);

[0093] Among them, B P represents the influence matrix of the air conditioner specific power matrix, B Q Represents the impact matrix of the workshop production equipment capacity matrix, B pc The influence matrix of the workshop production product type matrix, B tc represents the influence matrix of the air conditioning cooling water temperature matrix, B voc The influence matrix of the air conditioning cooling water valve opening matrix, B FP represents the influence matrix of the air conditioning fan frequency matrix, B FT represents the influence matrix of the air conditioning fan outlet temperature matrix, B S represents the influence matrix of the seasonal category matrix, B W Represents the influence matrix of weather type matrix, B TO represents the influence matrix of the factory external temperature matrix, B vof Represents the influence matrix of the fresh air system valve opening matrix.

[0094] S203b: Constructing a linear regression equation group corresponding to the predicted temperature mapping relationship based on the feature training set;

[0095] In some embodiments, based on the feature training set, the obtained ambient temperature matrix X(t) before the control of the t-th control cycle and the state parameter matrix Q of the i-th type of equipment in the t-th control cycle are obtained. i (t), bring in the predicted temperature mapping relationship, and obtain the corresponding mathematical expression, that is, complete the construction of the linear regression equation group corresponding to the predicted temperature mapping relationship.

[0096] S203c: Solve the linear regression equations based on the least squares method to obtain the temperature influence matrix A and the i-th type equipment state parameter influence matrix B i The initial value of .

[0097] In some embodiments, after obtaining the linear regression equation group corresponding to the predicted temperature mapping relationship, the linear regression equation group can be solved based on the least squares method to obtain the temperature influence matrix A and the i-th type device state parameter influence matrix B i For example, the device state parameter influence matrix B i When all the above types are included, the temperature influence matrix A and the influence matrix B of the air conditioner specific power matrix are calculated by the least squares method. P , the impact matrix B of the workshop production equipment capacity matrix Q , the impact matrix B of the workshop production product type matrix pc , the influence matrix B of the air conditioning cooling water temperature matrix tc , the matrix weight B of the air conditioning cooling water valve opening matrix voc , the influence matrix B of the air conditioning fan frequency matrix FP , the influence matrix B of the air conditioning fan outlet temperature matrix FT , the influence matrix B of the seasonal category matrix S , the influence matrix B of the weather type matrix W , the impact matrix B of the factory external temperature matrix TO , and the influence matrix B of the fresh air system valve opening matrix vof .

[0098] In some embodiments, the linear regression equations are solved based on the least squares method to obtain the temperature influence matrix A and the i-th type device state parameter influence matrix B. i After the initial value, the following steps are also included:

[0099] S11: Based on the predicted temperature mapping relationship and the feature training set, a predicted temperature mapping equation for this round of training is constructed.

[0100] S12: Based on the Kalman filter algorithm, the predicted temperature mapping equation is solved to obtain the target predicted ambient temperature matrix for this round of training.

[0101] S13: Determine whether the difference between the target predicted ambient temperature matrix of this round of training and the target ambient temperature matrix of this round of training is less than a second set threshold. If so, execute S14; if not, execute S15.

[0102] S14: If it is not less than, then solve the linear regression equations again based on the least squares method, and update the temperature influence matrix A and the i-th type equipment state parameter influence matrix B. i and returns S11.

[0103] S15: If it is less than, the model obtained in this round of training is used as the air conditioning cluster control model and the training ends.

[0104] In some embodiments, in order to make the target predicted ambient temperature matrix predicted based on the currently trained predicted temperature mapping equation closer to the actual target ambient temperature matrix, the predicted temperature mapping equation can be continuously optimized by looping S11 to S15 until the difference between the target predicted ambient temperature matrix of this round of training and the target ambient temperature matrix of this round of training is less than the second set threshold, and then the air conditioning cluster control model is obtained.

[0105] In some embodiments, the specific method of optimizing the predicted temperature mapping equation can be to optimize and update A and B in the predicted temperature mapping equation. i The value of A and the i-th type device B after optimization is i It is more suitable for the input values currently calculated by the prediction temperature mapping equation, namely the temperature influence matrix and the i-th type equipment state parameter influence matrix, so as to achieve the purpose of making the target prediction ambient temperature matrix predicted by the prediction temperature mapping equation closer to the actual target ambient temperature matrix.

[0106] In some embodiments, after several rounds of training, if it is determined that the difference between the target predicted ambient temperature matrix of this round of training and the target ambient temperature matrix of this round of training is less than a second set threshold, the model obtained in this round of training can be used as the air conditioning cluster control model and the training is terminated.

[0107] In some embodiments, after completing the training of the air-conditioning cluster control model through the feature training set, the trained air-conditioning cluster control model can also be evaluated based on the test data set to obtain the evaluation result of the trained air-conditioning cluster control model, and determine whether the evaluation result meets the preset evaluation effect. If the preset evaluation effect is not achieved, the model can continue to be trained until the preset evaluation effect is achieved.

[0108] In some embodiments, after obtaining an air conditioning cluster control model that meets preset evaluation results, the air conditioning cluster control model can be deployed online, thereby enabling actual regulation of the air conditioning cluster at the industrial production site. During the deployment process, the model can also be configured for mode selection. For example, the types of equipment state parameter matrices can be set, that is, adaptive settings can be made based on actual on-site conditions and actual needs to meet regulation requirements.

[0109] Figure 4 A flow chart of an air conditioning cluster control method provided in one embodiment of the present application.

[0110] Reference Figure 4 As shown, the air conditioning cluster control method may include the following steps:

[0111] S401: Obtaining the device state parameter matrix of the air conditioning cluster.

[0112] In some embodiments, when the next control cycle comes, the current equipment status parameter matrix of the air-conditioning cluster can be obtained. In one embodiment, the types of the equipment status parameter matrix include one or more of the air-conditioning specific power matrix, the workshop production equipment capacity matrix, the workshop production product type matrix, the air-conditioning refrigeration water temperature matrix, the air-conditioning refrigeration water valve opening matrix, the air-conditioning fan frequency matrix, the air-conditioning fan outlet temperature matrix, the season type matrix, the weather type matrix, the factory external temperature matrix and the fresh air system valve opening matrix.

[0113] S402: Input the target ambient temperature matrix and the device state parameter matrix into the trained air conditioning cluster control model to obtain the predicted device action parameter matrix.

[0114] In some embodiments, after obtaining the device state parameter matrix, the target ambient temperature matrix can also be obtained, and the target ambient temperature matrix and the device state parameter matrix are used as input data to input the trained air conditioning cluster control model. Figure 1 In the scenario diagram shown, the target ambient temperature matrix includes the target ambient temperature of each area in area A1, area A2, area A3, area A4, area A5, and area A6.

[0115] By inputting the target ambient temperature matrix and the equipment state parameter matrix into the trained air-conditioning cluster control model, the predicted equipment action parameter matrix output by the trained air-conditioning cluster control model can be obtained, wherein the predicted equipment action parameter matrix may include the control parameters of each air-conditioning equipment in the air-conditioning cluster.

[0116] S403: Control the air conditioning cluster according to the predicted device action parameter matrix to adjust the ambient temperature at the installation location of each device in the air conditioning cluster to a temperature corresponding to the target ambient temperature matrix.

[0117] In some embodiments, after obtaining the predicted device action parameter matrix output by the trained air-conditioning cluster control model, the corresponding air-conditioning equipment in the air-conditioning cluster can be independently regulated based on the control parameters of each air-conditioning equipment in the air-conditioning cluster. Therefore, after completing the independent regulation of each air-conditioning equipment in the air-conditioning cluster, the ambient temperature at the installation location of each equipment is adjusted to the temperature corresponding to the target ambient temperature matrix. In other words, the actual temperature of each area of the industrial production site can meet the temperature conditions of the corresponding area, and then the air-conditioning system can be automatically and efficiently operated with energy saving, effectively saving energy consumption on the basis of being close to the set value.

[0118] An embodiment of the present application also provides a training device for an air-conditioning cluster control model, including: a data acquisition module for acquiring historical data of the air-conditioning cluster, performing feature extraction on the historical data, and constructing a feature training set; a model construction module for constructing an initial air-conditioning cluster control model; wherein the air-conditioning cluster control model includes an input layer and an output layer; the input layer is used to receive the device state parameter matrix of the air-conditioning cluster under the target control cycle, as well as the pre-control ambient temperature matrix and the target ambient temperature matrix at the installation location of each device in the air-conditioning cluster; the output layer is used to output the predicted device action parameter matrix of the air-conditioning cluster; a training module is used to train the initial air-conditioning cluster control model based on the feature training set using a reward and punishment mechanism to obtain the air-conditioning cluster control model; wherein the reward and punishment mechanism is used to optimize the influencing parameter matrix of the predicted temperature mapping relationship according to the accuracy of the predicted temperature mapping relationship from the pre-control ambient temperature matrix to the target predicted ambient temperature matrix; the predicted temperature mapping relationship is used to obtain the target predicted ambient temperature matrix based on the degree of influence of other devices in the air-conditioning cluster on the ambient temperature at the installation location of any device and the current ambient temperature at the installation location of any device.

[0119] An embodiment of the present application also provides an air-conditioning cluster control device, including: a parameter acquisition module for obtaining the equipment status parameter matrix of the air-conditioning cluster; an input module for inputting the target ambient temperature matrix and the equipment status parameter matrix into the air-conditioning cluster control model obtained by the training method provided in the first aspect to obtain the predicted equipment action parameter matrix; a control module for controlling the air-conditioning cluster according to the predicted equipment action parameter matrix to adjust the ambient temperature at the installation location of each device in the air-conditioning cluster to the temperature corresponding to the target ambient temperature matrix.

[0120] An embodiment of the present application also provides an electronic device, including a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the training method of the air-conditioning cluster control model provided in any embodiment of the present application and the air-conditioning cluster control method provided in any embodiment of the present application.

[0121] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0122] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, which may be electrical, mechanical or other forms.

[0123] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0124] In addition, the functional units in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional units.

[0125] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to perform some steps of the method described in various embodiments of the present invention. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., various media that can store program code.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0127] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0128] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. That is, they may be located in one place or distributed across at least two network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0129] 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 above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A training method for an air conditioning cluster control model, characterized in that: The method comprises: Acquire historical data of the air conditioning cluster, perform feature extraction on the historical data, and construct a feature training set; Construct an initial air conditioning cluster control model; wherein the air conditioning cluster control model includes an input layer and an output layer; the input layer is used to receive the device state parameter matrix of the air conditioning cluster under the target control cycle, as well as the pre-control ambient temperature matrix and the target ambient temperature matrix at the installation location of each device in the air conditioning cluster; the output layer is used to output the predicted device action parameter matrix of the air conditioning cluster; Based on the feature training set, the initial air-conditioning cluster control model is trained using a reward and punishment mechanism to obtain the air-conditioning cluster control model; wherein, the reward and punishment mechanism is used to optimize the influencing parameter matrix of the predicted temperature mapping relationship according to the accuracy of the predicted temperature mapping relationship from the pre-control ambient temperature matrix to the target predicted ambient temperature matrix; the predicted temperature mapping relationship is used to obtain the target predicted ambient temperature matrix based on the degree of influence of each device in the air-conditioning cluster on the ambient temperature of each installation location of the device and the current ambient temperature at the installation location of each device.

2. The training method according to claim 1, characterized in that The method of training the initial air conditioning cluster control model based on the feature training set and utilizing a reward and punishment mechanism to obtain the air conditioning cluster control model includes: Construct the predicted temperature mapping relationship; wherein the mathematical expression of the predicted temperature mapping relationship includes: X(t+1)=A·X(t)+∑ i [B i ·Q i (t)]; Wherein, X(t+1) is the target predicted ambient temperature matrix of the t+1th control cycle, X(t) is the ambient temperature matrix before control of the tth control cycle, A is the temperature influence matrix containing the intrinsic influence weights between the devices in the air-conditioning cluster, Q i (t) is the state parameter matrix of the equipment of type i in the t-th control cycle, B i is the i-th type of equipment state parameter influence matrix containing the intrinsic influence weights between the equipment in the air-conditioning cluster; Based on the feature training set, construct a linear regression equation group corresponding to the predicted temperature mapping relationship; Solve the linear regression equations based on the least squares method to obtain the temperature influence matrix A and the i-th type equipment state parameter influence matrix B. i The initial value of .

3. The training method according to claim 2, characterized in that The linear regression equations are solved based on the least squares method to obtain the temperature influence matrix A and the i-th type equipment state parameter influence matrix B. i After the initial value of , the method further comprises: Step 11: constructing a predicted temperature mapping equation for this round of training based on the predicted temperature mapping relationship and the feature training set; Step 12: Solve the predicted temperature mapping equation based on the Kalman filter algorithm to obtain the target predicted ambient temperature matrix for this round of training; Step 13, determining whether the difference between the target predicted ambient temperature matrix of this round of training and the target ambient temperature matrix of this round of training is less than a second set threshold; Step 14: If it is not less than, then re-solve the linear regression equations based on the least squares method and update the temperature influence matrix A and the i-th type equipment state parameter influence matrix B. i The value of , and return to step 11; Step 15: If it is less than , the model obtained in this round of training is used as the air-conditioning cluster control model and the training is ended.

4. The training method according to claim 2, characterized in that The types of the equipment state parameter matrix include one or more of an air conditioning specific power matrix, a workshop production equipment capacity matrix, a workshop production product type matrix, an air conditioning refrigeration water temperature matrix, an air conditioning refrigeration water valve opening matrix, an air conditioning fan frequency matrix, an air conditioning fan outlet temperature matrix, a season type matrix, a weather type matrix, a factory external temperature matrix, and a fresh air system valve opening matrix.

5. The training method according to claim 1, wherein: The acquiring of historical data of the air conditioning cluster, performing feature extraction on the historical data, and constructing a feature training set includes: Cleaning and / or filling the historical data of the air-conditioning cluster to obtain pre-processed historical data; Performing feature extraction processing on the pre-processed historical data to obtain original feature data; Performing derivative processing on the original feature data to obtain expanded feature data; Feature screening is performed on the expanded feature data to obtain final feature data and construct the feature training set.

6. The training method according to claim 5, characterized in that The step of screening the expanded feature data to obtain final feature data and constructing the feature training set includes: Performing variance value screening on the expanded feature data to obtain first screened feature data; Performing multicollinearity screening on the first screening feature data to obtain second screening feature data; Based on a hyperparameter optimization algorithm, parameter combinations are performed on the feature types in the second screening feature data, and the second screening feature data are screened according to the results of cross-validation to obtain third screening feature data; Using the XGBoost model, one or more types of feature data that are most sensitive to the target predicted ambient temperature matrix are screened out from the third screened feature data to obtain the final feature data.

7. An air conditioning cluster control method, characterized in that: The method comprises: Obtain the equipment status parameter matrix of the air conditioning cluster; Inputting the target ambient temperature matrix and the device state parameter matrix into the air conditioning cluster control model obtained by the training method according to any one of claims 1 to 6 to obtain a predicted device action parameter matrix; The air conditioning cluster is controlled according to the predicted device action parameter matrix to adjust the ambient temperature at the installation location of each device in the air conditioning cluster to a temperature corresponding to the target ambient temperature matrix.

8. A training device for an air conditioning cluster control model, characterized in that: The training device comprises: A data acquisition module is used to obtain historical data of the air conditioning cluster, perform feature extraction on the historical data, and construct a feature training set; A model building module is configured to construct an initial air conditioning cluster control model; wherein the air conditioning cluster control model includes an input layer and an output layer; the input layer is configured to receive the device state parameter matrix of the air conditioning cluster under the target control cycle, as well as the pre-control ambient temperature matrix and the target ambient temperature matrix at the installation location of each device in the air conditioning cluster; the output layer is configured to output the predicted device action parameter matrix of the air conditioning cluster; A training module is used to train the initial air-conditioning cluster control model based on the feature training set using a reward and punishment mechanism to obtain the air-conditioning cluster control model; wherein the reward and punishment mechanism is used to optimize the influencing parameter matrix of the predicted temperature mapping relationship according to the accuracy of the predicted temperature mapping relationship from the pre-control ambient temperature matrix to the target predicted ambient temperature matrix; the predicted temperature mapping relationship is used to obtain the target predicted ambient temperature matrix based on the degree of influence of each device in the air-conditioning cluster on the ambient temperature of each installation location of the device and the current ambient temperature at the installation location of each device.

9. An air conditioning cluster control device, characterized in that: The device comprises: Parameter acquisition module, used to obtain the equipment status parameter matrix of the air conditioning cluster; An input module, configured to input the target ambient temperature matrix and the device state parameter matrix into the air conditioning cluster control model obtained by the training method provided in the first aspect, to obtain a predicted device action parameter matrix; A control module is used to control the air conditioning cluster according to the predicted device action parameter matrix to adjust the ambient temperature at the installation location of each device in the air conditioning cluster to a temperature corresponding to the target ambient temperature matrix.

10. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement the method according to any one of claims 1 to 7.

Images