Control method and control device of air conditioner, computer readable storage medium and air conditioner

Abstract

The application provides a control method and device of an air conditioner, a computer readable storage medium and the air conditioner, and the method comprises the following steps: obtaining historical air conditioner state parameters of a plurality of air conditioners, the historical air conditioner state parameters being air conditioner state parameters before the current time; clustering the air conditioners according to air conditioner running state parameters to obtain a plurality of air conditioner classes; training a deep neural network by using the air conditioner running state parameters of each air conditioner class and corresponding control strategies to obtain a plurality of strategy models, the strategy models corresponding to the air conditioner classes one by one, and the control strategies comprising start control strategies, stop control strategies and setting strategies of set temperatures; inputting air conditioner state parameters of a target air conditioner into a target strategy model to obtain a control strategy of the target air conditioner, the target strategy model being a strategy model corresponding to an air conditioner class to which the target air conditioner belongs; and controlling the target air conditioner to run according to the control strategy of the target air conditioner, thereby solving the problem that the deep neural network training for generating an air conditioner control strategy in the prior art is time-consuming.

Description

Technical Field

[0001] This invention relates to the field of air conditioning control technology, and more specifically, to an air conditioning control method, control device, computer-readable storage medium, and air conditioner. Background Technology

[0002] Most reinforcement learning methods for air conditioning energy conservation focus on individual devices, such as a single air conditioner or a single building. However, for some large-scale, centralized air conditioning models, such as low-temperature air source heat pump systems (a type of air conditioner) in coal-to-electricity projects, their geographical locations are relatively concentrated (e.g., a village or town), user habits are similar (long-term operation during the heating season with little temperature adjustment), and equipment models are consistent, exhibiting certain homogeneous characteristics. Due to the large number of units and their homogeneous nature, training a deep neural network for each unit to obtain an energy-saving control strategy model is obviously time-consuming. Therefore, a large-scale parallel reinforcement learning framework is urgently needed to accelerate the generation efficiency of energy-saving control strategies. Summary of the Invention

[0003] The main objective of this application is to provide an air conditioner control method, control device, computer-readable storage medium, and air conditioner, so as to at least solve the problem that the training of deep neural networks for generating air conditioner control strategies is time-consuming in the prior art.

[0004] To achieve the above objectives, according to one aspect of this application, an air conditioner control method is provided, comprising: acquiring historical air conditioner state parameters of multiple air conditioners, wherein the historical air conditioner state parameters are air conditioner state parameters prior to the current moment, and the air conditioner state parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV; clustering the air conditioners according to the air conditioner operating state parameters to obtain multiple air conditioner classes, wherein each air conditioner class includes at least one of the air conditioners; training a deep neural network using the air conditioner operating state parameters of each air conditioner class and corresponding control strategies to obtain multiple strategy models, wherein each strategy model corresponds one-to-one with the air conditioner class, and the strategy model is used to determine the optimal control strategy, wherein the control strategy includes a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy; inputting the air conditioner state parameters of a target air conditioner into a target strategy model to obtain the control strategy of the target air conditioner, wherein the target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs; and controlling the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0005] Optionally, the air conditioners are clustered according to the air conditioner operating status parameters to obtain multiple air conditioner classes, including: calculating the average value of each historical air conditioner status parameter to obtain a baseline air conditioner status parameter; calculating the distance between each historical air conditioner status parameter and the baseline air conditioner status parameter to obtain multiple target distances, wherein each target distance corresponds one-to-one with an air conditioner, and the distance is any one of dtw distance, cosine distance, and Euclidean distance;

[0006] Clustering algorithms are used to cluster the target distances to obtain multiple air conditioner clusters. The difference in the target distances between any two air conditioners in an air conditioner cluster is less than a distance threshold.

[0007] Optionally, calculating the average value of each of the historical air conditioning status parameters to obtain the baseline air conditioning status parameters includes: dividing the historical time period into multiple historical sub-time periods, wherein the historical time period is a time period covering the collection time of all the historical air conditioning status parameters; a first calculation step, calculating the average value of all the historical air conditioning status parameters of an air conditioner in each of the historical sub-time periods to obtain a reference air conditioning status parameter group, wherein the reference air conditioning status parameter group includes multiple reference air conditioning status parameters, wherein the reference air conditioning status parameters correspond one-to-one with the historical sub-time periods; repeating the first calculation step at least once until the reference air conditioning status parameter groups of all the air conditioners are obtained; a second calculation step, calculating the average value of the reference air conditioning status parameters of each air conditioner in the target historical sub-time period to obtain the baseline air conditioning status parameters, wherein the target historical sub-time period is any one of the historical sub-time periods; repeating the second calculation step at least once until all the baseline air conditioning status parameters are obtained, wherein the baseline air conditioning status parameters correspond one-to-one with the historical sub-time periods.

[0008] Optionally, calculating the distance between each of the historical air conditioning state parameters and the reference air conditioning state parameters to obtain multiple target distances includes: a third calculation step, calculating the distance between each of the reference air conditioning state parameters of an air conditioner and the corresponding reference air conditioning state parameters to obtain multiple candidate distances; a fourth calculation step, calculating the ratio of the sum of the multiple candidate distances to the number of the historical sub-time periods to obtain a target distance; repeating the third calculation step and the fourth calculation step at least once in sequence until all the target distances are obtained.

[0009] Optionally, a reinforcement learning algorithm is trained using the air conditioner operating state parameters and corresponding control strategies for each of the aforementioned air conditioner types to obtain multiple strategy models, including: an acquisition step, acquiring all training parameters, wherein the training parameters are air conditioner operating state parameters for a target air conditioner type, and the target air conditioner type can be any of the aforementioned air conditioner types; a fifth calculation step, calculating the reward of the control strategy corresponding to each training parameter based on the power consumption and the PMV, wherein the reward is used to measure the contribution of the control strategy to energy saving and human thermal comfort; a determination step, determining the control strategy corresponding to the largest reward as the optimal control strategy; a training step, training the reinforcement learning algorithm based on all the training parameters and the optimal control strategy to obtain the strategy model for the target air conditioner type; and repeating the acquisition step, the fifth calculation step, the determination step, and the training step at least once in sequence to obtain the strategy models for all the aforementioned air conditioner types.

[0010] Optionally, calculating the reward of the control strategy corresponding to each of the training parameters based on the power consumption and the PMV includes: determining the weight of the power consumption and the weight of the PMV, wherein the weights are determined based on multiple trials, the weight of the power consumption is less than 0 and the weight of the PMV is greater than 0; calculating the weighted average of the power consumption and the PMV based on the weight of the power consumption and the weight of the PMV to obtain the reward of the control strategy.

[0011] Optionally, the historical air conditioner status parameters of multiple air conditioners are obtained, including one of the following: obtaining the historical air conditioner status parameters collected by the multiple air conditioners during a historical time period; or obtaining the historical air conditioner status parameters of the multiple air conditioners generated by simulation software, wherein the simulation software is used to simulate the operation of the multiple air conditioners in a corresponding environment.

[0012] According to another aspect of this application, an air conditioner control device is provided, comprising: an acquisition unit, configured to acquire historical air conditioner state parameters of multiple air conditioners, the historical air conditioner state parameters being air conditioner state parameters prior to the current moment, the air conditioner state parameters including exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV; a clustering unit, configured to cluster the air conditioners according to the air conditioner operating state parameters to obtain multiple air conditioner classes, the air conditioner class including at least one of the air conditioners; a training unit, configured to train a deep neural network using the air conditioner operating state parameters of each air conditioner class and corresponding control strategies to obtain multiple strategy models, the strategy models corresponding one-to-one with the air conditioner classes, the strategy models being used to determine the optimal control strategy, the control strategy including a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy; an analysis unit, configured to input the air conditioner state parameters of a target air conditioner into a target strategy model to obtain the control strategy of the target air conditioner, the target strategy model being the strategy model corresponding to the air conditioner class to which the target air conditioner belongs; and a control unit, configured to control the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0013] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform any of the methods described.

[0014] According to another aspect of this application, an air conditioner is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including methods for performing any one of the methods described.

[0015] By applying the technical solution of this application, in the above-mentioned air conditioner control method, multiple air conditioner classes are obtained by clustering multiple historical air conditioner state parameters. The air conditioner state parameters of the air conditioners in the air conditioner classes are relatively similar. Therefore, the air conditioner state parameters of all the air conditioners in the air conditioner classes can be used to train the reinforcement learning algorithm, avoiding the problem of long training time caused by insufficient training samples. This solves the problem of long training time for reinforcement learning algorithms that generate air conditioner control strategies in the prior art. In addition, the training process is usually time-consuming. The multi-threaded / multi-process parallel approach can make full use of the multi-core advantage of the computer and accelerate the training. Attached Figure Description

[0016] Figure 1 A hardware structure block diagram of a mobile terminal for performing an air conditioning control method according to an embodiment of this application is shown;

[0017] Figure 2 A schematic flowchart of an air conditioner control method according to an embodiment of this application is shown;

[0018] Figure 3 A flowchart illustrating the training process of a strategy model according to an embodiment of this application is shown.

[0019] Figure 4 A schematic flowchart of another air conditioner control method provided according to an embodiment of this application is shown;

[0020] Figure 5 A structural block diagram of an air conditioner control device according to an embodiment of this application is shown;

[0021] Figure 6 A flowchart of an air conditioning reinforcement learning method provided according to an embodiment of this application is shown. Detailed Implementation

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] For ease of description, the following explains some of the nouns or terms used in the embodiments of this application:

[0026] PMV, or Predicted Mean Vote, is an indicator of the predicted average number of votes and can measure human thermal comfort.

[0027] As described in the background section, the reinforcement learning algorithms for generating air conditioning control strategies in the prior art are time-consuming to train. To solve this problem, embodiments of this application provide an air conditioning control method, control device, computer-readable storage medium, and air conditioner.

[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0029] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for an air conditioner control method according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0030] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the device information display method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0031] This embodiment provides a method for controlling an air conditioner that runs on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0032] Figure 2 This is a flowchart of an air conditioner control method according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0033] Step S201: Obtain historical air conditioner status parameters of multiple air conditioners. The historical air conditioner status parameters are the air conditioner status parameters before the current time. The air conditioner status parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature and PMV.

[0034] Specifically, the operating status of air conditioners is characterized by exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV, so as to accurately find air conditioners with similar operating status and expand the training sample.

[0035] Step S202: Cluster the air conditioners according to the air conditioner operating status parameters to obtain multiple air conditioner categories, each air conditioner category including at least one of the air conditioners.

[0036] Specifically, clustering algorithms, such as k-means, Gaussian mixture algorithm, and Mean Shift, are used to determine the similarity of the operating status parameters of each air conditioner. Air conditioners with high similarity in their operating status parameters are then grouped into one air conditioner class, resulting in multiple air conditioner classes, each containing one or more air conditioners.

[0037] Step S203: Train a deep neural network using the air conditioner operating status parameters and corresponding control strategies of each of the above-mentioned air conditioner types to obtain multiple strategy models. Each strategy model corresponds to one of the above-mentioned air conditioner types. The strategy models are used to determine the optimal control strategy. The control strategies include a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy.

[0038] Specifically, each air conditioner operating status parameter corresponds to a control strategy. For example, under one air conditioner operating status parameter, the air conditioner executes the start-up control strategy, and under another air conditioner operating status parameter, the air conditioner executes the set temperature setting strategy. By training a deep neural network with all the air conditioner operating status parameters and their control strategies for an air conditioner class, a strategy model that determines the optimal control strategy is obtained. The optimal control strategy is the one that ensures human thermal comfort while consuming the least amount of electricity. For example, if the current environment is no longer cooling, and human thermal comfort is not affected, the shutdown control strategy is the optimal control strategy. Each air conditioner class is trained to obtain a corresponding strategy model.

[0039] Step S204: Input the air conditioner status parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs.

[0040] Specifically, each air conditioner class has a corresponding strategy model. By inputting the air conditioner state parameters of the target air conditioner into the strategy model of its respective air conditioner class, the optimal control strategy can be obtained and used as the control strategy for the target air conditioner.

[0041] Step S205: Control the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0042] Specifically, if the target air conditioner's control strategy is an on-time control strategy, then the target air conditioner will be turned on; if the target air conditioner's control strategy is an off-time control strategy, then the target air conditioner will be turned off; if the target air conditioner's control strategy is a set temperature setting strategy, then the target air conditioner's set temperature will be adjusted to the set temperature corresponding to the control strategy.

[0043] In the aforementioned air conditioner control method, firstly, historical air conditioner state parameters of multiple air conditioners are acquired. These historical air conditioner state parameters are the air conditioner state parameters prior to the current moment, including exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV. Then, the air conditioners are clustered based on these operating state parameters to obtain multiple air conditioner classes, each class including at least one of the aforementioned air conditioners. Next, a deep neural network is trained using the operating state parameters of each air conditioner class and the corresponding control strategy to obtain multiple strategy models. Each strategy model corresponds one-to-one with an air conditioner class and is used to determine the optimal control strategy, including a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy. Then, the air conditioner state parameters of the target air conditioner are input into the target strategy model to obtain the control strategy for the target air conditioner. This target strategy model corresponds to the strategy model of the air conditioner class to which the target air conditioner belongs. Finally, the target air conditioner is controlled to operate according to the aforementioned control strategy. This method clusters multiple historical air conditioner state parameters to obtain multiple air conditioner classes. The air conditioner state parameters of the air conditioners in the air conditioner class are relatively similar. Therefore, the air conditioner state parameters of all the air conditioners in the air conditioner class can be used to train the reinforcement learning algorithm, avoiding the problem of long training time caused by insufficient training samples. This solves the problem of long training time of reinforcement learning algorithms for generating air conditioner control strategies in the prior art. In addition, the training process is usually time-consuming. The multi-threaded / multi-process parallel approach can make full use of the multi-core advantage of the computer and accelerate the training.

[0044] To simplify the clustering process, in one optional implementation, step S202 includes:

[0045] Step S2021: Calculate the average value of each of the above historical air conditioning status parameters to obtain the baseline air conditioning status parameters;

[0046] Step S2022: Calculate the distance between each of the above-mentioned historical air conditioning state parameters and the above-mentioned reference air conditioning state parameters to obtain multiple target distances. The above-mentioned target distances correspond one-to-one with the above-mentioned air conditioning units. The above-mentioned distances are any one of dtw distance, cosine distance and Euclidean distance.

[0047] Step S2023: Clustering algorithm is used to cluster the target distances to obtain multiple air conditioner clusters. The difference between the target distances of any two air conditioners in an air conditioner cluster is less than a distance threshold.

[0048] Specifically, such as Figure 3As shown, the average historical air conditioner status parameters from air conditioner 1 to air conditioner N are used as the baseline air conditioner status parameters. That is, the average historical air conditioner status parameters from 1 to N are calculated to obtain the baseline air conditioner status parameters. The target distances from 1 to N are obtained by calculating the distances between each of the historical air conditioner status parameters and the baseline air conditioner status parameters. These target distances correspond one-to-one with air conditioners 1 to N. Air conditioners 1 to N are divided into K air conditioner classes, i.e., air conditioner class 1 to air conditioner class K, by a clustering algorithm. This ensures that the difference between the target distances of any two air conditioners in a given air conditioner class is less than a distance threshold, thus ensuring that the operating states of the air conditioners in the air conditioner class are highly similar.

[0049] To further ensure the accuracy of clustering, in an optional implementation, step S2021 above includes:

[0050] Step S20211: Divide the historical time period into multiple historical sub-time periods. The aforementioned historical time periods are the time periods that cover the collection times of all the aforementioned historical air conditioning status parameters.

[0051] Step S20212, the first calculation step, calculates the average value of all the above-mentioned historical air conditioner state parameters of the above-mentioned air conditioner in each of the above-mentioned historical sub-time periods, and obtains a reference air conditioner state parameter group. The reference air conditioner state parameter group includes multiple reference air conditioner state parameters, and the reference air conditioner state parameters correspond one-to-one with the above-mentioned historical sub-time periods.

[0052] Step S20213: Repeat the first calculation step above at least once until all the above reference air conditioner state parameter groups of the above air conditioners are obtained;

[0053] Step S20214, the second calculation step, calculate the average value of the reference air conditioner state parameters of each of the above air conditioners in the target historical sub-time period, and obtain the above-mentioned benchmark air conditioner state parameters. The above-mentioned target historical sub-time period is any one of the above-mentioned historical sub-time periods.

[0054] Step S20215: Repeat the second calculation step above at least once until all the above-mentioned reference air conditioning state parameters are obtained, and the above-mentioned reference air conditioning state parameters correspond one-to-one with the above-mentioned historical sub-time periods.

[0055] Specifically, the different collection frequencies of air conditioner status parameters for each air conditioner result in inconsistent status parameters at different points in time, making it difficult to determine their similarity. By dividing the historical time period into multiple historical sub-time periods, for example, dividing a day into 24 hours, and calculating the average of the historical air conditioner status parameters for each hour, each hour can correspond to a historical air conditioner status parameter, denoted as the reference air conditioner status parameter. The average of the reference air conditioner status parameters corresponding to all air conditioners in this hour can be used to obtain the baseline air conditioner status parameter. A baseline air conditioner status parameter can be calculated for each hour. Through historical sub-time periods, each air conditioner has 24 reference air conditioner status parameters, which are not affected by the collection frequency and are more representative than the historical air conditioner status parameters at a single moment. This makes the determined baseline air conditioner status parameter more accurate and the clustering effect better.

[0056] To further ensure the accuracy of clustering, in an optional implementation, step S2022 above includes:

[0057] Step S20221, the third calculation step, calculates the distance between each of the above-mentioned reference air conditioner state parameters and the corresponding above-mentioned baseline air conditioner state parameters of an air conditioner, and obtains multiple candidate distances;

[0058] Step S20222, the fourth calculation step, calculates the ratio of the sum of multiple candidate distances to the number of historical sub-time periods to obtain a target distance.

[0059] Step S20223: Repeat the above third calculation step and the above fourth calculation step at least once, until all the above target distances are obtained.

[0060] Specifically, since the baseline air conditioner state parameters correspond one-to-one with the historical sub-time periods, a single historical sub-time period cannot reflect the air conditioner's operating status throughout the day. By calculating a distance for each historical sub-time period and averaging the distances as the target distances, the similarity of the air conditioner's operating status throughout the day can be reflected, resulting in better clustering effects.

[0061] To further achieve energy savings, in one optional implementation, step S203 includes:

[0062] Step S2031, obtaining all training parameters, wherein the training parameters are one of the air conditioner operation status parameters of the target air conditioner class, and the target air conditioner class is any one of the air conditioner classes;

[0063] Step S2032, the fifth calculation step, calculates the reward of the control strategy corresponding to each of the above training parameters based on the above power consumption and the above PMV. The reward is used to measure the contribution of the control strategy to energy saving and human thermal comfort.

[0064] Step S2033, Determine the optimal control strategy by determining the control strategy corresponding to the largest of the above-mentioned rewards.

[0065] Step S2034, training step: Train the reinforcement learning algorithm according to all the above training parameters and the above optimal control strategy to obtain the above strategy model of the target air conditioner class.

[0066] Step S2035: Repeat the above acquisition step, the above fifth calculation step, the above determination step, and the above training step at least once in sequence to obtain all the above-mentioned strategy models for the above-mentioned air conditioning types.

[0067] Specifically, such as Figure 3 As shown, the air conditioner operating status parameters of the air conditioner class through the data channel are all universal. The air conditioner operating status parameters are usually represented by S, and the corresponding actions of the control strategy are represented by A, which are instructions that cause changes in the environment, such as turning on / off and setting the temperature. The reward R is a scalar, which can be obtained by the reward function π(a|s) and is used to measure the quality of the action A corresponding to the control strategy. The training objective is to maximize the cumulative reward ∑R. The reward function π can be expressed as minimizing the power consumption of the air conditioner while satisfying the user's thermal comfort. The control strategy corresponding to the maximum reward is determined as the optimal control strategy. The reinforcement learning algorithm is trained based on all training parameters and the optimal control strategy to obtain the strategy model of the target air conditioner class. Each air conditioner class corresponds to a strategy model, namely strategy model 1 to strategy model K. For example, strategy model 1 can generate control strategies for each air conditioner 1 to air conditioner M, namely S11 to S1M. The energy-saving purpose can be achieved by using the control strategies (set of actions) generated by the strategy model.

[0068] To simplify the calculation, in an optional implementation, step S2032 above includes:

[0069] Step S20321: Determine the weight of the power consumption and the weight of the PMV. The weights are determined based on multiple experiments. The weight of the power consumption is less than 0 and the weight of the PMV is greater than 0.

[0070] Step S20322: Calculate the weighted average of the power consumption and PMV based on the weight of the power consumption and PMV to obtain the reward of the control strategy.

[0071] Specifically, the formula for the above reward function is π(a|s)=-α*power_consumption+β*PMV, where power_consumption represents the power consumption of the air conditioner, PMV is used to measure human thermal comfort, α and β are decimals between 0 and 1, -α is the power consumption weight, and β is the PMV weight, in order to balance the contribution of the two to the reward value. α and β can be obtained through multiple experiments, and the calculation process is simple and convenient, improving training efficiency.

[0072] It should be noted that α + β = 1. Once one is determined, the other is also determined. α and β can be determined using a lookup table, such as selecting one α sequentially from the table below [0.1, 0.2, 0.33, 0.36, ... 0.9, 1.0] to verify the effect. Of course, the table can also contain other data increasing from 0 to 1. Alternatively, a stepwise search method can be used for α and β. First, determine the initial value (e.g., 0) and the search step size (e.g., 0.03). Use a loop to select α sequentially to determine the optimal coefficient (which may also be a local optimum). The basis for determining the optimal coefficient is that, under the condition that other factors remain unchanged, this coefficient obtains the maximum reward or reaches a threshold.

[0073] To further enrich the training samples, in one optional implementation, step S201 above includes one of the following:

[0074] Step S2011: Obtain the historical air conditioner status parameters collected from multiple air conditioners during historical time periods;

[0075] Step S2012: Obtain the historical air conditioner status parameters of the multiple air conditioners generated by the simulation software. The simulation software is used to simulate the operation of the multiple air conditioners in the corresponding environment.

[0076] Specifically, historical air conditioning state parameters can be recorded from actual air conditioning operation or generated through simulation, further enriching the training samples. The environment can be a real environment or a virtual environment. For low-temperature air source heat pump systems in the air conditioning category, training reinforcement learning algorithms with real unit equipment is too costly and impractical. Simulating the state of air conditioning in a virtual environment is more feasible. To construct a virtual low-temperature air source heat pump system environment model, building simulation software such as EnergyPlus, DeST, and Transys can be used, or data-driven methods such as ARIMA, LSTM, and XGBoost can be used to predict future states based on historical unit states.

[0077] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the air conditioner control method of this application will be described in detail below with reference to specific embodiments.

[0078] This embodiment relates to a specific air conditioner control method, such as... Figure 4 As shown, it includes the following steps:

[0079] Step S1: Data preprocessing. Obtain historical air conditioner status parameters for multiple air conditioners. These historical air conditioner status parameters are those prior to the current moment, including exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV. Divide the historical time period into multiple historical sub-time periods, which are time periods covering the collection times of all the historical air conditioner status parameters. First calculation step: Calculate the average value of all the historical air conditioner status parameters for one of the air conditioners in each of the historical sub-time periods to obtain a reference air conditioner status parameter group. The reference air conditioner status parameter group includes multiple reference air conditioner status parameters, which correspond one-to-one with the historical sub-time periods. Repeat the first calculation step at least once until the reference air conditioner status parameter groups for all the air conditioners are obtained.

[0080] Step S2: Second calculation step, calculate the average value of the reference air conditioner status parameters of each of the above air conditioners in the target historical sub-time period to obtain the above benchmark air conditioner status parameters, the above target historical sub-time period is any of the above historical sub-time periods; repeat the above second calculation step at least once until all the above benchmark air conditioner status parameters are obtained, the above benchmark air conditioner status parameters correspond one-to-one with the above historical sub-time periods.

[0081] Step S3: Determine the historical sub-time period as the target time period;

[0082] Step S4: Third calculation step, calculate the distance between the reference air conditioner state parameter and the corresponding baseline air conditioner state parameter of the air conditioner in the target time period to obtain the target distance; repeat the third calculation step at least once until all the target distances are obtained, the target distances correspond one-to-one with the air conditioners, and the distances are any one of dtw distance, cosine distance and Euclidean distance;

[0083] Step S5: Use a clustering algorithm to cluster the target distances to obtain multiple air conditioner clusters. The difference between the target distances corresponding to any two air conditioners in an air conditioner cluster is less than a distance threshold.

[0084] Step S6: Train a deep neural network using the air conditioner operating status parameters and corresponding control strategies of each of the above-mentioned air conditioner types to obtain multiple strategy models. Each strategy model corresponds to one of the above-mentioned air conditioner types. The strategy models are used to determine the optimal control strategy. The control strategies include the start-up control strategy, the shutdown control strategy, and the set temperature setting strategy.

[0085] Step S7: Input the above air conditioning state parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioning class to which the target air conditioner belongs.

[0086] Step S8: Control the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0087] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0088] This application also provides an air conditioner control device. It should be noted that the air conditioner control device of this application embodiment can be used to execute the air conditioner control device method provided in this application embodiment. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0089] The control device for an air conditioner provided in the embodiments of this application will be described below.

[0090] Figure 5 This is a structural block diagram of an air conditioner control device according to an embodiment of this application. Figure 5 As shown, the device includes:

[0091] The acquisition unit 10 is used to acquire historical air conditioner status parameters of multiple air conditioners. The historical air conditioner status parameters are the air conditioner status parameters before the current time. The air conditioner status parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature and PMV.

[0092] Specifically, the operating status of air conditioners is characterized by exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV, so as to accurately find air conditioners with similar operating status and expand the training sample.

[0093] Clustering unit 20 is used to cluster the air conditioners according to the air conditioner operating status parameters to obtain multiple air conditioner classes, wherein each air conditioner class includes at least one of the air conditioners.

[0094] Specifically, clustering algorithms, such as k-means, Gaussian mixture algorithm, and Mean Shift, are used to determine the similarity of the operating status parameters of each air conditioner. Air conditioners with high similarity in their operating status parameters are then grouped into one air conditioner class, resulting in multiple air conditioner classes, each containing one or more air conditioners.

[0095] Training unit 30 is used to train a deep neural network using the air conditioner operating state parameters and corresponding control strategies of each of the above-mentioned air conditioner types to obtain multiple strategy models. Each strategy model corresponds to one of the above-mentioned air conditioner types. The strategy model is used to determine the optimal control strategy. The control strategy includes a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy.

[0096] Specifically, each air conditioner operating status parameter corresponds to a control strategy. For example, under one air conditioner operating status parameter, the air conditioner executes the start-up control strategy, and under another air conditioner operating status parameter, the air conditioner executes the set temperature setting strategy. By training a deep neural network with all the air conditioner operating status parameters and their control strategies for an air conditioner class, a strategy model that determines the optimal control strategy is obtained. The optimal control strategy is the one that ensures human thermal comfort while minimizing power consumption. For example, if the current environment is no longer cooling, and human thermal comfort is not affected, the shutdown control strategy is the optimal control strategy. Each air conditioner class is trained to obtain a corresponding strategy model.

[0097] Analysis unit 40 is used to input the air conditioning state parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioning class to which the target air conditioner belongs.

[0098] Specifically, each air conditioner class has a corresponding strategy model. By inputting the air conditioner state parameters of the target air conditioner into the strategy model of its respective air conditioner class, the optimal control strategy can be obtained and used as the control strategy for the target air conditioner.

[0099] Control unit 50 is used to control the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0100] Specifically, if the target air conditioner's control strategy is an on-time control strategy, then the target air conditioner will be turned on; if the target air conditioner's control strategy is an off-time control strategy, then the target air conditioner will be turned off; if the target air conditioner's control strategy is a set temperature setting strategy, then the target air conditioner's set temperature will be adjusted to the set temperature corresponding to the control strategy.

[0101] In the aforementioned air conditioner control device, the acquisition unit acquires historical air conditioner state parameters of multiple air conditioners. These historical air conditioner state parameters are the air conditioner state parameters prior to the current moment, including exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV. The clustering unit clusters the air conditioners according to the aforementioned air conditioner operating state parameters, obtaining multiple air conditioner classes, each class including at least one of the aforementioned air conditioners. The training unit trains a deep neural network using the air conditioner operating state parameters of each air conditioner class and the corresponding control strategy, obtaining multiple strategy models. Each strategy model corresponds one-to-one with the aforementioned air conditioner class, and the strategy model is used to determine the optimal control strategy, including a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy. The analysis unit inputs the air conditioner state parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs. The control unit controls the operation of the target air conditioner according to the aforementioned control strategy. This device clusters multiple historical air conditioner state parameters to obtain multiple air conditioner classes. The air conditioner state parameters of the air conditioners in the air conditioner classes are relatively similar. Therefore, the air conditioner state parameters of all the air conditioners in the air conditioner classes can be used to train the reinforcement learning algorithm, avoiding the problem of long training time caused by insufficient training samples. This solves the problem of long training time of reinforcement learning algorithms for generating air conditioner control strategies in the prior art. In addition, the training process is usually time-consuming. The multi-threaded / multi-process parallel approach can make full use of the multi-core advantage of the computer and accelerate the training.

[0102] To simplify the clustering process, in one optional implementation, the clustering unit includes:

[0103] The first calculation module is used to calculate the average value of each of the above-mentioned historical air conditioning state parameters to obtain the baseline air conditioning state parameters.

[0104] The second calculation module is used to calculate the distance between each of the above-mentioned historical air conditioning state parameters and the above-mentioned baseline air conditioning state parameters, and obtain multiple target distances. The above-mentioned target distances correspond one-to-one with the above-mentioned air conditioning. The above-mentioned distances are any one of dtw distance, cosine distance and Euclidean distance.

[0105] The clustering module is used to cluster the target distances using a clustering algorithm to obtain multiple air conditioner clusters. The difference between the target distances corresponding to any two air conditioners in an air conditioner cluster is less than a distance threshold.

[0106] Specifically, such as Figure 3As shown, the average historical air conditioner status parameters from air conditioner 1 to air conditioner N are used as the baseline air conditioner status parameters. That is, the average historical air conditioner status parameters from 1 to N are calculated to obtain the baseline air conditioner status parameters. The target distances from 1 to N are obtained by calculating the distances between each of the historical air conditioner status parameters and the baseline air conditioner status parameters. These target distances correspond one-to-one with air conditioners 1 to N. Air conditioners 1 to N are divided into K air conditioner classes, i.e., air conditioner class 1 to air conditioner class K, by a clustering algorithm. This ensures that the difference between the target distances of any two air conditioners in a given air conditioner class is less than a distance threshold, thus ensuring that the operating states of the air conditioners in the air conditioner class are highly similar.

[0107] To further ensure the accuracy of clustering, in one optional implementation, the first calculation module includes:

[0108] The processing submodule is used to divide the historical time period into multiple historical sub-time periods, which are time periods that cover the collection times of all the above-mentioned historical air conditioning status parameters.

[0109] The first calculation submodule is used to perform the first calculation step, calculate the average value of all the above-mentioned historical air conditioner status parameters of the above-mentioned air conditioner in each of the above-mentioned historical sub-time periods, and obtain a reference air conditioner status parameter group. The reference air conditioner status parameter group includes multiple reference air conditioner status parameters, and the reference air conditioner status parameters correspond one-to-one with the above-mentioned historical sub-time periods.

[0110] The first repeating submodule is used to repeat the first calculation step above at least once until all the above-mentioned reference air conditioner state parameter groups of the above-mentioned air conditioners are obtained.

[0111] The second calculation submodule is used to perform the second calculation step, calculate the average value of the reference air conditioner state parameters of each of the above air conditioners in the target historical sub-time period, and obtain the reference air conditioner state parameters. The target historical sub-time period is any one of the above historical sub-time periods.

[0112] The second repeating submodule is used to repeat the second calculation step above at least once until all the above-mentioned reference air conditioning state parameters are obtained, and the above-mentioned reference air conditioning state parameters correspond one-to-one with the above-mentioned historical sub-time periods.

[0113] Specifically, the different collection frequencies of air conditioner status parameters for each air conditioner result in inconsistent status parameters at different points in time, making it difficult to determine their similarity. By dividing the historical time period into multiple historical sub-time periods, for example, dividing a day into 24 hours, and calculating the average of the historical air conditioner status parameters for each hour, each hour can correspond to a historical air conditioner status parameter, denoted as the reference air conditioner status parameter. The average of the reference air conditioner status parameters corresponding to all air conditioners in this hour can be used to obtain the baseline air conditioner status parameter. A baseline air conditioner status parameter can be calculated for each hour. Through historical sub-time periods, each air conditioner has 24 reference air conditioner status parameters, which are not affected by the collection frequency and are more representative than the historical air conditioner status parameters at a single moment. This makes the determined baseline air conditioner status parameter more accurate and the clustering effect better.

[0114] To further ensure the accuracy of clustering, in one optional implementation, the second calculation module includes:

[0115] The third calculation submodule is used to perform the third calculation step, calculate the distance between each of the above-mentioned reference air conditioner state parameters and the corresponding above-mentioned baseline air conditioner state parameters of an air conditioner, and obtain multiple candidate distances.

[0116] The fourth calculation submodule is used to perform the fourth calculation step, which calculates the ratio of the sum of multiple candidate distances to the number of historical sub-time periods to obtain a target distance.

[0117] The third repeating submodule is used to repeat the third calculation step and the fourth calculation step at least once in sequence until all the target distances are obtained.

[0118] Specifically, since the baseline air conditioner state parameters correspond one-to-one with the historical sub-time periods, a single historical sub-time period cannot reflect the air conditioner's operating status throughout the day. By calculating a distance for each historical sub-time period and averaging the distances as the target distances, the similarity of the air conditioner's operating status throughout the day can be reflected, resulting in better clustering effects.

[0119] To further achieve energy savings, in one optional implementation, the training unit includes:

[0120] The first acquisition module is used to perform the acquisition step and acquire all training parameters. The training parameters are the air conditioner operation status parameters of the target air conditioner class, and the target air conditioner class is any air conditioner class.

[0121] The third calculation module is used to perform the fifth calculation step, which calculates the reward of the control strategy corresponding to each of the above training parameters based on the above power consumption and the above PMV. The reward is used to measure the contribution of the control strategy to energy saving and human thermal comfort.

[0122] The determination module is used to perform determination steps to determine the optimal control strategy corresponding to the highest of the above-mentioned rewards.

[0123] The training module is used to execute the training steps, train the reinforcement learning algorithm based on all the above training parameters and the above optimal control strategy, and obtain the above strategy model for the target air conditioner class.

[0124] The repeat module is used to repeat the above acquisition step, the above fifth calculation step, the above determination step and the above training step at least once in sequence to obtain all the above-mentioned strategy models of the above-mentioned air conditioning types.

[0125] Specifically, such as Figure 3 As shown, the air conditioner operating status parameters of the air conditioner class through the data channel are all universal. The air conditioner operating status parameters are usually represented by S, and the corresponding actions of the control strategy are represented by A, which are instructions that cause changes in the environment, such as turning on / off and setting the temperature. The reward R is a scalar, which can be obtained by the reward function π(a|s) and is used to measure the quality of the action A corresponding to the control strategy. The training objective is to maximize the cumulative reward ∑R. The reward function π can be expressed as minimizing the power consumption of the air conditioner while satisfying the user's thermal comfort. The control strategy corresponding to the maximum reward is determined as the optimal control strategy. The reinforcement learning algorithm is trained based on all training parameters and the optimal control strategy to obtain the strategy model of the target air conditioner class. Each air conditioner class corresponds to a strategy model, namely strategy model 1 to strategy model K. For example, strategy model 1 can generate control strategies for each air conditioner 1 to air conditioner M, namely S11 to S1M. The energy-saving purpose can be achieved by using the control strategies (set of actions) generated by the strategy model.

[0126] To simplify the calculation, in one optional implementation, the third calculation module includes:

[0127] The determination submodule is used to determine the weight of the power consumption and the weight of the PMV. The weights are determined based on multiple experiments, and the weight of the power consumption is less than 0 and the weight of the PMV is greater than 0.

[0128] The fifth calculation submodule is used to calculate the weighted average of the power consumption and PMV based on the weight of the power consumption and the weight of PMV, so as to obtain the reward of the control strategy.

[0129] Specifically, the formula for the above reward function is π(a|s)=-α*power_consumption+β*PMV, where power_consumption represents the power consumption of the air conditioner, PMV is used to measure human thermal comfort, α and β are decimals between 0 and 1, -α is the power consumption weight, and β is the PMV weight, in order to balance the contribution of the two to the reward value. α and β can be obtained through multiple experiments, and the calculation process is simple and convenient, improving training efficiency.

[0130] It should be noted that α + β = 1. Once one is determined, the other is also determined. α and β can be determined using a lookup table, such as selecting one α sequentially from the table below [0.1, 0.2, 0.33, 0.36, ... 0.9, 1.0] to verify the effect. Of course, the table can also contain other data increasing from 0 to 1. Alternatively, a stepwise search method can be used for α and β. First, determine the initial value (e.g., 0) and the search step size (e.g., 0.03). Use a loop to select α sequentially to determine the optimal coefficient (which may also be a local optimum). The basis for determining the optimal coefficient is that, under the condition that other factors remain unchanged, this coefficient obtains the maximum reward or reaches a threshold.

[0131] To further enrich the training samples, in one optional implementation, the acquisition unit includes one of the following:

[0132] The second acquisition module is used to acquire the historical air conditioner status parameters collected from multiple air conditioners during historical time periods.

[0133] The third acquisition module is used to acquire the historical air conditioner status parameters of the multiple air conditioners generated by the simulation software. The simulation software is used to simulate the operation of the multiple air conditioners in the corresponding environment.

[0134] Specifically, historical air conditioning state parameters can be recorded from actual air conditioning operation or generated through simulation, further enriching the training samples. The environment can be a real environment or a virtual environment. For low-temperature air source heat pump systems in the air conditioning category, training reinforcement learning algorithms with real unit equipment is too costly and impractical. Simulating the state of air conditioning in a virtual environment is more feasible. To construct a virtual low-temperature air source heat pump system environment model, building simulation software such as EnergyPlus, DeST, and Transys can be used, or data-driven methods such as ARIMA, LSTM, and XGBoost can be used to predict future states based on historical unit states.

[0135] The control device for the aforementioned air conditioner includes a processor and a memory. The acquisition unit, clustering unit, training unit, analysis unit, and control unit are all stored as program units in the memory. The processor executes these program units stored in the memory to achieve the corresponding functions. All of the above modules reside in the same processor; alternatively, the modules may be located in different processors in any combination.

[0136] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can address the problem of long training times in reinforcement learning algorithms for generating air conditioning control strategies in existing technologies.

[0137] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0138] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the air conditioner control method.

[0139] Specifically, the control methods for air conditioning include:

[0140] Step S201: Obtain historical air conditioner status parameters of multiple air conditioners. The historical air conditioner status parameters are the air conditioner status parameters before the current time. The air conditioner status parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature and PMV.

[0141] Specifically, the operating status of air conditioners is characterized by exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV, so as to accurately find air conditioners with similar operating status and expand the training sample.

[0142] Step S202: Cluster the air conditioners according to the air conditioner operating status parameters to obtain multiple air conditioner categories, each air conditioner category including at least one of the air conditioners.

[0143] Specifically, clustering algorithms, such as k-means, Gaussian mixture algorithm, and Mean Shift, are used to determine the similarity of the operating status parameters of each air conditioner. Air conditioners with high similarity in their operating status parameters are then grouped into one air conditioner class, resulting in multiple air conditioner classes, each containing one or more air conditioners.

[0144] Step S203: Train a deep neural network using the air conditioner operating status parameters and corresponding control strategies of each of the above-mentioned air conditioner types to obtain multiple strategy models. Each strategy model corresponds to one of the above-mentioned air conditioner types. The strategy models are used to determine the optimal control strategy. The control strategies include a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy.

[0145] Specifically, each air conditioner operating status parameter corresponds to a control strategy. For example, under one air conditioner operating status parameter, the air conditioner executes the start-up control strategy, and under another air conditioner operating status parameter, the air conditioner executes the set temperature setting strategy. By training a deep neural network with all the air conditioner operating status parameters and their control strategies for an air conditioner class, a strategy model that determines the optimal control strategy is obtained. The optimal control strategy is the one that ensures human thermal comfort while consuming the least amount of electricity. For example, if the current environment is no longer cooling, and human thermal comfort is not affected, the shutdown control strategy is the optimal control strategy. Each air conditioner class is trained to obtain a corresponding strategy model.

[0146] Step S204: Input the air conditioner status parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs.

[0147] Specifically, each air conditioner class has a corresponding strategy model. By inputting the air conditioner state parameters of the target air conditioner into the strategy model of its respective air conditioner class, the optimal control strategy can be obtained and used as the control strategy for the target air conditioner.

[0148] Step S205: Control the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0149] Specifically, if the target air conditioner's control strategy is an on-time control strategy, then the target air conditioner will be turned on; if the target air conditioner's control strategy is an off-time control strategy, then the target air conditioner will be turned off; if the target air conditioner's control strategy is a set temperature setting strategy, then the target air conditioner's set temperature will be adjusted to the set temperature corresponding to the control strategy.

[0150] This invention provides a processor for running a program, wherein the program executes the air conditioner control method during operation.

[0151] Step S201: Obtain historical air conditioner status parameters of multiple air conditioners. The historical air conditioner status parameters are the air conditioner status parameters before the current time. The air conditioner status parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature and PMV.

[0152] Specifically, the operating status of air conditioners is characterized by exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV, so as to accurately find air conditioners with similar operating status and expand the training sample.

[0153] Step S202: Cluster the air conditioners according to the air conditioner operating status parameters to obtain multiple air conditioner categories, each air conditioner category including at least one of the air conditioners.

[0154] Specifically, clustering algorithms, such as k-means, Gaussian mixture algorithm, and Mean Shift, are used to determine the similarity of the operating status parameters of each air conditioner. Air conditioners with high similarity in their operating status parameters are then grouped into one air conditioner class, resulting in multiple air conditioner classes, each containing one or more air conditioners.

[0155] Step S203: Train a deep neural network using the air conditioner operating status parameters and corresponding control strategies of each of the above-mentioned air conditioner types to obtain multiple strategy models. Each strategy model corresponds to one of the above-mentioned air conditioner types. The strategy models are used to determine the optimal control strategy. The control strategies include a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy.

[0156] Specifically, each air conditioner operating status parameter corresponds to a control strategy. For example, under one air conditioner operating status parameter, the air conditioner executes the start-up control strategy, and under another air conditioner operating status parameter, the air conditioner executes the set temperature setting strategy. By training a deep neural network with all the air conditioner operating status parameters and their control strategies for an air conditioner class, a strategy model that determines the optimal control strategy is obtained. The optimal control strategy is the one that ensures human thermal comfort while consuming the least amount of electricity. For example, if the current environment is no longer cooling, and human thermal comfort is not affected, the shutdown control strategy is the optimal control strategy. Each air conditioner class is trained to obtain a corresponding strategy model.

[0157] Step S204: Input the air conditioner status parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs.

[0158] Specifically, each air conditioner class has a corresponding strategy model. By inputting the air conditioner state parameters of the target air conditioner into the strategy model of its respective air conditioner class, the optimal control strategy can be obtained and used as the control strategy for the target air conditioner.

[0159] Step S205: Control the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0160] Specifically, if the target air conditioner's control strategy is an on-time control strategy, then the target air conditioner will be turned on; if the target air conditioner's control strategy is an off-time control strategy, then the target air conditioner will be turned off; if the target air conditioner's control strategy is a set temperature setting strategy, then the target air conditioner's set temperature will be adjusted to the set temperature corresponding to the control strategy.

[0161] This invention provides an air conditioner, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps:

[0162] Step S201: Obtain historical air conditioner status parameters of multiple air conditioners. The historical air conditioner status parameters are the air conditioner status parameters before the current time. The air conditioner status parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature and PMV.

[0163] Specifically, the operating status of air conditioners is characterized by exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV, so as to accurately find air conditioners with similar operating status and expand the training sample.

[0164] Step S202: Cluster the air conditioners according to the air conditioner operating status parameters to obtain multiple air conditioner categories, each air conditioner category including at least one of the air conditioners.

[0165] Specifically, clustering algorithms, such as k-means, Gaussian mixture algorithm, and Mean Shift, are used to determine the similarity of the operating status parameters of each air conditioner. Air conditioners with high similarity in their operating status parameters are then grouped into one air conditioner class, resulting in multiple air conditioner classes, each containing one or more air conditioners.

[0166] Step S203: Train a deep neural network using the air conditioner operating status parameters and corresponding control strategies of each of the above-mentioned air conditioner types to obtain multiple strategy models. Each strategy model corresponds to one of the above-mentioned air conditioner types. The strategy models are used to determine the optimal control strategy. The control strategies include a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy.

[0167] Specifically, each air conditioner operating status parameter corresponds to a control strategy. For example, under one air conditioner operating status parameter, the air conditioner executes the start-up control strategy, and under another air conditioner operating status parameter, the air conditioner executes the set temperature setting strategy. By training a deep neural network with all the air conditioner operating status parameters and their control strategies for an air conditioner class, a strategy model that determines the optimal control strategy is obtained. The optimal control strategy is the one that ensures human thermal comfort while consuming the least amount of electricity. For example, if the current environment is no longer cooling, and human thermal comfort is not affected, the shutdown control strategy is the optimal control strategy. Each air conditioner class is trained to obtain a corresponding strategy model.

[0168] Step S204: Input the air conditioner status parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs.

[0169] Specifically, each air conditioner class has a corresponding strategy model. By inputting the air conditioner state parameters of the target air conditioner into the strategy model of its respective air conditioner class, the optimal control strategy can be obtained and used as the control strategy for the target air conditioner.

[0170] Step S205: Control the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0171] Specifically, if the target air conditioner's control strategy is an on-time control strategy, then the target air conditioner will be turned on; if the target air conditioner's control strategy is an off-time control strategy, then the target air conditioner will be turned off; if the target air conditioner's control strategy is a set temperature setting strategy, then the target air conditioner's set temperature will be adjusted to the set temperature corresponding to the control strategy.

[0172] Optionally, step S202 includes: step S2021, calculating the average value of each of the historical air conditioner state parameters to obtain the baseline air conditioner state parameters; step S2022, calculating the distance between each of the historical air conditioner state parameters and the baseline air conditioner state parameters to obtain multiple target distances, wherein each target distance corresponds one-to-one with an air conditioner, and the distance is any one of dtw distance, cosine distance, and Euclidean distance; step S2023, using a clustering algorithm to cluster the target distances to obtain multiple air conditioner classes, wherein the difference between the target distances corresponding to any two air conditioners in an air conditioner class is less than a distance threshold.

[0173] Optionally, step S2021 includes: step S20211, dividing the historical time period into multiple historical sub-time periods, wherein the historical time period is the time period covering the collection time of all the historical air conditioning status parameters; step S20212, a first calculation step, calculating the average value of all the historical air conditioning status parameters of one of the air conditioners in each of the historical sub-time periods to obtain a reference air conditioning status parameter group, wherein the reference air conditioning status parameter group includes multiple reference air conditioning status parameters, and the reference air conditioning status parameters correspond one-to-one with the historical sub-time periods; step S20213, repeating the first calculation step at least once until the reference air conditioning status parameter group of all the air conditioners is obtained; step S20214, a second calculation step, calculating the average value of the reference air conditioning status parameters of each of the air conditioners in a target historical sub-time period to obtain the benchmark air conditioning status parameters, wherein the target historical sub-time period is any one of the historical sub-time periods; step S20215, repeating the second calculation step at least once until all the benchmark air conditioning status parameters are obtained, and the benchmark air conditioning status parameters correspond one-to-one with the historical sub-time periods.

[0174] Optionally, step S2022 includes: step S20221, a third calculation step, calculating the distance between each of the above-mentioned reference air conditioner state parameters and the corresponding above-mentioned baseline air conditioner state parameters of an air conditioner to obtain multiple candidate distances; step S20222, a fourth calculation step, calculating the ratio of the sum of the multiple candidate distances to the number of the above-mentioned historical sub-time periods to obtain a target distance; step S20223, repeating the above-mentioned third calculation step and the above-mentioned fourth calculation step at least once in sequence until all the above-mentioned target distances are obtained.

[0175] Optionally, step S203 includes: step S2031, an acquisition step, acquiring all training parameters, wherein the training parameters are one of the air conditioner operation state parameters of the target air conditioner class, and the target air conditioner class is any of the air conditioner classes; step S2032, a fifth calculation step, calculating the reward of the control strategy corresponding to each of the training parameters based on the power consumption and PMV, wherein the reward is used to measure the contribution of the control strategy to energy saving and human thermal comfort; step S2033, a determination step, determining the control strategy corresponding to the largest reward as the optimal control strategy; step S2034, a training step, training the reinforcement learning algorithm based on all the training parameters and the optimal control strategy to obtain the strategy model of the target air conditioner class; and step S2035, repeating the acquisition step, the fifth calculation step, the determination step, and the training step at least once in sequence to obtain the strategy models of all the air conditioner classes.

[0176] Optionally, step S2032 includes: step S20321, determining the weight of the power consumption and the weight of the PMV, wherein the weights are determined based on multiple experiments, the weight of the power consumption is less than 0 and the weight of the PMV is greater than 0; step S20322, calculating the weighted average of the power consumption and the PMV based on the weights of the power consumption and the PMV to obtain the reward of the control strategy.

[0177] Optionally, step S201 includes one of the following: step S2011, obtaining the historical air conditioner status parameters collected by the multiple air conditioners in a historical time period; step S2012, obtaining the historical air conditioner status parameters of the multiple air conditioners generated by simulation software, wherein the simulation software is used to simulate the operation of the multiple air conditioners in the corresponding environment.

[0178] In addition, the process of air conditioning performing reinforcement learning to train a deep neural network is as follows: Figure 6 As shown, the intelligent agent refers to the carrier of the deep reinforcement learning algorithm, which can be the air conditioner itself or the air conditioner's backend server. It can run relevant algorithms based on the environmental state and rewards to generate actions corresponding to the control strategy. The predicted air conditioner virtual environment is an environmental model that simulates the state of the air conditioner equipment by constructing an environment model through prediction methods (such as LSTM, XGBoost, ARIMAX, etc.). It can generate corresponding states and rewards for the intelligent agent's actions. The action At is the instruction generated by the intelligent agent, such as turning the machine on or off, adjusting the set temperature, etc. The subscript t represents a certain moment. The state St, St+1: the state of the air conditioner at a certain moment, such as temperature, pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, etc. The reward Rt, Rt+1 is the reward value at a certain moment, which is a scalar and is used to measure the quality of the action.

[0179] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps:

[0180] Step S201: Obtain historical air conditioner status parameters of multiple air conditioners. The historical air conditioner status parameters are the air conditioner status parameters before the current time. The air conditioner status parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature and PMV.

[0181] Specifically, the operating status of air conditioners is characterized by exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV, so as to accurately find air conditioners with similar operating status and expand the training sample.

[0182] Step S202: Cluster the air conditioners according to the air conditioner operating status parameters to obtain multiple air conditioner categories, each air conditioner category including at least one of the air conditioners.

[0183] Specifically, clustering algorithms, such as k-means, Gaussian mixture algorithm, and Mean Shift, are used to determine the similarity of the operating status parameters of each air conditioner. Air conditioners with high similarity in their operating status parameters are then grouped into one air conditioner class, resulting in multiple air conditioner classes, each containing one or more air conditioners.

[0184] Step S203: Train a deep neural network using the air conditioner operating status parameters and corresponding control strategies of each of the above-mentioned air conditioner types to obtain multiple strategy models. Each strategy model corresponds to one of the above-mentioned air conditioner types. The strategy models are used to determine the optimal control strategy. The control strategies include a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy.

[0185] Specifically, each air conditioner operating status parameter corresponds to a control strategy. For example, under one air conditioner operating status parameter, the air conditioner executes the start-up control strategy, and under another air conditioner operating status parameter, the air conditioner executes the set temperature setting strategy. By training a deep neural network with all the air conditioner operating status parameters and their control strategies for an air conditioner class, a strategy model that determines the optimal control strategy is obtained. The optimal control strategy is the one that ensures human thermal comfort while minimizing power consumption. For example, if the current environment is no longer cooling, and human thermal comfort is not affected, the shutdown control strategy is the optimal control strategy. Each air conditioner class is trained to obtain a corresponding strategy model.

[0186] Step S204: Input the air conditioner status parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs.

[0187] Specifically, each air conditioner class has a corresponding strategy model. By inputting the air conditioner state parameters of the target air conditioner into the strategy model of its respective air conditioner class, the optimal control strategy can be obtained and used as the control strategy for the target air conditioner.

[0188] Step S205: Control the operation of the target air conditioner according to the control strategy of the target air conditioner.

[0189] Specifically, if the target air conditioner's control strategy is an on-time control strategy, then the target air conditioner will be turned on; if the target air conditioner's control strategy is an off-time control strategy, then the target air conditioner will be turned off; if the target air conditioner's control strategy is a set temperature setting strategy, then the target air conditioner's set temperature will be adjusted to the set temperature corresponding to the control strategy.

[0190] Optionally, step S202 includes: step S2021, calculating the average value of each of the historical air conditioner state parameters to obtain the baseline air conditioner state parameters; step S2022, calculating the distance between each of the historical air conditioner state parameters and the baseline air conditioner state parameters to obtain multiple target distances, wherein each target distance corresponds one-to-one with an air conditioner, and the distance is any one of dtw distance, cosine distance, and Euclidean distance; step S2023, using a clustering algorithm to cluster the target distances to obtain multiple air conditioner classes, wherein the difference between the target distances corresponding to any two air conditioners in an air conditioner class is less than a distance threshold.

[0191] Optionally, step S2021 includes: step S20211, dividing the historical time period into multiple historical sub-time periods, wherein the historical time period is the time period covering the collection time of all the historical air conditioning status parameters; step S20212, a first calculation step, calculating the average value of all the historical air conditioning status parameters of one of the air conditioners in each of the historical sub-time periods to obtain a reference air conditioning status parameter group, wherein the reference air conditioning status parameter group includes multiple reference air conditioning status parameters, and the reference air conditioning status parameters correspond one-to-one with the historical sub-time periods; step S20213, repeating the first calculation step at least once until the reference air conditioning status parameter group of all the air conditioners is obtained; step S20214, a second calculation step, calculating the average value of the reference air conditioning status parameters of each of the air conditioners in a target historical sub-time period to obtain the benchmark air conditioning status parameters, wherein the target historical sub-time period is any one of the historical sub-time periods; step S20215, repeating the second calculation step at least once until all the benchmark air conditioning status parameters are obtained, and the benchmark air conditioning status parameters correspond one-to-one with the historical sub-time periods.

[0192] Optionally, step S2022 includes: step S20221, a third calculation step, calculating the distance between each of the above-mentioned reference air conditioner state parameters and the corresponding above-mentioned baseline air conditioner state parameters of an air conditioner to obtain multiple candidate distances; step S20222, a fourth calculation step, calculating the ratio of the sum of the multiple candidate distances to the number of the above-mentioned historical sub-time periods to obtain a target distance; step S20223, repeating the above-mentioned third calculation step and the above-mentioned fourth calculation step at least once in sequence until all the above-mentioned target distances are obtained.

[0193] Optionally, step S203 includes: step S2031, an acquisition step, acquiring all training parameters, wherein the training parameters are one of the air conditioner operation state parameters of the target air conditioner class, and the target air conditioner class is any of the air conditioner classes; step S2032, a fifth calculation step, calculating the reward of the control strategy corresponding to each of the training parameters based on the power consumption and PMV, wherein the reward is used to measure the contribution of the control strategy to energy saving and human thermal comfort; step S2033, a determination step, determining the control strategy corresponding to the largest reward as the optimal control strategy; step S2034, a training step, training the reinforcement learning algorithm based on all the training parameters and the optimal control strategy to obtain the strategy model of the target air conditioner class; and step S2035, repeating the acquisition step, the fifth calculation step, the determination step, and the training step at least once in sequence to obtain the strategy models of all the air conditioner classes.

[0194] Optionally, step S2032 includes: step S20321, determining the weight of the power consumption and the weight of the PMV, wherein the weights are determined based on multiple experiments, the weight of the power consumption is less than 0 and the weight of the PMV is greater than 0; step S20322, calculating the weighted average of the power consumption and the PMV based on the weights of the power consumption and the PMV to obtain the reward of the control strategy.

[0195] Optionally, step S201 includes one of the following: step S2011, acquiring historical air conditioner status parameters collected from multiple air conditioners over a historical time period; step S2012, acquiring the historical air conditioner status parameters of multiple air conditioners generated by simulation software, wherein the simulation software is used to simulate the operation of multiple air conditioners in a corresponding environment. Obviously, those skilled in the art should understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0196] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0197] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0198] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0199] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0200] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0201] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0202] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0203] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0204] As can be seen from the above description, the embodiments of this application achieve the following technical effects:

[0205] 1) In the air conditioner control method of this application, firstly, historical air conditioner state parameters of multiple air conditioners are obtained. The historical air conditioner state parameters are the air conditioner state parameters before the current time. The air conditioner state parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV. Then, the air conditioners are clustered according to the air conditioner operating state parameters to obtain multiple air conditioner classes. Each air conditioner class includes at least one of the air conditioners. Next, a deep neural network is trained using the air conditioner operating state parameters of each air conditioner class and the corresponding control strategy to obtain multiple strategy models. Each strategy model corresponds one-to-one with the air conditioner class. The strategy models are used to determine the optimal control strategy. The control strategy includes a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy. Then, the air conditioner state parameters of the target air conditioner are input into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs. Finally, the target air conditioner is controlled to operate according to the control strategy. This method clusters multiple historical air conditioner state parameters to obtain multiple air conditioner classes. The air conditioner state parameters of the air conditioners in the air conditioner class are relatively similar. Therefore, the air conditioner state parameters of all the air conditioners in the air conditioner class can be used to train the reinforcement learning algorithm, avoiding the problem of long training time caused by insufficient training samples. This solves the problem of long training time of reinforcement learning algorithms for generating air conditioner control strategies in the prior art. In addition, the training process is usually time-consuming. The multi-threaded / multi-process parallel approach can make full use of the multi-core advantage of the computer and accelerate the training.

[0206] 2) In the air conditioner control device of this application, the acquisition unit acquires historical air conditioner state parameters of multiple air conditioners. The historical air conditioner state parameters are the air conditioner state parameters before the current time. The air conditioner state parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV. The clustering unit clusters the air conditioners according to the air conditioner operating state parameters to obtain multiple air conditioner classes. Each air conditioner class includes at least one of the air conditioners. The training unit trains a deep neural network using the air conditioner operating state parameters of each air conditioner class and the corresponding control strategy to obtain multiple strategy models. Each strategy model corresponds one-to-one with the air conditioner class. The strategy model is used to determine the optimal control strategy. The control strategy includes a start-up control strategy, a shutdown control strategy, and a set temperature setting strategy. The analysis unit inputs the air conditioner state parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs. The control unit controls the operation of the target air conditioner according to the control strategy. This device clusters multiple historical air conditioner state parameters to obtain multiple air conditioner classes. The air conditioner state parameters of the air conditioners in the air conditioner classes are relatively similar. Therefore, the air conditioner state parameters of all the air conditioners in the air conditioner classes can be used to train the reinforcement learning algorithm, avoiding the problem of long training time caused by insufficient training samples. This solves the problem of long training time of reinforcement learning algorithms for generating air conditioner control strategies in the prior art. In addition, the training process is usually time-consuming. The multi-threaded / multi-process parallel approach can make full use of the multi-core advantage of the computer and accelerate the training.

[0207] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for controlling an air conditioner, characterized in that, include: Obtain historical air conditioner status parameters for multiple air conditioners. These historical air conditioner status parameters are those prior to the current moment, and include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV. The air conditioners are clustered according to the air conditioner status parameters to obtain multiple air conditioner classes, and each air conditioner class includes at least one of the air conditioners. A deep neural network is trained using the air conditioner state parameters and corresponding control strategies of each of the aforementioned air conditioner types to obtain multiple strategy models. Each strategy model corresponds one-to-one with the air conditioner type. The strategy model is used to determine the optimal control strategy, which includes a start-up control strategy, a stop-up control strategy, and a set temperature setting strategy. The air conditioner status parameters of the target air conditioner are input into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs. The target air conditioner is controlled to operate according to the control strategy of the target air conditioner. The air conditioners are clustered according to the air conditioner status parameters to obtain multiple air conditioner categories, including: calculating the average value of each of the historical air conditioner status parameters to obtain the baseline air conditioner status parameters; Calculate the distance between each of the historical air conditioner state parameters and the baseline air conditioner state parameters to obtain multiple target distances. Each target distance corresponds one-to-one with an air conditioner. The distance can be any one of dtw distance, cosine distance, and Euclidean distance. Use a clustering algorithm to cluster the target distances to obtain multiple air conditioner classes. The difference between the target distances corresponding to any two air conditioners in an air conditioner class is less than a distance threshold.

2. The method according to claim 1, characterized in that, Calculate the average value of each of the historical air conditioning state parameters to obtain the baseline air conditioning state parameters, including: The historical time period is divided into multiple historical sub-time periods, and the historical time period is the time period that covers the collection time of all the historical air conditioning status parameters; The first calculation step is to calculate the average value of all the historical air conditioner status parameters of the air conditioner in each of the historical sub-time periods to obtain a reference air conditioner status parameter group. The reference air conditioner status parameter group includes multiple reference air conditioner status parameters, and the reference air conditioner status parameters correspond one-to-one with the historical sub-time periods. Repeat the first calculation step at least once until all the reference air conditioner state parameter groups of the air conditioners are obtained; The second calculation step is to calculate the average value of the reference air conditioner state parameters corresponding to each air conditioner in the target historical sub-time period, and obtain the benchmark air conditioner state parameters. The target historical sub-time period is any one of the historical sub-time periods. Repeat the second calculation step at least once until all the reference air conditioning state parameters are obtained, and the reference air conditioning state parameters correspond one-to-one with the historical sub-time periods.

3. The method according to claim 2, characterized in that, Calculate the distance between each of the historical air conditioning state parameters and the baseline air conditioning state parameters to obtain multiple target distances, including: The third calculation step is to calculate the distance between each of the reference air conditioner state parameters and the corresponding base air conditioner state parameters of the air conditioner to obtain multiple candidate distances; The fourth calculation step is to calculate the ratio of the sum of multiple candidate distances to the number of historical sub-time periods to obtain a target distance; Repeat the third and fourth calculation steps at least once in sequence until all the target distances are obtained.

4. The method according to claim 1, characterized in that, A reinforcement learning algorithm is trained using the air conditioner state parameters and corresponding control strategies for each of the aforementioned air conditioner types to obtain multiple strategy models, including: The acquisition step involves acquiring all training parameters, where the training parameters are air conditioner state parameters of a target air conditioner class, and the target air conditioner class can be any air conditioner class. The fifth calculation step is to calculate the reward of the control strategy corresponding to each of the training parameters based on the power consumption and the PMV. The reward is used to measure the contribution of the control strategy to energy saving and human thermal comfort. The steps are defined to determine the optimal control strategy corresponding to the maximum reward. The training step involves training the reinforcement learning algorithm based on all the training parameters and the optimal control strategy to obtain the strategy model for the target air conditioner class. The acquisition step, the fifth calculation step, the determination step, and the training step are repeated at least once in sequence to obtain the strategy models for all air conditioning types.

5. The method according to claim 4, characterized in that, The reward for the control strategy corresponding to each training parameter is calculated based on the power consumption and the PMV, including: The weights of the power consumption and the PMV are determined based on multiple experiments, wherein the weight of the power consumption is less than 0 and the weight of the PMV is greater than 0. The weighted average of the power consumption and the PMV is calculated based on the weight of the power consumption and the weight of the PMV to obtain the reward of the control strategy.

6. The method according to any one of claims 1 to 5, characterized in that, Obtain historical air conditioner status parameters for multiple air conditioners, including one of the following: Obtain the historical air conditioner status parameters collected over a historical time period from multiple air conditioners; The historical air conditioner status parameters of multiple air conditioners generated by simulation software are obtained, and the simulation software is used to simulate the operation of multiple air conditioners in a corresponding environment.

7. A control device for an air conditioner, characterized in that, include: The acquisition unit is used to acquire historical air conditioner status parameters of multiple air conditioners. The historical air conditioner status parameters are the air conditioner status parameters before the current time. The air conditioner status parameters include exhaust pressure, power consumption, compressor frequency, indoor temperature, outdoor temperature, and PMV. A clustering unit is used to cluster the air conditioners according to the air conditioner status parameters to obtain multiple air conditioner classes, wherein each air conditioner class includes at least one of the air conditioners. The training unit is used to train a deep neural network using the air conditioner state parameters and corresponding control strategies of each air conditioner class to obtain multiple strategy models. Each strategy model corresponds one-to-one with the air conditioner class. The strategy model is used to determine the optimal control strategy. The control strategy includes a start-up control strategy, a stop-up control strategy, and a set temperature setting strategy. The analysis unit is used to input the air conditioner status parameters of the target air conditioner into the target strategy model to obtain the control strategy of the target air conditioner. The target strategy model is the strategy model corresponding to the air conditioner class to which the target air conditioner belongs. The control unit is used to control the operation of the target air conditioner according to the control strategy of the target air conditioner. The clustering unit includes: a first calculation module, used to calculate the average value of each of the historical air conditioner state parameters to obtain a baseline air conditioner state parameter; a second calculation module, used to calculate the distance between each of the historical air conditioner state parameters and the baseline air conditioner state parameter to obtain multiple target distances, wherein each target distance corresponds one-to-one with an air conditioner, and the distance is any one of dtw distance, cosine distance, and Euclidean distance; and a clustering module, used to cluster the target distances using a clustering algorithm to obtain multiple air conditioner classes, wherein the difference between the target distances corresponding to any two air conditioners in an air conditioner class is less than a distance threshold.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method according to any one of claims 1 to 6.

9. An air conditioner, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising methods for performing any one of claims 1 to 6.

Images