A fluorine pump air conditioner temperature stabilization control method and system

Abstract

The present invention provides a fluorine pump air conditioner temperature stabilization control method and system. When formulating a control strategy for a target air conditioner, the target analysis element data set of the air conditioner objective parameter elements and the first air conditioner control record data are analyzed. During the analysis, the air conditioner implicit array of the target air conditioner can represent the needs of the user of the target air conditioner for the air conditioner control strategy and the needs of the involved air conditioners of the target air conditioner for the air conditioner control strategy. In this way, a more comprehensive air conditioner implicit array representing the needs of the user of the target air conditioner is obtained. When the target control strategy is obtained in a control command set based on the air conditioner implicit array of the target air conditioner, the air conditioner control strategy can be comprehensively determined.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a method and system for controlling the temperature stability of a fluorine pump air conditioner. Background Art

[0002] With the advancement of science and technology and the improvement of people's living standards, air conditioning equipment has become an indispensable part of modern life, especially in commercial buildings, offices and residential buildings. In order to provide a more comfortable and energy-saving indoor environment, intelligent control of air conditioning systems is particularly important. However, most existing air conditioning control systems rely on preset temperature settings and simple switch control logic, lacking comprehensive consideration of user personalized needs and environmental changes, resulting in unsatisfactory control effects.

[0003] In the existing technology, although some advanced air conditioning control systems have begun to try to introduce sensor data and environmental parameters for feedback control, these systems are often limited to the status monitoring of a single device and fail to effectively integrate the mutual influence between multiple devices and the user's behavior habits, thus limiting the further improvement of the overall performance of the system. In addition, when formulating control strategies, traditional control systems mostly rely on fixed algorithms or rules, lacking flexibility and adaptability, and are difficult to cope with complex and changing usage scenarios and user needs.

[0004] On the other hand, the rapid development of the Internet of Things, big data and artificial intelligence technologies has provided new possibilities for the intelligent upgrade of air-conditioning control systems. By collecting and analyzing a large amount of air-conditioning operation data, user behavior data and environmental parameter data, we can deeply understand the actual needs of users for air-conditioning control, and then formulate more accurate and efficient control strategies. However, how to effectively collect, process and use this data while ensuring data privacy and security is an important challenge facing the intelligent upgrade of air-conditioning control systems. Summary of the invention

[0005] The object of the present invention is to provide a method and system for stabilizing the temperature of a fluorine pump air conditioner. The present invention is achieved by:

[0006] In the first aspect, the present invention provides a method for stabilizing the temperature of a fluorine pump air conditioner, comprising: obtaining a target analysis factor data set of a target air-conditioning device to be controlled and a data set of involved analysis factors of air-conditioning devices involved in the target air-conditioning device, wherein the target analysis factor data set includes: an air-conditioning objective parameter element and a first air-conditioning control record data; obtaining an objective parameter implicit array of the air-conditioning objective parameter element based on an objective parameter characterization component of an edge side network, and obtaining a control record implicit array of the first air-conditioning control record data based on a control record characterization component of an edge side network; obtaining a fused implicit array corresponding to the air-conditioning objective parameter element and the involved analysis factor data set based on a weight focusing component of an edge side network; integrating the objective parameter implicit array, the control record implicit array and the fused implicit array based on an integration component of an edge side network to obtain an air-conditioning device implicit array of the target air-conditioning device; obtaining a target control strategy from a control command set based on the air-conditioning device implicit array of the target air-conditioning device, and controlling the target air-conditioning device based on the target control strategy.

[0007] In a second aspect, the present invention provides a computer system comprising: one or more processors; a memory; one or more computer programs; wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method as described above is implemented.

[0008] Beneficial effects of the present invention: When formulating a control strategy for a target air-conditioning device, the present invention analyzes the target analysis element data set of the air-conditioning objective parameter elements and the first air-conditioning control record data, and then mines and extracts the objective parameter implicit array of the air-conditioning objective parameter elements based on the objective parameter characterization component of the edge side network, and mines and extracts the control record implicit array of the first air-conditioning control record data based on the control record characterization component of the edge side network, and refers to the characteristics of the target air-conditioning device of the objective parameter implicit array and the control record implicit array, so as to characterize the needs of the user of the target air-conditioning device for the air-conditioning control strategy. In addition, based on obtaining the involved analysis element data set of the involved air-conditioning device of the target air-conditioning device, and dynamically learning the needs of the user of the involved air-conditioning device for the air-conditioning control strategy in the air-conditioning objective parameter elements and the involved analysis element data set based on the weight focusing component of the edge side network, for example, it is reflected by the fused implicit array corresponding to the air-conditioning objective parameter elements and the involved analysis element data set output by the weight focusing component of the edge side network. Finally, the integration component based on the edge network integrates the objective parameter implicit array, the control record implicit array and the fusion implicit array to obtain the air-conditioning equipment implicit array of the target air-conditioning equipment. In other words, the air-conditioning equipment implicit array of the target air-conditioning equipment can together represent the needs of the user of the target air-conditioning equipment for the air-conditioning control strategy and the needs of the involved air-conditioning equipment of the target air-conditioning equipment for the air-conditioning control strategy. In this way, a more comprehensive air-conditioning equipment implicit array representing the needs of the user of the target air-conditioning equipment is obtained. When the target control strategy is obtained in the control command set based on the air-conditioning equipment implicit array of the target air-conditioning equipment, the air-conditioning control strategy can be comprehensively determined. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 The present invention provides a flow chart of a method for controlling the temperature stability of a fluorine pump air conditioner.

[0010] Figure 2 It is a schematic diagram of the composition of a computer system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0011] The executor of the fluorine pump air conditioner temperature stabilization control method in the embodiment of the present invention is a computer system, including but not limited to a server, a personal computer, a notebook computer, a tablet computer, a smart phone, etc. Figure 1 As shown, the method includes:

[0012] Step S100: obtaining a target analysis factor data set of a target air-conditioning device to be controlled and a data set of analysis factors of related air-conditioning devices of the target air-conditioning device, wherein the target analysis factor data set includes: air-conditioning objective parameter factors and first air-conditioning control record data.

[0013] In step S100, for the target analysis element data set, the air conditioner objective parameter element and the first air conditioner control record data are included. The air conditioner objective parameter element covers the parameters related to the inherent attributes and operating status of the air conditioner itself. For example, the power parameter of the air conditioner may be different values ​​such as 1.5 horsepower or 3 horsepower, which is an important indicator reflecting the cooling or heating capacity of the air conditioner; the cooling capacity is also an objective parameter element, such as the common 3500W cooling capacity and other values, which is directly related to the amount of heat that the air conditioner can transfer per unit time; and the energy efficiency ratio of the air conditioner, such as 3.0 or higher values, the higher the energy efficiency ratio, the better the cooling or heating effect of the air conditioner when consuming the same amount of electricity. In addition to the objective parameters of the air conditioner, the objective parameter elements of the air conditioner can also include environmental data parameters of the air conditioner, such as ambient temperature, humidity, etc.

[0014] The first air conditioning control record data records the past control operation of the target air conditioning device. The involved analysis element data set of the involved air conditioning device is also an important content that the computer system needs to obtain. The involved air conditioning device refers to other air conditioning devices that have some association with the target air conditioning device. For example, in a large office space, there may be a mutual influence relationship between the air conditioning devices in multiple offices. The air conditioning device in one office is the target air conditioning device, and the air conditioning devices in other offices may be involved air conditioning devices. The computer system obtains the involved analysis element data set of the involved air conditioning device, and the content in this data set is the integration of the relevant information of the involved air conditioning device. Each involved air conditioning device has its corresponding involved analysis element data, which can reflect the relationship between the involved air conditioning device and the target air conditioning device and some characteristics and control conditions of the involved air conditioning device itself. For example, in a multi-room building environment, if there is heat exchange between the room where the target air conditioning device is located and the air conditioning device (involved air conditioning device) in the adjacent room, then some operating parameters of the involved air conditioning device, such as temperature, operating mode, and other information, as well as the relative position relationship between the involved air conditioning device and the target air conditioning device (which may affect heat conduction or air flow, etc.) will be included in the involved analysis element data set.

[0015] The target analysis factor dataset provides a data source for the subsequent analysis of the target air-conditioning equipment's own characteristics and user usage habits; the involved analysis factor dataset helps the computer system consider the relationship between the target air-conditioning equipment and other related equipment, so that it can start from a more comprehensive perspective when formulating control strategies. Only by accurately obtaining these datasets can the computer system perform various subsequent data processing operations based on the edge network, such as mining various implicit arrays of the target air-conditioning equipment, and comprehensively considering the target air-conditioning equipment and the involved air-conditioning equipment to determine the final air-conditioning control strategy.

[0016] Step S200: obtaining an objective parameter implicit array of air conditioner objective parameter elements based on an objective parameter representation component of an edge side network, and obtaining a control record implicit array of first air conditioner control record data based on a control record representation component of an edge side network.

[0017] In step S200, the objective parameter characterization component based on the edge side network obtains the objective parameter implicit array of the objective parameter elements of the air conditioner. The edge side network includes an embedded coding component, and the objective parameter characterization component includes a feedforward network component. The computer system first embeds the objective parameter implicit array obtained by embedding the objective parameter elements of the air conditioner by the embedded coding component based on the edge side network as the loading data of the objective parameter characterization component.

[0018] Take the objective parameter elements of air conditioners such as power, cooling capacity and energy efficiency ratio as an example. Assume that the power of the air conditioner is 3 horsepower, the cooling capacity is 7200W, and the energy efficiency ratio is 3.2. The computer system processes these values ​​through embedded coding components. Embedded coding can convert these actual values ​​into a coding form suitable for further processing by the computer system. For example, the values ​​of power, cooling capacity and energy efficiency ratio are mapped to a specific coding space according to certain rules to obtain the objective parameter embedded implicit array. Each element in this array may represent specific information after these parameters are encoded.

[0019] Then, the computer system performs information mapping on the objective parameter embedding implicit array based on the feedforward network component of the objective parameter characterization component, thereby obtaining the objective parameter implicit array of the air conditioning objective parameter elements. The feedforward network component has multiple neurons and connection weights, and it will calculate the input objective parameter embedding implicit array according to a predefined algorithm. For example, each neuron in the feedforward network component may perform a weighted summation of the input encoded information and perform a nonlinear transformation through an activation function. Assuming that the objective parameter embedding implicit array is [10, 20, 30] (this is just an assumed encoding value for the convenience of example), the weight matrix in the feedforward network component is [[0.5, 0.3, 0.2], [0.1, 0.6, 0.3], [0.4, 0.2, 0.4]], the neuron calculates the sum of the product of the input value and the weight, and then after processing by the activation function (such as the ReLU function), obtains a new value, and finally forms the objective parameter implicit array. This implicit array of objective parameters can characterize the objective parameter elements of air conditioning from a new perspective and explore the potential relationships and characteristics between these parameters.

[0020] Secondly, the computer system obtains the control record implicit array of the first air-conditioning control record data based on the control record representation component of the edge network. The first air-conditioning control record data includes a plurality of multi-dimensional control command data arranged in time sequence according to the control moment. The computer system first embeds the plurality of air-conditioning control records of the first air-conditioning control record data obtained by the embedded coding component based on the edge network into the implicit array as the loading data of the control record representation component of the edge network.

[0021] For example, the first air conditioning control record data records the temperature and wind speed control commands at different times of the day. At 8 a.m., the temperature is set to 25°C and the wind speed is low; at 12 noon, the temperature is adjusted to 23°C and the wind speed is high, etc. The computer system converts these temperature and wind speed control commands at different times into multiple air conditioning control record embedded implicit arrays through embedded coding components. Each array may contain coding information for temperature and wind speed, as well as coding information related to the control time.

[0022] Next, the computer system processes multiple air-conditioning control record embedded implicit arrays based on the control record representation component of the edge side network to obtain the control record implicit array of the first air-conditioning control record data. The control record representation component of the edge side network can obtain the control record implicit array based on multiple multi-dimensional control command data with a control duration greater than a preset duration. For example, if the preset duration is 2 hours, the computer system will select the air-conditioning control record embedded implicit array corresponding to the control record data with a control duration greater than 2 hours for processing. Alternatively, the control record representation component of the edge side network can also be based on its Transformer component to obtain multiple air-conditioning control record mapping implicit arrays corresponding to the multiple air-conditioning control record embedded implicit arrays, and then based on the downsampling module of the control record representation component, the multiple air-conditioning control record mapping implicit arrays are processed to obtain the control record implicit array of the first air-conditioning control record data. Assuming that multiple air-conditioning control records are embedded in implicit arrays [[15, 5], [20, 10], [25, 15]], the Transformer component will map these arrays, which may change the values ​​and relationships of the elements in the arrays, and obtain multiple air-conditioning control record mapping implicit arrays, such as [[18, 6], [22, 12], [26, 16]], and then the downsampling module will downsample these mapping implicit arrays, such as selecting some elements or merging elements according to certain rules, and finally obtain the control record implicit array. This control record implicit array can mine the potential patterns and features in the first air-conditioning control record data, reflecting the user's habits and preferences for air-conditioning control in different time periods.

[0023] Step S300: Obtaining a fused implicit array corresponding to the air-conditioning objective parameter elements and the involved analysis element data set based on the weight focusing component of the edge side network.

[0024] In step S300, the involved analysis element data set includes x involved analysis element data corresponding to x involved air conditioning equipment (x≥1), the edge side network includes an embedded coding component, and the weight focusing component includes x equipment weight focusing modules, a Transformer component, and a downsampling module. The computer system first embeds the objective parameter implicit array obtained by embedding the air conditioning objective parameter elements based on the embedded coding component of the edge side network, and the x involved embedded coding implicit arrays obtained by embedding the x involved analysis element data based on the embedded coding component of the edge side network as the loading data of the weight focusing component.

[0025] For example, suppose there are two involved air conditioners, and the air conditioner power in the objective parameter element of the air conditioner is 3 horsepower, the cooling capacity is 7200W, and the energy efficiency ratio is 3.2. After being encoded by the embedded coding component, the objective parameter embedded implicit array [10, 20, 30] is obtained. For the involved air conditioner 1, its corresponding involved analysis element data such as the operating time is 8 hours and the current set temperature is 24℃, after encoding, the involved embedded coding implicit array [5, 15] is obtained; the involved analysis element data of the involved air conditioner 2 such as the operating mode is cooling and the wind speed is low wind speed, after encoding, the involved embedded coding implicit array [8, 3] is obtained.

[0026] Next, the computer system processes the objective parameter embedding implicit array and the x involved embedded coding implicit arrays based on the x device weight focusing modules of the weight focusing component to obtain x involved weight implicit arrays. Taking the processing of one of the device weight focusing modules as an example, for the involved embedded coding implicit array involving air-conditioning equipment 1 in the above example, the computer system obtains the Kronecker product array between the objective parameter embedding implicit array and the involved embedded coding implicit array. Assume that the Kronecker product array of the objective parameter embedding implicit array [10, 20, 30] and the involved embedded coding implicit array [5, 15] is [50, 150, 100, 300, 150, 450] (the calculation process is to multiply the corresponding elements to obtain a new array). Then, the objective parameter embedding implicit array, the first involved embedding coding implicit array and the Kronecker product array are concatenated to obtain a new array, such as [10, 20, 30, 5, 15, 50, 150, 100, 300, 150, 450], and the concatenated array is loaded into the nonlinear transformation unit of the first device weight focusing module to obtain the execution result of the unit. This execution result is then determined as the loading data of the fully connected unit of the first device weight focusing module, and the first involved weight implicit array is obtained based on the fully connected unit, such as [25, 35].

[0027] Afterwards, the computer system obtains the implicit array of involvement mapping corresponding to the x implicit arrays of involvement weights based on the Transformer component of the weight focusing component. The Transformer component can perform feature extraction and transformation on the input implicit array of involvement weights, change the relationship and value between array elements, for example, convert the implicit array of involvement weights [25,35] obtained above into [30, 40] to obtain the implicit array of involvement mapping. Finally, the computer system processes the implicit array of involvement mapping based on the downsampling module of the weight focusing component to obtain a fused implicit array. The downsampling module may perform data compression or feature extraction on the implicit array of involvement mapping according to certain rules, such as obtaining a fused implicit array

[35] (for example only) after downsampling the implicit array of involvement mapping [30, 40]. This fused implicit array can comprehensively reflect the relationship between the objective parameter elements of air conditioning and the data of involvement analysis elements, and provide a data basis for subsequent operations.

[0028] Step S400: integrating the objective parameter implicit array, the control record implicit array and the fusion implicit array based on the integration component of the edge side network to obtain the air-conditioning equipment implicit array of the target air-conditioning equipment.

[0029] In step S400, first, the related arrays obtained in the previous steps are reviewed. The implicit array of objective parameters is obtained by the computer system after processing the objective parameter elements of the air conditioner based on the objective parameter characterization component of the edge network. For example, it is assumed that the objective parameter elements of the air conditioner include the power, cooling capacity and energy efficiency ratio of the air conditioner. The power is 3 horsepower, the cooling capacity is 7200W, and the energy efficiency ratio is 3.2. After a series of processing (such as embedded coding, information mapping of the feedforward network component, etc.), the implicit array of objective parameters obtained may be [10, 20, 30] (here is only a hypothetical value for the convenience of example). This array represents the potential relationship and characteristics between the objective parameter elements of the air conditioner in a form that has been mined and converted. The implicit array of control records is obtained by the computer system after processing the first air conditioner control record data based on the control record characterization component of the edge network. The first air conditioner control record data contains multi-dimensional control command data such as temperature and wind speed at different times. For example, the temperature is set to 25°C and the wind speed is low at 8 o'clock in the morning of the day; the temperature is adjusted to 23°C and the wind speed is high at 12 noon, etc. After embedding coding, possible Transformer components and downsampling modules, the obtained implicit array of control records is assumed to be [5, 10, 15] (also an assumed value for the convenience of example). This array reflects the user's habits and patterns of air conditioning control operations and other potential information.

[0030] The fused implicit array is obtained by the computer system in step S300 based on the weight focusing component of the edge network, which comprehensively considers the relationship between the objective parameter elements of the air conditioner and the involved analysis element data set. For example, the involved analysis element data set contains relevant information of other air conditioners (involved air conditioners) associated with the target air conditioner, such as the operating time and set temperature of the involved air conditioner. After a series of operations (including embedded coding, device weight focusing module, Transformer component and downsampling module processing), the fused implicit array is obtained, which is assumed to be [8, 12].

[0031] Then, the computer system integrates the three implicit arrays using the integration component of the edge network. The integration component may use a specific algorithm or rule to merge the information in these arrays. One possible way is to perform a weighted sum of the elements in the three arrays according to certain weights. For example, the weight of the objective parameter implicit array is set to 0.3, the weight of the control record implicit array is set to 0.4, and the weight of the fusion implicit array is set to 0.3.

[0032] For the objective parameter implicit array [10, 20, 30], after weighting, we get [10×0.3, 20×0.3, 30×0.3]=[3, 6, 9]; for the control record implicit array [5, 10, 15], after weighting, we get [5×0.4, 10×0.4, 15×0.4]=[2, 4, 6]; for the fusion implicit array [8, 12], after weighting, we get [8×0.3, 12×0.3]=[2.4,3.6].

[0033] Next, the weighted array elements are merged. A simple way to merge is to directly add the elements at the corresponding positions to obtain a new array [3+2+2.4,6+4+3.6,9+6]=[7.4,13.6,15]. This new array is the implicit array of the air conditioner of the target air conditioner. Through such an integration process, the implicit array of the air conditioner combines the objective parameters of the air conditioner itself, the user control records, and the information related to the involved air conditioners. It can comprehensively reflect the various characteristics and related relationships of the target air conditioner. For example, from this array, it is possible to mine the comprehensive state of the air conditioner under specific user control habits (reflected in the control record implicit array), combined with the air conditioner's own performance (objective parameter implicit array) and the influence of related air conditioners (fusion implicit array).

[0034] Step S500: based on the air-conditioning equipment implicit array of the target air-conditioning equipment, obtaining a target control strategy from a control command set, and controlling the target air-conditioning equipment based on the target control strategy.

[0035] In step S500, first, the implicit array of the air conditioner of the target air conditioner is obtained by integrating the implicit array of objective parameters, the implicit array of control records, and the fusion implicit array in the previous step (S400). This implicit array of air conditioners integrates information such as the objective parameter elements of the air conditioner, the first air conditioner control record data, and the data set of the involved analysis elements. For example, assuming that the implicit array of the air conditioner is [20, 30, 40], each element in this array contains specific information related to the target air conditioner, which may include a comprehensive representation of factors such as air conditioner performance, user usage habits, and related equipment impacts.

[0036] The control command set stores multiple air conditioning control strategies, and each air conditioning control strategy has a corresponding control strategy implicit array based on the central side network. For example, there are three air conditioning control strategies in the control command set, namely strategy A, strategy B, and strategy C. For strategy A, its control strategy implicit array may be [15, 25, 35]; the control strategy implicit array of strategy B is [22, 32, 42]; and the control strategy implicit array of strategy C is [18, 28, 38].

[0037] The computer system needs to obtain the target control strategy from the control command set. The control strategy implicit array of this target control strategy matches the air conditioning equipment implicit array of the target air conditioning equipment. The matching method may be based on a certain distance measurement or similarity calculation method. For example, the Euclidean distance is used to calculate the distance between the air conditioning equipment implicit array [20, 30, 40] and each control strategy implicit array. Calculate the Euclidean distance with the control strategy implicit array [15, 25, 35] of strategy A:

[0038] ;

[0039] Calculate the Euclidean distance with the implicit array [22, 32, 42] of the control strategy of strategy B:

[0040] ;

[0041] Calculate the Euclidean distance to the implicit array [18, 28, 38] of the control policy of policy C:

[0042] ;

[0043] Here, the Euclidean distances of strategies B and C with the implicit array of the target air-conditioning equipment are the same and the smallest (assuming that in actual situations there are more strategies, the only strategy with the smallest distance is selected, this is just an example), and the computer system can determine strategy B as the target control strategy based on other preset rules, such as strategy priority or random selection (if the priority is the same). Finally, the computer system controls the target air-conditioning equipment based on the determined target control strategy. For example, if strategy B includes control commands such as setting the temperature to 23°C and the wind speed to medium wind speed, the computer system will control the target air-conditioning equipment according to these commands, thereby achieving the expected effects such as stable control of the temperature of the target air-conditioning equipment.

[0044] As an implementation method, the control command set stores multiple air-conditioning control strategies, each of which has a corresponding control strategy implicit array constructed based on the central side network, and the control strategy implicit array of the target control strategy obtained from the control command set matches the air-conditioning equipment implicit array of the target air-conditioning equipment. Based on this, the central side network and the edge side network are obtained based on collaborative tuning, and the cost parameters determined by the cost metric function used in the collaborative tuning link include: the air-conditioning equipment implicit array of the sample air-conditioning equipment obtained based on the edge side network during the tuning process, the control strategy implicit array of the control strategy learning sample obtained based on the central side network during the tuning process, and the prior mark determined based on the maintenance of the control strategy learning sample by the sample air-conditioning equipment.

[0045] The operation of the computer system under this implementation involves many aspects such as a control command set, a center-side network, an edge-side network, and the coordinated adjustment between them.

[0046] First, the control command set stores multiple air conditioning control strategies. Each air conditioning control strategy has a corresponding control strategy implicit array built based on the central side network. For example, suppose there are three air conditioning control strategies in the control command set, namely strategy P, strategy Q and strategy R.

[0047] For strategy P, the central side network constructs its control strategy implicit array based on the various parameters of strategy P. Assume that strategy P mainly involves parameters such as temperature adjustment range of 22-26℃, wind speed adjustment of low, medium and high gears, and compressor operating frequency between 30-50Hz. The central side network constructs the control strategy implicit array of these parameters through specific encoding and mapping methods, for example, it may be [10, 15, 20] (the numerical values ​​here are only assumptions for the convenience of example, and the actual construction process will involve complex network processing). Similarly, for strategies Q and R, different control strategy implicit arrays will be constructed based on their respective parameters. For example, the control strategy implicit array corresponding to strategy Q is [12, 18, 22], and the corresponding control strategy implicit array for strategy R is [8,14, 18].

[0048] When obtaining the target control strategy from the control command set, the control strategy implicit array of the target control strategy must match the air conditioning implicit array of the target air conditioning device. The air conditioning implicit array of the target air conditioning device is a comprehensive representation array (as constructed in the previous step S400) obtained by the computer system after comprehensively considering the objective parameter elements of the air conditioning, the control record data, and the data related to the air conditioning device. Assume that the air conditioning implicit array of the target air conditioning device is [15, 20,25]. The computer system will use a matching mechanism to determine the target control strategy. This matching mechanism may be based on a distance metric or a similarity calculation method. For example, the cosine similarity is used to calculate the similarity between the air conditioning implicit array of the target air conditioning device and the implicit arrays of each control strategy. Calculate the cosine similarity with the control strategy implicit array [10, 15, 20] of strategy P:

[0049] ;

[0050] Similarly, the cosine similarity with strategy Q and strategy R is calculated. Assuming that after calculation, the cosine similarity of strategy Q is the highest, then strategy Q will be determined as the target control strategy.

[0051] On this basis, the center-side network and the edge-side network are obtained based on collaborative tuning. In the collaborative tuning link, the cost metric function is used to determine the cost parameters. One of the cost parameters is an implicit array of air conditioners of sample air conditioners obtained based on the edge-side network during the tuning process. For example, suppose there are 5 sample air conditioners, namely sample air conditioner A, sample air conditioner B, sample air conditioner C, sample air conditioner D, and sample air conditioner E.

[0052] For the sample air conditioner A, the edge network constructs the implicit array of the air conditioner A based on its objective air conditioner parameter elements (such as power of 2 horsepower, cooling capacity of 5000W, etc.), control record data (such as temperature and wind speed settings at different times), and related data of the air conditioner (such as heat interaction with adjacent air conditioners, etc.), assuming it is [8, 12, 16]. Similarly, for other sample air conditioners BE, their own implicit arrays will also be constructed.

[0053] Another cost parameter is an implicit array of control strategies based on the control strategy learning samples obtained by the central side network during the tuning process. Assume that there are three control strategy learning samples during the tuning process, namely, learning sample strategy M, learning sample strategy N, and learning sample strategy O.

[0054] For the learning sample strategy M, the central side network constructs its control strategy implicit array, for example, [6, 10, 14]. For the learning sample strategy N and the learning sample strategy O, the corresponding control strategy implicit array is also constructed.

[0055] Another cost parameter is a priori mark determined based on the sample air conditioner's maintenance of the control strategy learning sample. For example, for sample air conditioner A, when applying the learning sample strategy M, if sample air conditioner A maintains the strategy for more than the set time threshold (assuming the time threshold is 2 hours), it means that the customer accepts the control strategy. At this time, a priori mark will be determined for this combination (sample air conditioner A and learning sample strategy M). This priori mark may be a specific value or code, such as a mark of 1, indicating acceptance. If sample air conditioner A changes the strategy before exceeding the time threshold when applying the learning sample strategy M, the priori mark may be 0, indicating rejection.

[0056] In the collaborative tuning process, the computer system considers these cost parameters through a cost metric function. The cost metric function may be a complex formula, such as:

[0057] ;

[0058] Among them, D1 represents the difference measure between the implicit array of the air-conditioning equipment of the sample air-conditioning equipment obtained based on the edge network and the ideal state (this ideal state can be the optimal state array obtained based on a large amount of data statistics), and α is its corresponding weight coefficient; D2 represents the difference measure between the implicit array of the control strategy learning sample obtained based on the central network and the implicit array of the optimal control strategy, and β is its corresponding weight coefficient; D3 represents a measure calculated based on the prior label (for example, the more combinations with the prior label of 0, the larger the value of D3, indicating that the overall acceptance is low), and γ is its corresponding weight coefficient. By continuously adjusting the parameters of the central network and the edge network, the value of the cost metric function is minimized, thereby realizing the coordinated adjustment of the central network and the edge network. This coordinated adjustment can make the matching between the implicit array of the control strategy constructed by the central network and the implicit array of the air-conditioning equipment constructed by the edge network more accurate, thereby improving the accuracy of obtaining the appropriate target control strategy from the control command set, and finally achieving more accurate and effective control of the air-conditioning equipment, such as meeting the user's comfort requirements while improving the energy efficiency and operation stability of the air-conditioning.

[0059] As an implementation method, the method may further include:

[0060] Step S600: obtaining a control command sequence of a target air conditioning control strategy, the control command sequence including: distribution position data and control commands of different control dimensions;

[0061] Step S700: generating dimensional implicit arrays of different control dimensions according to each control command and distribution position data, and integrating the dimensional implicit arrays of different control dimensions to obtain a multi-dimensional integrated implicit array;

[0062] Step S800: The Transformer component of the central side network obtains a multi-dimensional transformation implicit array corresponding to the multi-dimensional integration implicit array;

[0063] Step S900: performing information mapping on the multi-dimensional transformation implicit array based on the feedforward network component of the central side network to obtain a control strategy implicit array of the target air conditioning control strategy; the target air conditioning control strategy is one of the multiple air conditioning control strategies.

[0064] In step S600, a control command sequence of a target air conditioning control strategy is obtained. The control command sequence includes distribution position data and control commands of different control dimensions. For example, consider a target air conditioning control strategy, the purpose of which is to achieve comfortable temperature and air flow regulation in a specific environment.

[0065] For the distribution location data in the control command sequence, it may represent the distribution information of the control command in time or space. Assuming it is an air conditioning control scenario with time series characteristics, the distribution location data can be the identifier of different time periods. For example, the 24 hours of a day are divided into several time intervals, 0-4 o'clock is the low activity period at night, 5-8 o'clock is the morning preheating period, 9-17 o'clock is the office period, and 18-23 o'clock is the evening period. These different time period identifiers are part of the distribution location data, which provides a basis for subsequent targeted control according to different time periods.

[0066] The control commands of different control dimensions cover multiple aspects of air conditioning control. Take temperature and wind speed as examples. In the temperature control dimension, the control command may be to set the target temperature value for different time periods.

[0067] In step S700, dimensional implicit arrays of different control dimensions are generated according to the control commands and distribution position data obtained in step S600. Still taking the temperature and wind speed control dimensions as an example.

[0068] For the temperature control dimension, assume that the computer system constructs a dimensional implicit array based on the distribution location data (different time periods) and the temperature control commands (target temperature values ​​for different time periods). The computer system first encodes these data and converts the time period information and temperature values ​​into a form suitable for calculating and representing potential relationships. For example, for the 22°C temperature setting during the morning preheating period (5-8 o'clock), after encoding, it may be represented as [0.1, 0.2] (this is just a simple example, and the actual encoding may involve more complex algorithms), where 0.1 may represent a certain encoding feature of the time period, and 0.2 represents the encoding feature of the 22°C temperature. With the encoding of the temperature control commands for each time period, the dimensional implicit array of the temperature control dimension is finally formed, such as [[0.1, 0.2], [0.3, 0.4], [0.5, 0.4]] (corresponding to the encoding representations of the morning preheating period, office period, etc.).

[0069] For the wind speed control dimension, the dimension implicit array is also constructed based on the distribution location data and the wind speed control command. For example, the low wind speed is encoded as 0.1 and the medium wind speed is encoded as 0.3. For the low wind speed setting during the morning preheating period, it may be encoded as [0.1, 0.1]. After encoding the wind speed control commands for each period, the dimension implicit array of the wind speed control dimension is obtained, such as [[0.1, 0.1], [0.3, 0.3], [0.2, 0.1]].

[0070] Then, the computer system integrates the dimensional implicit arrays of different control dimensions (such as temperature and wind speed) to obtain a multi-dimensional integrated implicit array. The integration method can be simple array concatenation or combination according to specific weights. For example, the dimensional implicit array of the temperature control dimension and the dimensional implicit array of the wind speed control dimension are concatenated in sequence to obtain a multi-dimensional integrated implicit array of [[0.1, 0.2, 0.1, 0.1], [0.3, 0.4, 0.3, 0.3], [0.5, 0.4,0.2, 0.1]]. This multi-dimensional integrated implicit array integrates the control information of temperature and wind speed in different time periods, and can more comprehensively reflect the characteristics of the target air conditioning control strategy.

[0071] In step S800, the Transformer component based on the central side network processes the multi-dimensional integrated implicit array to obtain a multi-dimensional transformed implicit array. The Transformer component has a unique structure and function, and can extract and transform features of the input multi-dimensional integrated implicit array.

[0072] Assume that the multi-dimensional integrated implicit array is [[0.1, 0.2, 0.1, 0.1], [0.3, 0.4, 0.3, 0.3],[0.5, 0.4, 0.2, 0.1]]. The Transformer component first encodes the position of each element in the array. Considering the position information of the element in the array, this helps capture the sequential relationship between different time periods and different control dimensions. Then, feature extraction is performed through a multi-layer self-attention mechanism.

[0073] In the self-attention mechanism, each element is calculated in association with other elements in the array to determine its importance weight. For example, for the first element 0.1 in the array (in the first position of the first subarray), it will be calculated with other elements such as 0.2, 0.1, 0.1 (other elements in the same subarray) and the corresponding position elements in other subarrays to obtain a new representation that integrates its relationship information with other elements. After being processed by multiple layers of self-attention mechanisms, the element values ​​in the array will change, and a multi-dimensional transformation implicit array will be obtained. Assume that after being processed by the Transformer component, the multi-dimensional transformation implicit array becomes [[0.15, 0.22, 0.12, 0.13], [0.33,0.45, 0.35, 0.32], [0.52, 0.43, 0.23, 0.11]]. This multi-dimensional transformation implicit array mines out the potential feature relationships in the multi-dimensional integration implicit array, making the association between control information in different dimensions and time periods clearer, and providing more representative data for subsequent operations.

[0074] In step S900, the feedforward network component of the central side network is used to perform information mapping on the multi-dimensional transformation implicit array, so as to obtain the control strategy implicit array of the target air-conditioning control strategy.

[0075] The feedforward network component consists of multiple neuron layers, each with a corresponding weight and activation function. Take the multi-dimensional transformation implicit array [[0.15, 0.22, 0.12, 0.13], [0.33, 0.45, 0.35, 0.32],[0.52, 0.43, 0.23, 0.11]] as an example.

[0076] First, the input layer of the feedforward network component receives the multi-dimensional transformation implicit array. Then, the data is passed between the neuron layers. Each neuron performs a weighted summation operation on the input data. For example, for the first sub-array of the input [0.15, 0.22, 0.12, 0.13], assuming that the weight vector of the first neuron is [0.2, 0.3, 0.1, 0.4], the weighted summation result of the neuron is (0.15×0.2 + 0.22×0.3 + 0.12×0.1 + 0.13×0.4=0.158). Then, the weighted summation result is nonlinearly transformed by an activation function (such as the ReLU function). If (0.158 >0), the result after the ReLU function is still (0.158). After calculations in multiple neuron layers, the control strategy implicit array of the target air conditioning control strategy is finally obtained. Assume that the control strategy implicit array obtained after processing by the feedforward network component is [0.2, 0.3, 0.25, 0.35]. This control strategy implicit array is a comprehensive representation of the target air-conditioning control strategy after multi-dimensional integration, transformation and information mapping. It can be used to match other arrays (such as the air-conditioning equipment implicit array of the target air-conditioning equipment) to achieve the reasonable selection and application of the air-conditioning control strategy, thereby achieving the goal of temperature stability control.

[0077] As an implementation manner, in the above step S600, obtaining control commands of different control dimensions of the target air conditioning control strategy may include:

[0078] Step S610: Acquire the temperature control strategy and wind speed control strategy included in the target air conditioning control strategy, and obtain a control command in the temperature dimension and a control command in the wind speed dimension;

[0079] Step S620: Based on the control command of the temperature dimension and the control command of the wind speed dimension, the control commands of different control dimensions of the target air conditioning control strategy are determined.

[0080] In step S610, the computer system specifically obtains the temperature control strategy and wind speed control strategy parts of the target air conditioning control strategy, thereby determining the control command of the temperature dimension and the control command of the wind speed dimension.

[0081] For the temperature control strategy, assume that the target air conditioning control strategy is set to adapt to the indoor environment requirements at different times of the day. In an office environment, when people first enter the office in the morning, the indoor temperature is low and needs to be quickly raised to a comfortable working temperature. At this time, the temperature control strategy may stipulate that during the period from 8 to 9 o'clock, the temperature is quickly raised from the initial low temperature (for example, 18°C) to 22°C. This is the control command of the temperature dimension during this period, which clarifies the target value of temperature control and the approximate time range.

[0082] As time goes by to noon (12:00-13:00), the indoor temperature may rise due to factors such as personnel activities and equipment heat dissipation. In order to maintain a comfortable working environment, the temperature control strategy stipulates that the temperature should be maintained at around 24°C. This is another control command for the temperature dimension at different time periods, which reflects the requirements for temperature adjustment based on different environmental factors.

[0083] The wind speed control strategy works in conjunction with the temperature control strategy. In the morning heating phase (8:00-9:00), in order to distribute the hot air quickly and evenly in the room, the wind speed control strategy may be set to high wind speed, which is the control command for the wind speed dimension during this period. High wind speed can speed up air circulation, allowing the hot air from the air conditioner to fill the entire room faster and improve heating efficiency.

[0084] At noon, during the temperature maintenance phase (12:00-13:00), although the temperature needs to be maintained at 24°C, the wind speed control strategy may be adjusted to a low wind speed to avoid discomfort caused by strong winds. This wind speed control command reflects the consideration of human comfort while meeting the temperature control requirements.

[0085] In step S620, after obtaining the control commands of the temperature dimension and the wind speed dimension, the control commands of different control dimensions of the target air conditioning control strategy are determined based on these commands. This step is not just a simple summary, but an integration and clarification of the relationship between the control commands of different dimensions.

[0086] Continuing with the air conditioning control in the above office environment as an example, the combination of the control commands of the temperature dimension and the wind speed dimension at different time periods forms a complete control logic. During the period of 8-9 in the morning, the temperature rises rapidly from 18°C ​​to 22°C and the wind speed is high. This combination of control commands fully describes how the air conditioning should operate during this period to meet the indoor environment requirements. The computer system determines this relationship between time period, temperature and wind speed as part of the control commands of different control dimensions of the target air conditioning control strategy.

[0087] Looking at the period from 12:00 to 13:00 noon, the temperature is maintained at 24℃ and the wind speed is low. This combination is also an important part of the control commands of different control dimensions of the target air conditioning control strategy. The computer system integrates the temperature and wind speed control commands in these different time periods to form a complete set of air conditioning control instructions for different environments and demand changes.

[0088] For example, assuming that there is an afternoon period (14:00-16:00) in a day, the indoor temperature distribution may be uneven due to factors such as changes in the direction of sunlight. At this time, the temperature control strategy requires maintaining 24°C, but some areas may need to fine-tune the temperature to achieve overall comfort. At the same time, the wind speed control strategy is adjusted to medium wind speed to better adjust the indoor temperature distribution while ensuring comfort. The computer system combines the temperature and wind speed control commands for this period to further enrich and improve the control commands of different control dimensions of the target air conditioning control strategy. In this way, the computer system constructs a control command system that includes multiple control dimensions such as temperature and wind speed that cooperate with each other in different time periods and under different environmental conditions.

[0089] As an implementation method, the method may further include:

[0090] Step S10: generating an air conditioning control strategy;

[0091] Step S20: constructing an implicit array of control strategies for generating air conditioning control strategies based on the central side network;

[0092] Step S30: performing a commonality measurement operation based on the control strategy implicit array of each generated air conditioning control strategy to obtain a commonality measurement result, wherein the commonality measurement result represents the degree of commonality between any two air conditioning control strategies in each generated air conditioning control strategy;

[0093] Step S40: performing redundancy removal operations on the generated air-conditioning control strategies based on the degree of commonality between any two of the generated air-conditioning control strategies; wherein, when constructing an implicit array of control strategies for each generated air-conditioning control strategy based on the central side network, the loading information of the central side network is the temperature control strategy and wind speed control strategy for each generated air-conditioning control strategy.

[0094] The computer system first executes step S10, that is, generating an air conditioning control strategy. The air conditioning control strategy is a set of rules and parameters for controlling the operation of the air conditioner, aiming to meet different needs, such as temperature adjustment, energy saving, comfort optimization, etc.

[0095] After generating the air conditioning control strategy, the computer system executes step S20 to construct control strategy implicit arrays for each generated air conditioning control strategy based on the central side network. The central side network is a network structure that is pre-trained or specially designed for processing air conditioning control related data.

[0096] Take the air conditioning control strategy in the office environment mentioned above as an example. For the air conditioning control strategy during daytime office hours, the central network will process the various parameters. The parameter of setting the temperature to 24°C may be encoded as a specific value. Assume that according to the encoding rules of the central network, 24°C is encoded as [0.2, 0.4] (this is just an example, the actual encoding may be more complicated). The medium and low wind speeds also have corresponding encodings. Assume that the medium wind speed is encoded as [0.3, 0.6] and the low wind speed is encoded as [0.1, 0.2]. In terms of seasonal factors, summer is encoded as [0.8, 0.1] and winter is encoded as [0.1, 0.8]. The operating mode cooling is encoded as [0.9, 0.1] and heating is encoded as [0.1, 0.9].

[0097] Combining these codes together forms an implicit array of control strategies for the air conditioning control strategy during daytime office hours, for example, it may be [[0.2, 0.4], [0.3, 0.6], [0.8, 0.1], [0.9, 0.1]]. Similarly, for the air conditioning control strategy at night, parameters such as temperature, wind speed, season, and operation mode are encoded in a similar way and combined into an implicit array of control strategies, assuming it is [[0.26, 0.52], [0.1, 0.2], [0.1, 0.8], [0.1, 0.9]].

[0098] This control strategy implicit array is an abstract representation of the air conditioning control strategy. It integrates various parameters in a form suitable for subsequent processing by the central side network, and can better explore the potential relationships and characteristics between different control strategies.

[0099] In step S30, after constructing the control strategy implicit array of each air conditioning control strategy, the computer system executes step S30 to perform a commonality measurement operation. The commonality measurement operation is intended to find out the similarity or commonality degree between different air conditioning control strategies.

[0100] For example, consider the previously generated implicit arrays of control strategies for the air conditioning control strategies during daytime office hours and nighttime, which are [[0.2, 0.4], [0.3, 0.6], [0.8, 0.1], [0.9, 0.1]] and [[0.26, 0.52], [0.1, 0.2], [0.1, 0.8], [0.1, 0.9]]. The computer system can use a variety of methods to measure commonality, such as cosine similarity calculation. The cosine similarity value obtained by calculation is the commonality measurement result between the two air conditioning control strategies. If there are other air conditioning control strategies, the computer system will calculate the commonality measurement results between them in the same way. This commonality measurement result can reflect the similarity between different control strategies in terms of parameter settings and operating modes. For example, a higher cosine similarity value indicates that the two control strategies have similar settings in terms of temperature, wind speed, seasonal mode, etc.

[0101] In step S40, after obtaining the commonality measurement results between the air conditioning control strategies, the computer system executes step S40 to perform a redundancy removal operation. The purpose of the redundancy removal operation is to reduce control strategies with high similarity, so as to simplify the control strategy set, improve control efficiency and save storage space.

[0102] Assume that in addition to the air conditioning control strategies for daytime office hours and nighttime office hours mentioned above, there is also an air conditioning control strategy for weekend office environments. After the implicit array of its control strategies is subjected to the commonality measurement operation with other strategies, it is found that the commonality measurement results (such as cosine similarity) with the air conditioning control strategy for daytime office hours are very high, for example, reaching 0.9 (this is just an assumed high similarity value). This means that the two strategies are very similar in terms of temperature setting, wind speed adjustment, season, and operation mode. Based on this degree of commonality, the computer system determines that the air conditioning control strategy for the weekend office environment is redundant with the strategy for daytime office hours. Therefore, the computer system removes the air conditioning control strategy for the weekend office environment from the generated air conditioning control strategy set.

[0103] In this process, when the implicit array of control strategies for each generated air conditioning control strategy is constructed based on the central side network, the loading information of the central side network is the temperature control strategy and wind speed control strategy for each generated air conditioning control strategy. This means that in the process of constructing the implicit array of control strategies, the two key control factors of temperature and wind speed play an important role in mining the commonalities and differences between strategies. For example, if two strategies are mostly the same in the settings of temperature and wind speed, then they will show a high similarity in the overall implicit array of control strategies, and are more likely to be judged as redundant strategies and removed.

[0104] As an implementation mode, the edge side network includes an embedded coding component, and the objective parameter characterization component includes a feedforward network component. Based on this, in step S200, obtaining an objective parameter implicit array of air conditioner objective parameter elements based on the objective parameter characterization component of the edge side network may include:

[0105] Step S210: embedding the objective parameters obtained by embedding the air conditioner objective parameter elements by the embedded coding component based on the edge side network into an implicit array as loading data of the objective parameter representation component;

[0106] Step S220: Based on the feedforward network component of the objective parameter characterization component, information mapping is performed on the objective parameter embedded implicit array to obtain the objective parameter implicit array of the air-conditioning objective parameter elements.

[0107] In step S210, the computer system first uses the embedded coding component of the edge network to perform embedded coding operations on the air conditioner objective parameter elements, thereby obtaining an objective parameter embedded implicit array. The air conditioner objective parameter elements include a variety of parameters that reflect the characteristics of the air conditioner itself.

[0108] For example, assume that the objective parameter elements of an air conditioner include the cooling capacity, energy efficiency ratio, and compressor power of the air conditioner. The cooling capacity is 3500W, the energy efficiency ratio is 3.2, and the compressor power is 1000W. The embedded coding component of the edge network will encode these parameters according to the pre-set coding rules.

[0109] For the cooling capacity of 3500W, assuming that the coding rule is to encode every 1000W as a unit and use a simple linear coding method (the actual situation may be more complicated), then 3500W may be encoded as [3.5]. For the energy efficiency ratio of 3.2, assuming that it is mapped to a specific coding space according to its value range (for example, between 2-5), the code is [0.6] (this is just a simple example). For the compressor power of 1000W, the code is [1] according to the coding method similar to the cooling capacity.

[0110] Combining these codes together, we get the objective parameter embedded implicit array, such as [3.5, 0.6, 1]. This objective parameter embedded implicit array is the result of embedding the air conditioner objective parameter elements based on the embedded coding component of the edge network, and the computer system will use this array as the loading data of the objective parameter representation component.

[0111] In step S220, after obtaining the objective parameter embedded in the implicit array, the computer system uses the feedforward network component of the objective parameter characterization component to perform information mapping operation on it. The feedforward network component is a neural network structure composed of multiple neurons, which can perform complex transformations on input data and mine potential relationships in the data.

[0112] Take the objective parameter embedding implicit array [3.5, 0.6, 1] obtained just now as an example. Each neuron in the feedforward network component has its own weight and bias. Assume that the feedforward network component has three neurons and its weight matrix is , the bias vector is .

[0113] For the input objective parameter embedded implicit array [3.5, 0.6, 1], a weighted sum operation is first performed. The input of the first neuron is calculated as ; The input to the second neuron is calculated as ; The input to the third neuron is calculated as . Then, the weighted summation result is transformed nonlinearly through an activation function. Assuming the ReLU (Rectified Linear Unit) activation function is used, since 1.08 > 0, 0.99 > 0, 1.97 > 0, the result after the ReLU function is still 1.08, 0.99, 1.97. These results form a new array [1.08, 0.99, 1.97], which is the objective parameter implicit array of the air-conditioning objective parameter elements obtained after the feedforward network component of the objective parameter representation component performs information mapping on the objective parameter embedded implicit array.

[0114] This implicit array of objective parameters represents the objective parameter elements of the air conditioner in a new form after mining potential relationships. Compared with the original objective parameter elements of the air conditioner, it can better reflect the complex interrelationships between these parameters and the comprehensive impact on the overall performance and control of the air conditioner. Through this information mapping, the comprehensive impact of cooling capacity, energy efficiency ratio and compressor power on the actual operating effect of the air conditioner (such as temperature adjustment speed, energy consumption, etc.) under a certain specific combination can be mined.

[0115] As an implementation manner, the edge side network includes an embedded coding component. In step S200, obtaining a control record implicit array of the first air conditioner control record data based on the control record representation component of the edge side network may include:

[0116] Step S230: embedding multiple air-conditioning control records of the first air-conditioning control record data obtained by the embedded coding component based on the edge side network into an implicit array as loading data of the control record representation component of the edge side network, wherein the first air-conditioning control record data includes multiple multi-dimensional control command data arranged in time sequence according to the control moment;

[0117] Step S240: The control record representation component based on the edge network processes the multiple air-conditioning control record embedded implicit arrays to obtain the control record implicit array of the first air-conditioning control record data;

[0118] Among them, the control record representation component of the edge side network obtains an implicit array of control records of the first air-conditioning control record data based on multiple multi-dimensional control command data with a control duration greater than a preset duration; or, the control record representation component of the edge side network obtains multiple implicit arrays of air-conditioning control record mappings corresponding to multiple air-conditioning control record embedding implicit arrays based on the Transformer component of the control record representation component, and processes the multiple implicit arrays of air-conditioning control record mappings based on the downsampling module of the control record representation component to obtain the implicit array of control records of the first air-conditioning control record data.

[0119] In step S230, the computer system processes the first air conditioning control record data through the embedded coding component of the edge network to obtain multiple air conditioning control record embedded implicit arrays. The first air conditioning control record data includes multiple multi-dimensional control command data arranged in time sequence according to the control time.

[0120] Taking the air conditioning control record in an office environment as an example, suppose the air conditioning control record within a day is as follows: at 8 o'clock in the morning, the temperature is set to 22°C, the wind speed is set to low wind speed, and the operation mode is heating; at 10 o'clock in the morning, the temperature is adjusted to 20°C, the wind speed is adjusted to high wind speed, and the operation mode remains unchanged; at 12 o'clock in the afternoon, the temperature is adjusted to 23°C again, the wind speed becomes medium wind speed, and the operation mode is still heating.

[0121] For the control record at 8 a.m., this is a multi-dimensional control command data, including three dimensions: temperature, wind speed, and operation mode. Assuming that the encoding rule of the embedded coding component of the edge network for temperature is to divide the actual temperature value by 10 and round it up (just for example), then 22°C is encoded as [2]; for the encoding of wind speed, set the low wind speed to [1], the high wind speed to [3], and the medium wind speed to [2]; for the operation mode, the heating is encoded as [1] and the cooling is encoded as [0]. Therefore, the air conditioning control record at 8 a.m. after the embedded coding component is passed through, the embedded implicit array of the air conditioning control record is [2, 1, 1].

[0122] In the same way, the air conditioning control record embedded in the implicit array obtained after encoding the control record at 10 a.m. is [2, 3, 1]; the air conditioning control record embedded in the implicit array obtained after encoding the control record at 12 noon is [2,2, 1]. These multiple air conditioning control record embedded implicit arrays obtained by the embedded coding component of the edge network, such as [2, 1, 1], [2, 3, 1], [2, 2, 1], etc., are used by the computer system as the loading data of the control record representation component of the edge network.

[0123] In step S240, the computer system uses the control record representation component of the edge network to process the loaded multiple air conditioner control record embedded implicit arrays to obtain the control record implicit array of the first air conditioner control record data. There are two possible processing methods.

[0124] 1. Based on multiple multi-dimensional control command data whose control duration is greater than a preset duration.

[0125] Assume that the preset duration is 2 hours. In the air conditioning control records of the above office environment, the control record duration from 8 am to 12 pm is greater than 2 hours. The computer system will select the air conditioning control records during this period and embed them into an implicit array, namely [2, 1, 1], [2, 3, 1], [2, 2, 1] for processing.

[0126] A simple way to deal with this is to find the average. Find the average of each dimension separately: temperature dimension: (2+2+2)÷3=2; wind speed dimension: (1+3+2)÷3=2; operation mode dimension: (1+1+1)÷3=1. So the implicit array of control records is [2, 2, 1]. This implicit array of control records reflects an average state of air conditioning control over a long control period. For example, during this period, the average temperature setting is close to 20°C (according to the coding rules), the average wind speed setting is close to medium wind speed, and the operation mode is always heating.

[0127] 2. Based on Transformer components and downsampling modules.

[0128] Firstly, the computer system processes multiple air-conditioning control record embedded implicit arrays based on the Transformer component of the control record representation component to obtain multiple air-conditioning control record mapping implicit arrays.

[0129] The Transformer component has features such as self-attention mechanism, which can capture the relationship between different elements in the input array. Take the previously obtained air conditioning control record embedded implicit arrays [2, 1, 1], [2, 3, 1], [2, 2, 1] as an example. The Transformer component will process these arrays, considering the position of each element in the sequence and its relationship with other elements. Assume that after being processed by the Transformer component, [2, 1, 1] is mapped to [2.5, 1.2, 1.1], [2, 3, 1] is mapped to [2.2, 3.2, 1.2], and [2, 2, 1] is mapped to [2.3, 2.1, 1.1], and multiple air conditioning control record mapping implicit arrays [2.5, 1.2, 1.1], [2.2, 3.2, 1.2], [2.3, 2.1, 1.1] are obtained. The computer system then processes these mapped implicit arrays using the downsampling module of the control record representation component. The downsampling module can reduce the amount of data while retaining important information. For example, the downsampling module can take a value for every other element according to a certain rule. For [2.5, 1.2, 1.1], the downsampling results in [2.5, 1.1]; for [2.2, 3.2, 1.2], the downsampling results in [2.2,1.2]; for [2.3, 2.1, 1.1], the downsampling results in [2.3, 1.1]. These downsampled arrays are then merged or further processed to obtain the final control record implicit array. One possible way to merge is to average the elements at the corresponding positions again, temperature dimension: (2.5 + 2.2 + 2.3) ÷ 3 = 2.33; wind speed dimension: (1.1 + 1.2 + 1.1) ÷ 3 = 1.13; operation mode dimension: (1 + 1 + 1) ÷ 3 = 1. So the implicit array of control records is [2.33, 1.13, 1].

[0130] As an implementation method, the involved analysis factor data set includes x involved analysis factor data corresponding to x involved air conditioning devices, x≥1; the edge side network includes an embedded coding component, and the weight focusing component includes x device weight focusing modules, a Transformer component, and a downsampling module. Based on this, step S300, based on the weight focusing component of the edge side network, obtains the fused implicit array corresponding to the air conditioning objective parameter element and the involved analysis factor data set, which may include:

[0131] Step S310: embedding the objective parameter implicit array obtained by embedding the air conditioner objective parameter elements by the embedded coding component based on the edge side network, and embedding the x involved analysis element data by the embedded coding component based on the edge side network to obtain the x involved embedded coding implicit arrays as the loading data of the weight focusing component;

[0132] Step S320: x device weight focusing modules based on the weight focusing component process the objective parameter embedding implicit array and x involved embedding coding implicit arrays to obtain x involved weight implicit arrays;

[0133] Step S330: The Transformer component based on the weight focusing component obtains the implicit arrays of involvement mappings corresponding to the x implicit arrays of involvement weights;

[0134] Step S340: The downsampling module based on the weighted focusing component processes the implicit array involved in the mapping to obtain a fused implicit array.

[0135] In step S310, first, for the objective parameter elements of the air conditioner, it is assumed that the objective parameter elements of the air conditioner include cooling capacity, energy efficiency ratio and compressor power. The cooling capacity is 5000W, the energy efficiency ratio is 3.5, and the compressor power is 1500W. The embedded coding component of the edge side network encodes these parameters according to specific rules.

[0136] For example, for a cooling capacity of 5000W, if 1000W is used as a coding unit and a linear coding method is used (the actual situation may be more complicated), the code is [5]; the energy efficiency ratio is 3.5, and according to its value range (assuming it is between 2-5), it is mapped to a specific coding space and the code is [0.7]; the compressor power is 1500W, and it is coded [1.5] in a similar coding method to the cooling capacity. These codes are combined to obtain the objective parameters embedded in the implicit array, for example, [5, 0.7, 1.5].

[0137] Assume that the data set of the involved analysis elements includes 2 involved air-conditioning devices (i.e., x=2). For involved air-conditioning device 1, its involved analysis element data includes the operating time (assuming it is 8 hours), the current set temperature (24°C), and the operating mode (cooling). Similarly, the embedded coding component is used for encoding. The operating time is 8 hours. If every 2 hours is a coding unit, it is encoded as [4]; the temperature of 24°C is encoded as [2] after dividing by 10 and rounding (only for example encoding rules); the cooling mode is encoded as [0], and the involved embedded coding implicit array of involved air-conditioning device 1 is [4, 2, 0]. For involved air-conditioning device 2, its involved analysis element data is the operating time of 6 hours, the set temperature of 23°C, and the operating mode of heating. According to the above encoding rules, the operating time is encoded as [3], 23°C is encoded as [2], and the heating mode is encoded as [1], and the involved embedded coding implicit array of involved air-conditioning device 2 is [3, 2, 1]. The computer system embeds the obtained objective parameters into an implicit array [5, 0.7, 1.5], as well as two implicit arrays [4, 2, 0] and [3, 2, 1] involving embedded coding as loading data of the weight focusing component.

[0138] In step S320, since x=2, two device weight focusing modules are involved in the processing. The processing process is described here in a relatively abstract way (the specific calculation method in steps S321-S323 is not involved for the time being). For the first device weight focusing module, it receives the objective parameter embedded implicit array [5, 0.7, 1.5] and the involved embedded coding implicit array [4, 2, 0] involving air-conditioning equipment 1. This module has specific algorithms and weight parameters inside to perform calculations on these two arrays. For example, it may be in a weighted summation manner, assuming that the weights of the objective parameter embedded implicit array in the module are [0.3, 0.3, 0.4], and the weights of the involved embedded coding implicit array are [0.2, 0.5, 0.3].

[0139] The calculation process is as follows:

[0140] For the first element, calculate 5×0.3 + 4×0.2=1.5 + 0.8=2.3;

[0141] For the second element, calculate 0.7×0.3 + 2×0.5=0.21 + 1=1.21;

[0142] For the third element, calculate 1.5×0.4 + 0×0.3=0.6 + 0=0.6.

[0143] The first implicit array of involved weights is obtained as [2.3, 1.21, 0.6].

[0144] Similarly, for the second device weight focusing module, the objective parameter embedding implicit array [5, 0.7, 1.5] and the involved embedding encoding implicit array [3, 2, 1] involving air-conditioning device 2 are received. The calculation is similar (assuming different weight parameters, for example, the weights for the objective parameter embedding implicit array are [0.4, 0.2, 0.4], and the weights for the involved embedding encoding implicit array are [0.3, 0.4, 0.3]).

[0145] For the first element, calculate 5×0.4 + 3×0.3=2 + 0.9=2.9;

[0146] For the second element, calculate 0.7×0.2 + 2×0.4=0.14 + 0.8=0.94;

[0147] For the third element, we calculate 1.5×0.4 + 1×0.3=0.6 + 0.3=0.9.

[0148] The second implicit array of involved weights is obtained as [2.9, 0.94, 0.9].

[0149] In step S330, the computer system processes the two implicit arrays of involved weights [2.3, 1.21, 0.6] and [2.9, 0.94, 0.9] obtained in the previous step through the Transformer component of the weight focusing component. The Transformer component has characteristics such as self-attention mechanism, which can capture the relationship between elements in the array and perform feature transformation on them.

[0150] Taking the implicit array of weights [2.3, 1.21, 0.6] as an example, the Transformer component will consider the relationship between each element and other elements. Assume that after being processed by the Transformer component, the relationship between the elements changes, and the values ​​also change accordingly, resulting in an implicit array of weights, such as [2.5, 1.3, 0.5] (here is just a simple illustration of the actual result after processing).

[0151] Similarly, for the implicit array of involvement weights [2.9, 0.94, 0.9], after being processed by the Transformer component, the corresponding implicit array of involvement mapping is obtained, which is assumed to be [3.1, 0.9, 0.8].

[0152] In step S340, the computer system processes the implicit array of the involvement mapping obtained above using the downsampling module of the weight focusing component. The function of the downsampling module is to reduce the amount of data while retaining important information. For example, the downsampling module may adopt a method of taking a value for every other element.

[0153] For the implicit array [2.5, 1.3, 0.5] involving the mapping, the downsampling results are [2.5, 0.5]; for the implicit array [3.1, 0.9, 0.8] involving the mapping, the downsampling results are [3.1, 0.8].

[0154] Then the downsampled results are combined or further processed to obtain a fused implicit array. One possible combination method is to add the elements at the corresponding positions to obtain a fused implicit array of [5.6, 1.3] (one decimal place is retained after the addition). This fused implicit array combines the information of the objective parameter elements of the air conditioner and the involved analysis element data after a series of processing. It reflects a comprehensive representation obtained after considering the air conditioner's own parameters and various factors related to the air conditioning equipment, and provides an important data basis for subsequent operations, such as integration with other implicit arrays to formulate air conditioning control strategies.

[0155] As an implementation mode, step S320, based on the x device weight focusing modules of the weight focusing component, processes the objective parameter embedding implicit array and the x involved embedding coding implicit arrays to obtain x involved weight implicit arrays, which may include:

[0156] Step S321: obtaining a Kronecker product array between an objective parameter embedding implicit array and a p-th involved embedding coding implicit array among x involved embedding coding implicit arrays, 1≤p≤x;

[0157] Step S322: concatenate the objective parameter embedding implicit array, the pth involved embedding coding implicit array and the Kronecker product array, and load the concatenated array into the nonlinear transformation unit of the pth device weight focusing module among the x device weight focusing modules to obtain the execution result of the nonlinear transformation unit of the pth device weight focusing module;

[0158] Step S323: Determine the execution result of the nonlinear transformation unit of the pth device weight focusing module as the loading data of the fully connected unit of the pth device weight focusing module, and obtain the pth implicit array of involved weights based on the fully connected unit of the pth device weight focusing module.

[0159] In step S321, it is assumed that in the previous step, the objective parameter embedding implicit array and the involved embedded coding implicit array have been obtained. For example, the objective parameter embedding implicit array is , here we assume (These values ​​are examples obtained by coding the air conditioning objective parameters mentioned above).

[0160] Assume that the involved analysis factor dataset includes 3 involved air conditioners (i.e. x=3). Now consider the p=1th involved embedded coding implicit array, assuming it is , here we assume .

[0161] The Kronecker product is a special matrix multiplication operation. For vectors, it can also be calculated according to similar rules. Calculate the Kronecker product array between A and B1. First, calculate the product of a1 and B1, and get Then calculate the product of a2 and B1, and we get Finally, we calculate the product of a3 and B1 and get . Combining these results, we get the Kronecker product array This Kronecker product array contains a special product relationship between the objective parameter embedding implicit array and the first involved embedding coding implicit array. This relationship can capture more complex interaction information between the two array elements and provide a richer data basis for subsequent processing.

[0162] In step S322, the computer system then embeds the objective parameters into the implicit array A=[2, 0.5, 1], the first implicit array involved in embedding coding and Kronecker product array The array D1 after splicing is . Then the concatenated array D1 is loaded into the nonlinear transformation unit of the first device weight focusing module (because p=1). Assume that this nonlinear transformation unit is a neural network-based structure, for example, it contains multiple neurons, each with its own weight and activation function. Assume that the first layer of this nonlinear transformation unit has 3 neurons, and its weight matrix is , the bias vector is For the input array D1, calculate the input value of each neuron. Taking the first neuron as an example, its input value z1 is (where d 1i is the i-th element in D1). After similar calculations, the input values ​​z1, z2, and z3 of the three neurons are obtained, and then nonlinear transformation is performed through the activation function. Assuming that the ReLU (Rectified Linear Unit) activation function is used, if z1>0, the output value y1=z1 after ReLU; if z1≤0, then y1=0. Similarly, z2 and z3 are processed to obtain y2 and y3. The result [y1, y2, y3] is the execution result of the nonlinear transformation unit of the first device weight focusing module. This execution result reflects a new feature representation after nonlinear transformation, which comprehensively considers objective parameters, parameters involving air-conditioning equipment, and the special product relationship between them.

[0163] In step S323, the computer system determines the execution result [y1, y2, y3] of the nonlinear transformation unit of the first device weight focusing module as the loading data of the fully connected unit of the device weight focusing module.

[0164] Assume that the fully connected unit of the first device weight focusing module is also a neural network structure, and its weight matrix is , the bias vector is .

[0165] Calculate the output of the fully connected unit. Taking the first element as an example, calculate .

[0166] Similarly, the second and third elements o2 and o3 are calculated to obtain the result [o1, o2, o3]. This result is the first implicit array of weight involvement. For example, suppose that after calculation, [o1, o2, o3] = [1.2, 0.8, 0.6] is obtained. This is the first implicit array of weight involvement obtained after considering the objective parameters, the parameters of the involved air-conditioning equipment, and the complex relationship between them (through operations such as Kronecker product, nonlinear transformation and fully connected units). If x>0, for other p (2≤p≤x), the computer system will follow the same steps and use the corresponding implicit array of involved embedding code to calculate and obtain different implicit arrays of weight involvement. These implicit arrays of weight involvement can comprehensively consider the relationship with objective parameters from the perspective of different involved air-conditioning equipment, provide an important data basis for subsequent further processing (such as obtaining fused implicit arrays through Transformer components and downsampling modules, etc.), and help to formulate air-conditioning control strategies more comprehensively and accurately.

[0167] As an implementation mode, the edge side network includes an embedded coding component, which is used to embed the element data to be loaded, and the element data to be loaded includes at least one of the air conditioner objective parameter element, the first air conditioner control record data, and the involved analysis element data set;

[0168] If the element data to be loaded is binary coded data, embedding coding is performed on the element data to be loaded based on the embedded coding component;

[0169] If the element data to be loaded is multiple binary coded data, the element data to be loaded is embedded coded based on the embedded coding component, and the coded array obtained after the embedded coding is downsampled, and the dimension of the coded array obtained after the downsampling operation meets the preset requirements.

[0170] The embedded coding component in the edge network performs a specific coding operation on the element data to be loaded. The element data to be loaded includes the air conditioner objective parameter element, the first air conditioner control record data and at least one part of the analysis element data set involved.

[0171] First, for the objective parameter elements of the air conditioner, it is assumed that the objective parameter elements of the air conditioner include the power state of the air conditioner (on or off, which can be represented as binary, such as 1 for on and 0 for off), cooling mode (yes / no, which can also be represented as binary, 1 for yes and 0 for no), etc. If these parameters exist in the form of binary coded data, for example, the power state is 1 and the cooling mode is 1, the embedded coding component directly embeds and encodes them. This embedded coding may be to map these binary data to a higher-dimensional space to explore deeper relationships between data. Assume that the simple mapping rule is to multiply each binary bit by a specific coefficient and then add them together. For the power state 1, multiply by the coefficient 2 to get 2; for the cooling mode 1, multiply by the coefficient 3 to get 3, and the encoded result is [2, 3] (this is just a simple example, and the actual encoding method will be more complicated).

[0172] When the element data to be loaded is multiple binary coded data, take the first air conditioning control record data as an example. Assume that the first air conditioning control record data contains air conditioning control information at multiple time points, and the control information at each time point contains multiple binary coded parameters, such as wind speed (low, medium, high, which can be coded as 00, 01, 10), temperature adjustment direction (increase, decrease, unchanged, which can be coded as 00, 01, 10), etc. This constitutes multiple binary coded data.

[0173] The computer system first embeds and encodes these data based on the embedded coding component. Assume that the wind speed 01 is encoded as [1, 2], and the temperature adjustment direction 01 is encoded as [3, 4]. At a certain point in time, the combination of these two control information may result in a coding array, such as [1, 2, 3, 4].

[0174] Since this is multiple binary coded data, the dimension of the encoded array obtained after encoding may be high and does not meet the preset requirements for subsequent processing, so downsampling is required. There are many ways to downsample, such as taking every other element. For the encoded array [1, 2, 3, 4], after downsampling, [1, 3] is obtained, so that the dimension of the encoded array obtained after the downsampling operation meets the preset requirements.

[0175] Let's look at the involved analysis element data set. Assume that the involved analysis element data set contains information such as the operating status of other air-conditioning equipment associated with the target air-conditioning equipment. For example, the auxiliary heating function (on / off, binary coding) and humidity adjustment function (on / off, binary coding) of a certain involved air-conditioning equipment. If it is simple binary coded data, such as 1 and 0, the embedded coding component directly performs embedded coding, assuming that [5, 1] ​​is obtained (encoding example). If these data exist in the form of multiple binary coded data, for example, plus a special mode of the device (yes / no, binary coding), the three binary codes are combined for embedded coding to obtain a coded array, and then down-sampling is performed to make its dimension meet the preset requirements.

[0176] In this way, the embedded encoding component encodes the loaded factor data, whether it is simple binary encoded data or multiple binary encoded data, and converts it into a form suitable for subsequent processing (such as further calculation of implicit arrays and other operations in the edge network), ensuring that the data can be effectively processed in the entire process of the air conditioning temperature stabilization control method and meet the requirements of each processing link for data format and dimension.

[0177] As an implementation manner, the involvement analysis factor data set includes x involvement analysis factor data corresponding to x involved air-conditioning devices. In step S100, obtaining the involvement analysis factor data set of the air-conditioning device involved in the target air-conditioning device may include:

[0178] Step S110: obtaining second air-conditioning control record data of any involved air-conditioning device among the x involved air-conditioning devices, and objective parameter data of involvement between any involved air-conditioning device and the target air-conditioning device, wherein the objective parameter data of involvement includes involvement category data and involvement frequency data;

[0179] Step S120: Determine the second air-conditioning control record data and the involved objective parameter data as involved analysis element data corresponding to any involved air-conditioning equipment.

[0180] In step S110, in an environment including multiple air-conditioning devices, such as a large office space or a multi-room building, there is a target air-conditioning device and associated involved air-conditioning devices (assuming there are x in total). The computer system performs a data acquisition operation on any of the involved air-conditioning devices.

[0181] Taking the air conditioning system in an office as an example, suppose there is a target air conditioning device located in a certain office, and the air conditioning device in the adjacent office is the involved air conditioning device. For the involved air conditioning device, the computer system first obtains its second air conditioning control record data. The second air conditioning control record data contains the control operation records of the involved air conditioning device at different times.

[0182] At the same time, the computer system also needs to obtain the objective parameter data involved between the involved air-conditioning equipment and the target air-conditioning equipment. The involved category data is used to indicate the type of association between the two. For example, according to the classification of heat interaction, it can be divided into strong heat exchange association (such as air-conditioning equipment between adjacent and well-ventilated rooms), weak heat exchange association (such as air-conditioning equipment between rooms that are far apart or have insulation measures), and other categories. Assume that the involved category data here indicates that the involved air-conditioning equipment and the target air-conditioning equipment are strongly heat exchange associated, represented by code 1 (this is just a simple coding example). The number of involvement data reflects the number of times the association occurs between the two within a certain period of time. For example, in one day, if the air circulation between the two rooms causes the air conditioners to affect each other 5 times due to factors such as people walking, doors and windows opening and closing, then the number of involvement data here is 5.

[0183] In step S120, after obtaining the second air-conditioning control record data and the objective parameter data involving the air-conditioning device, the computer system determines these two parts of data as the involvement analysis element data corresponding to the involved air-conditioning device.

[0184] Let's continue with the example of the office space mentioned above. For the involved air-conditioning equipment in the adjacent office, the control record data such as the temperature setting of 22℃ and the wind speed of low wind speed from 8 to 9 in the morning, combined with the category data (code 1) indicating strong heat exchange association and the data of the number of involvement of 5, constitute the complete involved analysis element data of this involved air-conditioning equipment. This involved analysis element data can comprehensively reflect the self-control of the involved air-conditioning equipment and the relationship characteristics between it and the target air-conditioning equipment. For example, the operating mode tendency of the equipment can be understood from the temperature setting, and the wind speed setting can reflect the control strategy for air flow; while the involved category data and the involved number data can reflect the closeness of the association with the target air-conditioning equipment from the spatial relationship and interaction frequency. This information is of great significance for subsequent steps (such as various calculations and analysis operations based on the edge network to determine the final air-conditioning control strategy, etc.), and can help the computer system to more comprehensively consider the mutual influence between the various devices in the entire air-conditioning system, so as to formulate a more accurate and effective air-conditioning control strategy.

[0185] As an implementation mode, the above method may further include:

[0186] Step S1000: Acquire dimension data corresponding to the analysis dimension associated with the target air-conditioning equipment in the target control area and the variable to be analyzed;

[0187] Step S1100: Determine the weight of the dimension data of the analysis dimension according to the target air-conditioning equipment, and obtain the dimension weight of each analysis dimension;

[0188] Step S1200: filtering the analysis dimensions according to the dimension weights to obtain target analysis dimensions;

[0189] Step S1300: performing a chi-square binning operation on the dimensional data corresponding to the target analysis dimension to obtain binned dimensional data corresponding to the target analysis dimension;

[0190] Step S1400: performing causal analysis on the binned dimension data of the target analysis dimension according to the target air-conditioning equipment, and determining the suspicious analysis dimension causing the sudden change of the variable to be analyzed in the analysis dimension.

[0191] Taking indoor temperature stability as an example of the variable to be analyzed, the dimensional data corresponding to the analysis dimension obtained by the computer system may include the following aspects. For the target air-conditioning equipment, its power size is an analysis dimension. Assuming that the power of an air-conditioning equipment is 3 horsepower, this is the dimensional data corresponding to the power analysis dimension; the running time of the air-conditioning is also an analysis dimension. For example, in the past 24 hours, the air-conditioning has been running for 16 hours. These 16 hours are the dimensional data of the running time analysis dimension; there is also the wind speed setting of the air-conditioning. If the wind speed is set to medium wind speed, this is the dimensional data corresponding to the wind speed analysis dimension; in addition, environmental factors such as the indoor and outdoor temperature difference can also be used as an analysis dimension. Assuming that the indoor temperature is 24℃ and the outdoor temperature is 30℃, then the temperature difference of 6℃ is the dimensional data of the indoor and outdoor temperature difference analysis dimension.

[0192] In step S1100, after obtaining the dimensional data of each analysis dimension, the computer system will determine the dimensional weight of each analysis dimension based on the specific conditions of the target air-conditioning equipment for these analysis dimensions. Continuing with the example of the above indoor temperature stability as the variable to be analyzed. For the analysis dimension of power, if historical data or preliminary experimental analysis show that the power size has a greater impact on the indoor temperature stability, then the computer system may give a higher weight. Assume that according to a certain algorithm or experience setting, the weight of the analysis dimension of power is 0.3.

[0193] For the runtime analysis dimension, although it has a certain impact on temperature stability, the impact is slightly lower than that on power. For example, according to the analysis, its weight is determined to be 0.2.

[0194] The wind speed setting analysis dimension also plays an important role in the uniformity and stability of indoor temperature. It is assumed that it is given a weight of 0.25.

[0195] The indoor and outdoor temperature difference is an environmental factor analysis dimension. Because it is an external uncontrollable factor, but its impact on indoor temperature stability cannot be ignored, the computer system sets its weight to 0.25.

[0196] The weights here can be determined based on a variety of methods, such as statistical analysis of historical data, the learning results of machine learning algorithms (such as regression analysis, etc.) on a large amount of sample data, or prior weights set based on expert experience.

[0197] In step S1200, the computer system performs a filtering operation on the analysis dimension according to the determined dimension weight. For example, a weight threshold is set to 0.2.

[0198] For the above analysis dimensions, the power weight is 0.3, the running time weight is 0.2, the wind speed setting weight is 0.25, and the indoor and outdoor temperature difference weight is 0.25. Since the running time weight is exactly 0.2, which meets the weight threshold requirement, and the weights of power, wind speed setting, and indoor and outdoor temperature difference are all greater than 0.2, these analysis dimensions will be retained as target analysis dimensions.

[0199] The purpose of this step is to screen out the analysis dimensions that have a relatively large or critical impact on the target (such as indoor temperature stability) and remove those dimensions with smaller impacts, thereby reducing the amount of subsequent calculations and focusing on key factors.

[0200] In step S1300, the computer system performs a chi-square binning operation on the dimensional data corresponding to the retained target analysis dimension. Taking the target analysis dimension of wind speed setting as an example, assuming that the wind speed has three gears: low, medium, and high, the corresponding air conditioner operation time is 4 hours, 8 hours, and 4 hours respectively (this is part of the dimensional data corresponding to the wind speed setting analysis dimension).

[0201] When the computer system performs chi-square binning, it first sorts the data. The sorting is from small to large according to the running time: low wind speed - hours, high wind speed - 4 hours, medium wind speed - 8 hours.

[0202] Then, binning is performed according to the chi-square binning algorithm. Assume that according to the algorithm, low wind speed and high wind speed are divided into one box, and medium wind speed is a separate box. In this way, the binning dimension data corresponding to the target analysis dimension of wind speed setting is obtained. For the target analysis dimension of power, if the power has three values ​​of 1.5 horsepower, 2 horsepower, and 3 horsepower, the corresponding indoor temperature stability indicators (assuming that they are obtained through some quantitative method) are 80%, 85%, and 90% respectively. After the chi-square binning operation, 1.5 horsepower and 2 horsepower may be divided into one box, and 3 horsepower is a separate box. This is the binning dimension data corresponding to the target analysis dimension of power. The chi-square binning operation helps to reasonably group continuous or discrete data, explore the structural relationship within the data, and provide a more organized data form for subsequent causal analysis.

[0203] In step S1400, the computer system performs causal analysis based on the binned dimension data of the target analysis dimension of the target air-conditioning equipment.

[0204] Taking indoor temperature stability as an example, suppose that the variable to be analyzed suddenly changes, that is, the indoor temperature suddenly becomes unstable. For the target analysis dimension of wind speed setting, in the bin dimension data, if low wind speed and high wind speed were previously in one bin, and medium wind speed was in a separate bin, the computer system found that when the wind speed suddenly switched from medium wind speed to low wind speed or high wind speed, the indoor temperature stability index showed a significant decrease (mutation). This indicates that the wind speed setting analysis dimension may be a suspicious analysis dimension that causes a sudden change in indoor temperature stability. For the target analysis dimension of power, if the bin dimension data shows that 3 hp is in one bin, 1.5 hp and 2 hp are in one bin, when the air conditioner power is suddenly adjusted from 2 hp to 3 hp, the indoor temperature stability index fluctuates greatly. Then the power analysis dimension is also determined as a suspicious analysis dimension. In this way, the computer system finds suspicious analysis dimensions that may cause sudden changes in the variables to be analyzed in all analysis dimensions, which helps to further investigate the causes of abnormal operation of the air conditioning system, thereby providing a basis for adjusting the air conditioning control strategy or performing equipment maintenance.

[0205] In the specific causal analysis process, there may be many situations. For example, when there is a control period and a target period (such as the target period is the time period when the temperature is unstable, and the control period is the time period of normal operation), the computer system will compare the difference in the binning dimension data of the target analysis dimension in the two periods. If the wind speed frequently switches between different gears in the target period, while the wind speed is relatively stable in the control period, this further strengthens the possibility of the wind speed setting being a suspicious analysis dimension.

[0206] For another example, the causal confidence of the analysis dimension causing the mutation of the variable to be analyzed is calculated based on the binned dimension data. Assuming that for a certain analysis dimension, the causal confidence calculated by a specific algorithm is 0.8 (assuming the value range is 0-1, the closer to 1, the more likely it is to be the cause of the mutation), if this confidence exceeds the preset threshold (such as 0.7), then this analysis dimension is determined to be the analysis dimension that causes the mutation of the variable to be analyzed.

[0207] As an implementation mode, the target air-conditioning device includes a first target air-conditioning device that undergoes a mutation in a target period in which the variable to be analyzed mutates, and a second target air-conditioning device that maintains normal operation in a control period in which the variable to be analyzed is normal. Step S1100 determines the weight of the target air-conditioning device for the dimension data of the analysis dimension to obtain the dimension weight of each analysis dimension, which may include:

[0208] Step S1110: determining a device priori label of the target air-conditioning device based on the air-conditioning device intersection between the first target air-conditioning device that undergoes a sudden change in the target period and the second target air-conditioning device that maintains normal in the target period;

[0209] Step S1120: classifying the target air-conditioning equipment according to the dimension data of the target air-conditioning equipment for the analysis dimension to obtain category information;

[0210] Step S1130: Determine the dimension weight of each analysis dimension corresponding to the target air-conditioning equipment based on the category information and the equipment prior label.

[0211] In step S1110, in an air conditioning system, it is assumed that there are multiple air conditioning devices, and a target cycle is set. In this target cycle, some air conditioning devices may have some conditions, for example, when the indoor temperature stability is a variable to be analyzed, those affected air conditioning devices are the first target air conditioning devices, and the air conditioning devices whose indoor temperature stability is maintained normal in the same target cycle are the second target air conditioning devices.

[0212] For example, suppose there are 10 air conditioners, numbered from 1 to 10. During the target period, the indoor temperature stability of air conditioners 1, 3, and 5 changes suddenly, and these are the first target air conditioners; the indoor temperature stability of air conditioners 2, 4, 6, 8, and 10 remains normal, and these are the second target air conditioners. The computer system will find the intersection of air conditioners between the first target air conditioner and the second target air conditioner. Assume that air conditioner 4 is in the normal device set, and it has some association with the first target air conditioner, such as sharing the same power supply line or being located in a similar spatial area. This intersection will be used by the computer system to determine the device prior tag of the target air conditioner. The device prior tag is a predefined identifier of the characteristics or status of the target air conditioner, which may be a value or code. For example, if there is an intersection device, the computer system may set a device prior tag of 1 for the target air conditioner, indicating that there is a special association; if there is no intersection device, the device prior tag may be set to 0, indicating that there is no such special association. This device prior tag will play an important reference role in the subsequent weight determination process.

[0213] In step S1120, after determining the equipment prior label, the computer system performs classification operations on the dimensional data of each analysis dimension according to the target air-conditioning equipment to obtain category information.

[0214] Continuing with the example of indoor temperature stability, the analysis dimensions may include the power, wind speed setting, and operation time of the air conditioner. For the power analysis dimension, assume that the dimension data is different power values ​​of air conditioners, such as 1.5 HP, 2 HP, and 3 HP. The computer system can classify air conditioners according to these power values. For example, 1.5 HP air conditioners are classified into one category, 2 HP air conditioners are classified into another category, and 3 HP air conditioners are classified into another category. This is a type of classification information obtained based on the power analysis dimension.

[0215] For the analysis dimension of wind speed setting, if the dimension data contains three settings: low wind speed, medium wind speed, and high wind speed, the computer system will classify them according to the current wind speed setting of the target air-conditioning equipment. For example, air-conditioning equipment set to low wind speed is in one category, medium wind speed is in one category, and high wind speed is in one category. For the analysis dimension of operating time, it is assumed that the dimension data is the operating time value of each air-conditioning equipment in the target period, such as some air-conditioning equipment runs for 8 hours, some run for 12 hours, etc. The computer system can classify according to the range of operating time, such as operating time less than 10 hours is in one category, and operating time greater than or equal to 10 hours is in another category. In this way, the computer system obtains corresponding category information for each analysis dimension, and this category information can reflect the characteristic distribution of the target air-conditioning equipment under different analysis dimensions.

[0216] In step S1130, after obtaining the category information and the equipment prior label, the computer system begins to determine the dimension weight of each analysis dimension corresponding to the target air-conditioning equipment.

[0217] Still taking the indoor temperature stability as an example, for the analysis dimension of power. Assume that among the different power categories classified previously, the air conditioners with a power of 3 HP (category 1) account for a high proportion in the first target air conditioners (equipment with temperature stability mutations), and if the equipment is a priori marked as 1 (indicating the existence of special associations), the computer system may believe that the analysis dimension of power has a greater impact on the indoor temperature stability, and thus give it a higher dimension weight. For example, set the dimension weight of the analysis dimension of power to 0.3.

[0218] For the analysis dimension of wind speed setting, if the proportion of air-conditioning equipment with low wind speed settings in the first target air-conditioning equipment is also high, and the equipment prior mark is 1, this indicates that the low wind speed setting may be associated with the sudden change in indoor temperature stability. The computer system may give the wind speed setting analysis dimension a weight of 0.25.

[0219] As for the analysis dimension of operating time, if the proportion of air-conditioning equipment with an operating time greater than or equal to 10 hours in the first target air-conditioning equipment is relatively low, and the equipment is priori marked as 0 (no special association), the computer system may believe that the analysis dimension of operating time has a relatively small impact on indoor temperature stability, and the dimension weight given may be 0.15.

[0220] The determination of dimension weights here is the result of comprehensive consideration of category information and equipment prior labels. Category information reflects the distribution characteristics of target air-conditioning equipment under each analysis dimension, and equipment prior labels reflect the special associations that may exist between target air-conditioning equipment.

[0221] As an implementation mode, the analysis dimension includes a dimension index, and the dimension index corresponds to the binned dimension data. In step S1400, causal analysis is performed on the binned dimension data of the target analysis dimension according to the target air-conditioning equipment to determine the analysis dimension causing the mutation of the variable to be analyzed, including:

[0222] Step S1410: obtaining a reasoning result corresponding to the target analysis dimension based on the dimension data of the first target air-conditioning equipment that undergoes a sudden change in the target period with respect to the analysis dimension;

[0223] Step S1420: obtaining an actual value corresponding to the target analysis dimension based on the dimension data of the second target air-conditioning equipment that maintains normal operation during the comparison period for the analysis dimension;

[0224] Step S1430: Calculate based on the inference result and actual value corresponding to the target analysis dimension to determine the suspicious analysis dimension that causes the sudden change of the target weekly variable to be analyzed, and the suspicious dimension index associated with the sudden change of the variable to be analyzed under the suspicious analysis dimension.

[0225] In step S1410, during the operation of the air conditioning system, a target period is set, during which certain conditions may change, such as a sudden change in the variable to be analyzed, i.e., indoor temperature stability. The first target air conditioning equipment is an air conditioning equipment that has a phenomenon related to the sudden change in the variable to be analyzed during the target period.

[0226] Taking the sudden change of indoor temperature stability as an example, assume that the target analysis dimension is the wind speed setting of the air conditioner. For the first target air conditioner, the computer system will obtain the dimension data of the wind speed setting analysis dimension during the target period. Assume that the dimension data of the wind speed setting of a first target air conditioner during the target period is: it initially runs at a low wind speed for a period of time, then switches to a high wind speed, and then frequently switches between low and high wind speeds.

[0227] The computer system performs analytical reasoning based on such dimensional data. For example, based on a pre-established air conditioning operation model or based on empirical rules, the computer system can infer that such frequent switching of wind speed may cause unstable indoor air flow, thereby affecting the stability of indoor temperature. This is the reasoning result corresponding to the target analysis dimension (wind speed setting) obtained based on the dimensional data of the wind speed setting analysis dimension of the first target air conditioning device, that is, there is a potential causal relationship between the frequent switching of wind speed and the stability of indoor temperature.

[0228] In step S1420, the control period is a concept relative to the target period. The variables to be analyzed remain normal during the control period, and the second target air-conditioning device is an air-conditioning device that operates normally during the control period. Similarly, the wind speed setting is used as the target analysis dimension. For the second target air-conditioning device, the computer system obtains the dimensional data on the wind speed setting during the control period. It is assumed that the wind speed settings of these devices are relatively stable during the control period, and are maintained at medium wind speeds most of the time, and rarely switch.

[0229] This stable wind speed setting (medium wind speed) and stable operation status are the actual values ​​corresponding to the target analysis dimension (wind speed setting). This actual value reflects the normal state value of the wind speed setting analysis dimension when the indoor temperature is stable (the variable to be analyzed is normal).

[0230] In step S1430, the computer system performs calculation and analysis based on the previously obtained inference results and actual values. Continuing with the wind speed setting as an example, the inference results obtained from step S1410 indicate that frequent switching of wind speed (between low wind speed and high wind speed) may affect temperature stability, while the actual value obtained in step S1420 is a stable medium wind speed state. The computer system can determine that the wind speed setting is a suspicious analysis dimension by comparing these two states. For example, the difference between the two is calculated, assuming that a simple method of quantifying the difference is adopted to quantify the wind speed state, with low wind speed as 1, medium wind speed as 2, and high wind speed as 3. For the case of wind speed switching of the first target air-conditioning device, indicators such as the switching frequency and the average value of the numerical difference between different wind speeds can be calculated; for the stable medium wind speed state of the second target air-conditioning device, its value is 2 and there is no switching. By comparing these quantified values, it is found that there is a significant difference between the two, thereby determining that the wind speed setting is a suspicious analysis dimension that causes a sudden change in indoor temperature stability within the target period.

[0231] After determining that wind speed is set as a suspicious analysis dimension, further determine the suspicious dimension indicators. In this example, the switching frequency of wind speed and the difference between wind speeds are suspicious dimension indicators. This is because it is the frequent switching of wind speed (from low wind speed to high wind speed or vice versa) and the large difference between different wind speeds (such as the change from 1 to 3) that are inferred to be associated with the sudden change in indoor temperature stability.

[0232] In this way, the computer system can find out the suspicious analysis dimensions that may cause mutations in the variables to be analyzed and the suspicious dimension indicators associated with the mutations based on the inference results and actual values ​​of the target analysis dimensions in different cycles (target cycle and control cycle). This helps to more deeply analyze the factors affecting the variables to be analyzed in the air-conditioning system and provide a basis for subsequent air-conditioning control strategy adjustments or troubleshooting operations.

[0233] As an implementation mode, step S1430, performing calculations based on the inference results and actual values ​​corresponding to the target analysis dimension to determine the suspicious analysis dimension causing the sudden change of the target week variable to be analyzed, and the suspicious dimension indicators associated with the sudden change of the variable to be analyzed under the suspicious analysis dimension, may include:

[0234] Step S1431: According to the inference result and the actual value, the variance of the dimension indicator under the analysis dimension is determined to obtain the variance of the dimension indicator;

[0235] Step S1432: Based on the variance of the dimension index, determine the suspicious analysis dimension associated with the sudden change of the variable to be analyzed in the target period and the suspicious dimension index under the suspicious analysis dimension.

[0236] In step S1431, in the air-conditioning system, the analysis dimension includes multiple dimensional indicators, and the computer system determines the variance of these dimensional indicators by comparing the inference results and the actual values.

[0237] Taking the wind speed setting of the air conditioner as an analysis dimension, its dimension indicators include the wind speed and the frequency of wind speed switching. In the target period (assuming that the indoor temperature stability changes suddenly), for the first target air conditioner (the air conditioner with the problem), the wind speed setting can be known from the inference results. For example, the wind speed frequently switches between low wind speed (assuming that it is numerically 1) and high wind speed (numerically 3), and the wind speed switches 5 times in a specific time period.

[0238] During the control period (stable indoor temperature), for the second target air-conditioning equipment (normally operating air-conditioning equipment), the actual value is that the wind speed is always maintained at a medium wind speed (numerically 2), and the wind speed switching frequency is 0 times.

[0239] The computer system calculates the variance for the wind speed dimension. First, the average wind speed of each device in each cycle is calculated. For the first target air conditioning device, since the wind speed switches between 1 and 3, the average value is (1 +3) ÷ 2 = 2; for the second target air conditioning device, the wind speed is always 2, and the average value is also 2.

[0240] Then calculate the variance. The formula for calculating the variance is (where x i is each data point, is the mean value, and n is the number of data points). For the first target air conditioning equipment, assuming that each wind speed is regarded as a data point, the variance is For the second target air conditioning equipment, since the wind speed is always 2, the variance is 0.

[0241] For the wind speed switching frequency dimensional indicator, the variance is also calculated. The average switching frequency of the first target air-conditioning equipment is 5÷1=5 (assuming only one time period is considered), and the variance is 0 (because there is only one data point and no fluctuation); the average switching frequency of the second target air-conditioning equipment is (0÷1=0), and the variance is also 0.

[0242] By calculating the variance of dimensional indicators, the computer system can quantify the fluctuations of dimensional indicators in different periods. The larger the variance, the greater the difference between the target period and the control period of the dimensional indicator, and the more likely it is to be related to the mutation of the variable to be analyzed.

[0243] In step S1432, the computer system determines the suspicious analysis dimension and the suspicious dimension indicator according to the variance of the dimension indicator calculated in the previous step.

[0244] Continuing with the above example of wind speed setting, the variance of wind speed size is 1 in the first target air-conditioning device and 0 in the second target air-conditioning device; the variance of wind speed switching frequency is 0 in the first target air-conditioning device (this is due to the special case of a single data point, and in practice, different results may occur if there are multiple data points), and is 0 in the second target air-conditioning device.

[0245] Since the variance of wind speed is significantly different between the two devices (1 > 0), the computer system determines that wind speed is set as a suspicious analysis dimension associated with a sudden change in indoor temperature stability during the target period. Under this suspicious analysis dimension, the dimension indicator of wind speed is determined to be a suspicious dimension indicator due to its large variance.

[0246] Let's take the power of an air conditioner as an example of an analysis dimension. Its dimensional indicators include the power size and the power adjustment frequency. Assume that in the target period, the power size of the first target air conditioner is adjusted between 1.5 horsepower (numerically 1.5) and 3 horsepower (numerically 3) for 3 times; in the control period, the power size of the second target air conditioner is always maintained at 2 horsepower (numerically 2), and the adjustment frequency is 0 times. Calculate the variance of the power size. The average value of the first target air conditioner is (1.5+ 3)÷2=2.25, and the variance is ; The variance of the second target air conditioning device is 0. For the power adjustment frequency, the average value of the first target air conditioning device is 3÷1=3, and the variance is 0; the average value of the second target air conditioning device is 0÷1=0, and the variance is 0. Since the variance of power size is different between the two devices, 0.5625 > 0, the computer system determines that the air conditioning power is a suspicious analysis dimension, and the power size is a suspicious dimension indicator.

[0247] As another implementation, there are multiple control cycles. Step S1400, performing causal analysis on the binned dimension data of the target analysis dimension according to the target air-conditioning equipment to determine the analysis dimension causing the mutation of the variable to be analyzed, may include:

[0248] Step S1401: performing causal analysis based on the binned dimension data corresponding to each control period and the binned dimension data corresponding to the target period, and obtaining candidate analysis dimensions for which the binned dimension data corresponding to the target period has a sudden change when compared with the dimension data of each control period;

[0249] Step S1402: Determine the analysis dimension that causes the mutation of the variable to be analyzed within the target period based on the candidate dimension data.

[0250] In step S1401, in the scenario of air conditioning system operation, there are target cycles (cycles in which the variables to be analyzed change suddenly) and multiple control cycles (cycles in which the variables to be analyzed remain normal). The binning dimension data of the target analysis dimension is different in different cycles, and the computer system compares these data to find possible reasons.

[0251] Indoor temperature stability is used as the variable to be analyzed, and the operating mode and wind speed setting of the air conditioner are assumed to be the target analysis dimensions. For the operating mode dimension, it may be divided into cooling mode, heating mode, dehumidification mode, etc.; for the wind speed setting dimension, it may be divided into low wind speed, medium wind speed, and high wind speed.

[0252] In the target cycle, it is assumed that the binning dimension data of a target air-conditioning device is as follows: the operation mode frequently switches between cooling and dehumidification modes, and the wind speed setting suddenly changes from low wind speed to high wind speed.

[0253] For control cycle 1, the binning dimension data shows that the operation mode has been stable in cooling mode, and the wind speed setting has been medium wind speed. Comparing the data of the target cycle and control cycle 1, the computer system finds that the two dimensions of operation mode and wind speed setting have changed differently in the target cycle than in control cycle 1. Therefore, operation mode and wind speed setting become candidate analysis dimensions for which the corresponding binning dimension data in the target cycle has a sudden change when compared with the dimension data of control cycle 1.

[0254] Looking at the control period 2, its binning dimension data shows that the operating mode is mainly heating mode (occasionally switched to dehumidification mode), and the wind speed is set to low wind speed. After comparing with the data of the target period, the computer system once again found that the operating mode and wind speed settings have obvious differences in mode switching frequency and wind speed size. Therefore, under the comparison of the dimensional data of the control period 2, the operating mode and wind speed setting are still candidate analysis dimensions for the corresponding binning dimension data in the target period to have a sudden change.

[0255] In this way, the computer system performs similar comparative analysis for each control cycle, and selects the analysis dimensions that show significant differences from the target cycle when compared with multiple control cycles as candidate analysis dimensions.

[0256] In step S1402, after obtaining the candidate analysis dimensions, the computer system further determines the analysis dimensions that are ultimately related to causing the mutation of the variable to be analyzed within the target period based on the candidate dimension data.

[0257] Continuing with the above example, the operating mode and wind speed setting are candidate analysis dimensions. The computer system will further analyze the data characteristics of these two candidate dimensions.

[0258] For the operation mode, the computer system may count the switching frequency between different modes in the target cycle, the fluctuation amplitude of the indoor temperature at each switch, and other data. Assume that in the target cycle, the indoor temperature fluctuates by an average of 3°C each time the cooling and dehumidification modes are switched, while the temperature fluctuation amplitude is less than 1°C when different modes are switched (if any) in the control cycle. This significant difference in temperature fluctuations indicates that the operation mode is strongly associated with the sudden change of the variable to be analyzed, the indoor temperature stability. For the wind speed setting, the computer system analyzes the change of indoor temperature in a short period of time when switching from low wind speed to high wind speed in the target cycle, and the difference from the indoor temperature when the wind speed is stable in the control cycle. For example, it is found that the indoor temperature drops by 2°C in 10 minutes when the wind speed is switched in the target cycle, while the temperature changes very little when the wind speed is stable in the control cycle. This also shows that the wind speed setting is closely related to the sudden change of indoor temperature stability.

[0259] As another implementation, step S1400, performing causal analysis on the binned dimension data of the target analysis dimension according to the target air-conditioning equipment to determine the analysis dimension causing the mutation of the variable to be analyzed, may include:

[0260] Step S1400A: Calculate the causal confidence of the sudden change of the variable to be analyzed caused by the analysis dimension based on the binned dimension data;

[0261] Step S1400B: Determine the analysis dimension that causes the sudden change of the variable to be analyzed from the analysis dimensions based on the causal confidence.

[0262] First, for step S1400A, the computer system calculates the causal confidence of the sudden change of the variable to be analyzed caused by the analysis dimension based on the binned dimension data. The binned dimension data is obtained through the processing of the previous steps. Specifically, the computer system first obtains the dimension data corresponding to the target analysis dimension, sorts it (such as sorting the dimension data of the target air-conditioning equipment for the target analysis dimension in step S1310 to obtain the sorted dimension data), and then performs chi-square binning based on the sorted dimension data (such as step S1320), thereby obtaining the binned dimension data.

[0263] Causal confidence is an indicator that measures the reliability of the causal relationship between a certain analysis dimension and the mutation of the variable to be analyzed. When calculating causal confidence, the computer system needs to consider various factors in the binned dimension data and their relationship with the mutation of the variable to be analyzed. For example, assuming that the variable to be analyzed is the mutation of the cooling efficiency of the air conditioner, the target analysis dimension is the speed of the air conditioner compressor, the temperature of the evaporator, and the heat dissipation efficiency of the condenser. The computer system will calculate the causal confidence based on the binned dimension data of these dimensions. If the binned dimension data of the air conditioner compressor speed shows that the speed fluctuation range is strongly correlated with the mutation of the cooling efficiency in a certain period of time, for example, when the compressor speed changes in a certain range, the cooling efficiency always changes suddenly, then the causal confidence in this dimension may be high. For the dimension of evaporator temperature, if its binned dimension data shows that the situation of the sudden change of cooling efficiency in different temperature ranges is more complicated and interacts with other factors, then the causal confidence in this dimension may be relatively low.

[0264] Next is step S1400B, where the computer system determines the analysis dimension that causes the mutation of the variable to be analyzed from the analysis dimension based on the causal confidence. Here, the computer system makes a judgment based on the value of the causal confidence calculated in step S1400A. Continuing with the above example, assume that the computer system calculates that the causal confidence of the compressor speed dimension is 0.8, the causal confidence of the evaporator temperature dimension is 0.3, and the causal confidence of the condenser heat dissipation efficiency dimension is 0.6. The computer system will set a threshold, such as 0.5, and the analysis dimensions with a causal confidence higher than this threshold will be determined as the analysis dimensions that cause the mutation of the variable to be analyzed. In this example, the causal confidence of the compressor speed dimension and the condenser heat dissipation efficiency dimension is higher than 0.5, so the computer system will determine that these two dimensions are the analysis dimensions that cause the mutation of the refrigeration efficiency.

[0265] In the fluorine pump air conditioning system, there are multiple factors (analysis dimensions) that may affect the performance of the air conditioning (variables to be analyzed), and these factors are interrelated and complex. By calculating the causal confidence, the strength of the causal relationship between each analysis dimension and the mutation of the variable to be analyzed can be quantified, so as to accurately find the key influencing factors. This helps to adjust and optimize these key factors in a targeted manner in the subsequent formulation of air conditioning control strategies. For example, if it is determined that the compressor speed and the condenser heat dissipation efficiency are the key dimensions that affect the mutation of refrigeration efficiency, then in the control strategy, we can focus on the adjustment of the relevant parameters of these two dimensions, such as optimizing the control algorithm of the compressor speed, or improving the heat dissipation conditions of the condenser, so as to improve the temperature stability of the air conditioner and achieve better control effects. At the same time, this determination method also takes into account the synergistic relationship between the various components (the entity parts corresponding to the analysis dimensions) in the fluorine pump air conditioning system. The operating states of different components will affect each other, and the change of one component may affect other components and the final air conditioning performance in many ways. By calculating the causal confidence, these complex relationships can be comprehensively considered to avoid determining the influencing factors based on a single association, thereby improving the accuracy and reliability of the determination results. For example, although the evaporator temperature is related to the refrigeration efficiency, it may be affected by multiple factors such as the compressor speed and the condenser heat dissipation efficiency. Adjusting the evaporator temperature alone may not effectively solve the problem of sudden changes in refrigeration efficiency. By calculating the causal confidence, more critical factors can be accurately identified, providing a basis for formulating effective control strategies.

[0266] As an implementation manner, step S1300, performing a chi-square binning operation on the dimensional data corresponding to the target analysis dimension to obtain the binned dimensional data corresponding to the target analysis dimension, may include:

[0267] Step S1310: sorting the dimension data of the target air-conditioning equipment according to the target analysis dimension to obtain sorted dimension data corresponding to the target analysis dimension;

[0268] Step S1320: Perform chi-square binning based on the sorted dimensional data to obtain binned dimensional data of the target air-conditioning equipment for the target analysis dimension.

[0269] First, in step S1310, the computer system sorts the dimensional data of the target air-conditioning equipment for the target analysis dimension, thereby obtaining the sorted dimensional data corresponding to the target analysis dimension. The target analysis dimension is the analysis dimension associated with the target air-conditioning equipment and the variable to be analyzed determined in the previous step. For example, in a fluorine pump air-conditioning system, the variable to be analyzed is the energy consumption of the air-conditioning, and the target analysis dimension may include the operating time of the air-conditioning, the operating mode (cooling, heating, dehumidification, etc.), the ambient temperature, etc. After the computer system obtains the dimensional data corresponding to these target analysis dimensions, it starts the sorting operation. The purpose of this sorting operation is to prepare for the subsequent chi-square binning operation, so that the data has a certain order, which is convenient for data grouping and analysis. By sorting, the distribution law of the data can be observed more clearly, which helps to discover the potential patterns in the data. For example, from the sorted operating time data, it can be preliminarily seen that the approximate distribution range of the air-conditioning operating time is relatively concentrated or dispersed, which is of great significance for analyzing the relationship between air-conditioning energy consumption and operating time.

[0270] In step S1320, the computer system performs chi-square binning based on the sorted dimensional data, thereby obtaining binned dimensional data of the target air-conditioning equipment for the target analysis dimension. Chi-square binning is a method of data discretization that divides continuous data or numerical data into different intervals (boxes). The computer system first needs to determine the number of bins, assuming that it is determined to be 3 boxes. Then, the computer system determines the boundaries of each box based on the chi-square statistic. The chi-square statistic measures the degree of difference between the actual observed value and the theoretical expected value. In this example, the computer system calculates the chi-square statistic under different partitioning methods to find the optimal binning scheme. The computer system calculates the chi-square statistic based on the distribution of data in each box. If under a certain partitioning scheme, the distribution of data in each box is the smallest difference from the expected distribution (which can be determined based on the overall characteristics of the data or prior knowledge), that is, the chi-square statistic is the smallest, then this partitioning scheme is determined as the optimal scheme. For other target analysis dimensions, the computer system also performs chi-square binning operations. On the one hand, this chi-square binning operation discretizes continuous dimensional data, reducing the complexity of the data and making subsequent analysis and processing more convenient. For example, when analyzing the relationship between air conditioning energy consumption and ambient temperature, the binned ambient temperature data can be more intuitively associated with the energy consumption data for analysis without the need to process continuous temperature values. On the other hand, the binning operation helps to discover nonlinear relationships in the data. In the fluorine pump air conditioning system, the relationship between many variables is not a simple linear relationship, and these complex relationships can be better captured through chi-square binning. For example, the relationship between air conditioning energy consumption and operating time may show different relationship patterns in different operating time intervals, and the binned dimensional data after chi-square binning can more effectively reveal this relationship.

[0271] An embodiment of the present invention provides a computer system, such as Figure 2 As shown, the computer system 100 includes: a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, such as through a bus 102. Optionally, the computer system 100 may also include a transceiver 104. It should be noted that in actual applications, the transceiver 104 is not limited to one, and the structure of the computer system 100 does not constitute a limitation on the embodiments of the present invention.

Claims

1. A fluorine pump air conditioner temperature stabilization control method, characterized in that: include: Acquire a target analysis factor data set of a target air-conditioning device to be controlled and a data set of analysis factors of air-conditioning devices involved in the target air-conditioning device, wherein the target analysis factor data set includes: an air-conditioning objective parameter factor and first air-conditioning control record data; Obtaining an implicit array of objective parameters of the air conditioner objective parameter elements based on an objective parameter representation component of the edge side network, and obtaining an implicit array of control records of the first air conditioner control record data based on a control record representation component of the edge side network; A weighted focusing component based on the edge side network obtains a fused implicit array corresponding to the air conditioning objective parameter element and the involved analysis element data set; The integration component based on the edge side network integrates the objective parameter implicit array, the control record implicit array and the fusion implicit array to obtain the air-conditioning equipment implicit array of the target air-conditioning equipment; Based on the air conditioning equipment implicit array of the target air conditioning equipment, obtaining a target control strategy from a control command set, and controlling the target air conditioning equipment based on the target control strategy; The edge side network includes an embedded coding component, the objective parameter characterization component includes a feedforward network component, and the objective parameter characterization component based on the edge side network obtains an objective parameter implicit array of the air conditioner objective parameter elements, including: The objective parameters obtained by embedding the objective parameter elements of the air conditioner by the embedded coding component based on the edge side network are embedded into an implicit array as loading data of the objective parameter characterization component; Based on the feedforward network component of the objective parameter characterization component, information mapping is performed on the objective parameter embedded implicit array to obtain the objective parameter implicit array of the air conditioner objective parameter element; The edge side network includes an embedded coding component, and the control record representation component based on the edge side network obtains an implicit array of control records of the first air conditioning control record data, including: Embed multiple air-conditioning control records of the first air-conditioning control record data obtained by the embedded coding component of the edge side network into an implicit array as loading data of the control record representation component of the edge side network, wherein the first air-conditioning control record data includes multiple multi-dimensional control command data arranged in time sequence according to control moments; The control record representation component based on the edge side network processes the multiple air conditioning control record embedded implicit arrays to obtain the control record implicit array of the first air conditioning control record data; The control record representation component of the edge side network obtains an implicit array of control records of the first air conditioner control record data based on a plurality of multi-dimensional control command data whose control duration is greater than a preset duration; Alternatively, the control record representation component of the edge side network obtains multiple air-conditioning control record mapping implicit arrays corresponding to the multiple air-conditioning control record embedded implicit arrays based on the Transformer component of the control record representation component, and processes the multiple air-conditioning control record mapping implicit arrays based on the downsampling module of the control record representation component to obtain the control record implicit array of the first air-conditioning control record data; The involved analysis element data set includes x involved analysis element data corresponding to x involved air-conditioning equipment, x≥1; the edge side network includes an embedded coding component, and the weight focusing component includes x equipment weight focusing modules, a Transformer component and a downsampling module; the weight focusing component based on the edge side network obtains a fused implicit array corresponding to the air-conditioning objective parameter element and the involved analysis element data set, including: Embedding the objective parameter implicit array obtained by embedding the air conditioner objective parameter element by the embedded coding component based on the edge side network, and embedding the x involved analysis element data by the embedded coding component based on the edge side network to obtain x involved embedded coding implicit arrays as loading data of the weight focusing component; Based on the x device weight focusing modules of the weight focusing component, the objective parameter embedding implicit array and the x involved embedding coding implicit arrays are processed to obtain x involved weight implicit arrays; The Transformer component based on the weight focusing component obtains the implicit array of involvement mappings corresponding to the x implicit arrays of involvement weights; Processing the involved mapping implicit array based on the downsampling module of the weight focusing component to obtain the fused implicit array; The edge side network includes an embedded coding component, the embedded coding component is used to embed coding the element data to be loaded, the element data to be loaded includes at least one of the air conditioner objective parameter element, the first air conditioner control record data, and the involved analysis element data set; If the element data to be loaded is binary coded data, embedding-coding the element data to be loaded based on the embedded coding component; If the element data to be loaded is multiple binary coded data, embedding coding is performed on the element data to be loaded based on the embedded coding component, and a downsampling operation is performed on the coded array obtained after the embedded coding, and the dimension of the coded array obtained after the downsampling operation meets the preset requirements; The involved analysis element data set includes x involved analysis element data corresponding to x involved air-conditioning equipment, and obtaining the involved analysis element data set of the air-conditioning equipment involved in the target air-conditioning equipment includes: Acquire second air-conditioning control record data of any involved air-conditioning device among the x involved air-conditioning devices, and objective parameter data of involvement between the any involved air-conditioning device and the target air-conditioning device, wherein the objective parameter data of involvement includes involvement category data and involvement frequency data; The second air-conditioning control record data and the involved objective parameter data are determined as the involved analysis element data corresponding to any one of the involved air-conditioning devices.

2. The method according to claim 1, characterized in that The control command set stores a plurality of air conditioning control strategies, each of which has a corresponding control strategy implicit array constructed based on the central side network, and the control strategy implicit array of the target control strategy obtained from the control command set matches the air conditioning equipment implicit array of the target air conditioning equipment; The central side network and the edge side network are obtained based on collaborative tuning. The cost parameters determined by the cost measurement function used in the collaborative tuning link include: an implicit array of air-conditioning equipment of the sample air-conditioning equipment obtained based on the edge side network during the tuning process, an implicit array of control strategies of the control strategy learning samples obtained based on the central side network during the tuning process, and a priori marks determined based on the maintenance status of the control strategy learning samples by the sample air-conditioning equipment.

3. The method according to claim 2, characterized in that The method further comprises: Acquire a control command sequence of a target air conditioning control strategy, wherein the control command sequence includes: distribution position data and control commands of different control dimensions; Generate dimensional implicit arrays of different control dimensions according to each control command and the distribution position data, and integrate the dimensional implicit arrays of different control dimensions to obtain a multi-dimensional integrated implicit array; Obtaining a multi-dimensional transformation implicit array corresponding to the multi-dimensional integration implicit array based on the Transformer component of the central side network; Based on the feedforward network component of the central side network, information mapping is performed on the multi-dimensional transformation implicit array to obtain the control strategy implicit array of the target air conditioning control strategy; The target air-conditioning control strategy is one of the plurality of air-conditioning control strategies.

4. The method according to claim 3, characterized in that Get the control commands of different control dimensions of the target air conditioning control strategy, including: Acquire the temperature control strategy and wind speed control strategy included in the target air conditioning control strategy, and obtain a control command in the temperature dimension and a control command in the wind speed dimension; Based on the control command of the temperature dimension and the control command of the wind speed dimension, the control commands of different control dimensions of the target air-conditioning control strategy are determined.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Generate air conditioning control strategy; Constructing implicit arrays of control strategies for generating air conditioning control strategies based on the central side network; Performing a commonality measurement operation based on the implicit array of control strategies of each generated air-conditioning control strategy to obtain a commonality measurement result, wherein the commonality measurement result represents the degree of commonality between any two air-conditioning control strategies in each generated air-conditioning control strategy; performing a redundancy removal operation on the generated air conditioning control strategies based on the commonality between any two air conditioning control strategies in each generated air conditioning control strategy; When constructing the implicit array of control strategies for each generated air conditioning control strategy based on the central side network, the loading information of the central side network is the temperature control strategy and wind speed control strategy for each generated air conditioning control strategy.

6. The method according to claim 1, characterized in that The x device weight focusing modules based on the weight focusing component process the objective parameter embedding implicit array and the x involved embedding coding implicit arrays to obtain x involved weight implicit arrays, including: Obtaining a Kronecker product array between the objective parameter embedding implicit array and the pth involved embedding coding implicit array among the x involved embedding coding implicit arrays, 1≤p≤x; Embedding the objective parameter into an implicit array, the pth implicit array involving embedded coding, and the Kronecker product array for splicing, and loading the spliced ​​array into the nonlinear transformation unit of the pth device weight focusing module among the x device weight focusing modules, and obtaining the execution result of the nonlinear transformation unit of the pth device weight focusing module; The execution result of the nonlinear transformation unit of the p-th device weight focusing module is determined as the loading data of the fully connected unit of the p-th device weight focusing module, and the p-th implicit array of involved weights is obtained based on the fully connected unit of the p-th device weight focusing module.

7. A computer system, characterized in that: include: one or more processors; Memory; one or more computer programs; The one or more computer programs are stored in the memory and configured to be executed by the one or more processors, and when the one or more computer programs are executed by the processors, the method according to any one of claims 1 to 6 is implemented.

Images