A cloud server and a control method of an air conditioning system

Abstract

This invention discloses a control method for a cloud server and an air conditioning system, relating to the field of air conditioning technology, and aims to solve the problem of low accuracy in predicting user-defined modes using current RNN models. The method includes acquiring environmental parameters, operating parameters, input context, and transition context of a multi-split central air conditioning system. The input context includes time information and weather parameters for the current prediction of the user-defined mode, and the transition context includes the time difference between the current predicted user-defined mode and the previous predicted user-defined mode. The predicted user-defined mode is then determined based on the environmental parameters, operating parameters, input context, transition context, and a first neural network model.

Description

Technical Field

[0001] This invention relates to the field of air conditioning technology, and in particular to a cloud server and a control method for an air conditioning system. Background Technology

[0002] In the field of air conditioning technology, the prediction of user-defined modes is currently mainly achieved through traditional recurrent neural network (RNN) models. However, by inputting indoor environmental parameters (such as indoor temperature, indoor humidity, and indoor air volume) into the input matrix of the RNN model, the inherent limitations of the RNN model lead to low accuracy in predicting user-defined modes, thus failing to meet users' comfort requirements. Summary of the Invention

[0003] The embodiments of the present invention provide a control method for a cloud server and an air conditioning system, which solves the problem of low accuracy in predicting user-defined modes using RNN models.

[0004] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions.

[0005] In a first aspect, this application provides a cloud server that communicates with a central controller in a multi-split central air conditioning system. The multi-split central air conditioning system includes multiple indoor units. The cloud server includes: a parameter acquisition unit configured to acquire environmental parameters, operating parameters, input context, and transfer context of the multi-split central air conditioning system. The input context includes time information and weather parameters for the current prediction of the user-defined mode. The transfer context includes the time difference between the current prediction of the user-defined mode and the previous prediction of the user-defined mode. A mode determination unit configured to determine the current predicted user-defined mode based on the environmental parameters, operating parameters, input context, transfer context, and a first neural network model.

[0006] Therefore, this application determines the predicted user setting mode by acquiring the environmental parameters, operating parameters, input context, and transition context of the multi-split central air conditioning system, as well as a first neural network model. The input context includes the time information and weather parameters at the time of the current user setting mode prediction, and the transition context includes the time difference between the current predicted user setting mode and the previous predicted user setting mode. Compared to existing technologies, which predict user setting modes by inputting indoor environmental parameters into the input matrix of the RNN model, this application adds input context and transition context to the prediction of user setting modes, resulting in more accurate predictions.

[0007] In some embodiments, the pattern determination unit is further configured to determine the user setting mode for the current prediction based on the previous user setting mode, environmental parameters, operating parameters, input context, transition context, and the first neural network model when the time difference between the current predicted user setting mode and the previous predicted user setting mode is less than or equal to a preset threshold. When the time difference between the current predicted user setting mode and the previous predicted user setting mode is greater than the preset threshold, the user setting mode for the current prediction is determined based on the environmental parameters, operating parameters, input context, transition context, and the first neural network model obtained when the current predicted user setting mode is used.

[0008] In some embodiments, environmental parameters include at least one of the temperature and humidity of the room where the indoor unit is located during the current user-defined mode prediction. Operating parameters include the airflow of the indoor unit during the current user-defined mode prediction. Time information for the current user-defined mode prediction includes at least one weekday and ten-day period of the month during the current user-defined mode prediction. Weather parameters include the weather type during the current user-defined mode prediction. User-defined modes include heating mode, cooling mode, and dehumidification mode.

[0009] In some embodiments, the first neural network model includes multiple sets of correspondences, each set including a set of historical environmental parameters, historical operating parameters, historical input context, historical transition context, and a historical user-defined pattern correspondence. The pattern determination unit is further configured to determine the user-defined pattern for the current prediction based on the environmental parameters, operating parameters, input context, transition context, and multiple sets of correspondences obtained when predicting the user-defined pattern locally.

[0010] In some embodiments, the parameter acquisition unit is further configured to acquire multiple sets of training data, each set of training data including sampled data of multiple variables at different set temperatures. The multiple variables include historical environmental parameters, historical operating parameters, historical input context, historical transition context, and historical user-defined modes. The cloud server also includes a model training unit configured to use the multiple sets of training data as training data for a recurrent neural network (RNN) model to obtain a first neural network model.

[0011] Secondly, this application provides a control method for an air conditioning system, applied to a cloud server. The cloud server communicates with a central controller in a multi-split central air conditioning system, which includes multiple indoor units. The method includes: acquiring environmental parameters, operating parameters, input context, and transition context of the multi-split central air conditioning system. The input context includes time information and weather parameters for the current prediction of the user-defined mode. The transition context includes the time difference between the current prediction of the user-defined mode and the previous prediction of the user-defined mode. The predicted user-defined mode is determined based on the environmental parameters, operating parameters, input context, transition context, and a first neural network model.

[0012] Regarding the beneficial effects of the second aspect, please refer to the first aspect; it will not be elaborated upon here.

[0013] In some embodiments, determining the predicted user setting mode based on environmental parameters, operating parameters, input context, transition context, and the first neural network model includes: when the time difference between the current predicted user setting mode and the previous predicted user setting mode is less than or equal to a preset threshold, determining the current predicted user setting mode based on the previous predicted user setting mode, environmental parameters, operating parameters, input context, transition context, and the first neural network model. When the time difference between the current predicted user setting mode and the previous predicted user setting mode is greater than the preset threshold, determining the current predicted user setting mode based on the environmental parameters, operating parameters, input context, transition context, and the first neural network model.

[0014] In some embodiments, environmental parameters include at least one of the temperature and humidity of the room where the indoor unit is located during the current user-defined mode prediction. Operating parameters include the airflow of the indoor unit during the current user-defined mode prediction. Time information for the current user-defined mode prediction includes at least one weekday and ten-day period of the month during the current user-defined mode prediction. Weather parameters include the weather type during the current user-defined mode prediction. User-defined modes include heating mode, cooling mode, and dehumidification mode.

[0015] In some embodiments, the first neural network model includes multiple sets of correspondences, each set including a set of historical environmental parameters, historical operating parameters, historical input context, historical transition context, and a historical user setting mode correspondence. Determining the user setting mode for the current prediction based on the environmental parameters, operating parameters, input context, transition context, and the first neural network model includes: determining the user setting mode for the current prediction based on the environmental parameters, operating parameters, input context, transition context, and multiple sets of correspondences obtained during local prediction of the user setting mode.

[0016] In some embodiments, before determining the predicted user setting pattern based on environmental parameters, operating parameters, input context, transition context, and the first neural network model, the method further includes: acquiring multiple sets of training data, each set of training data including sampled data of multiple variables at different set temperatures. The multiple variables include historical environmental parameters, historical operating parameters, historical input context, historical transition context, and historical user setting patterns. The multiple sets of training data are then used as training data for a recurrent neural network (RNN) model to obtain the first neural network model. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a traditional RNN model.

[0018] Figure 2 This is a schematic diagram of the input and output of each layer of a traditional RNN model.

[0019] Figure 3 A schematic diagram of each layer of an optimized RNN model provided in this application;

[0020] Figure 4 A schematic diagram of an air conditioning system architecture is provided for this application;

[0021] Figure 5 A flowchart illustrating an air conditioning system control method provided in this application;

[0022] Figure 6 A schematic diagram of a model training structure provided in this application;

[0023] Figure 7 A schematic diagram of a model training process provided for this application;

[0024] Figure 8 A flowchart illustrating an air conditioning system control method provided in this application;

[0025] Figure 9 A flowchart illustrating another air conditioning system control method provided in this application;

[0026] Figure 10 This is a schematic diagram of the structure of a cloud server provided in this application. Detailed Implementation

[0027] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] In the description of this invention, it should be understood that "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" and "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0029] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. Furthermore, when describing pipelines or channels, the terms "connection" and "linking" used in this application have the meaning of establishing electrical conductivity. The specific meaning needs to be understood in conjunction with the context.

[0030] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0031] An Recurrent Neural Network (RNN) is an artificial neural network in which nodes are connected in a directed loop. The internal state of this network can exhibit dynamic temporal behavior. RNNs typically take sequential data as input and, through their internal structural design, can effectively capture the relationships between sequences, generally outputting sequences in their own form.

[0032] like Figure 1 The diagram illustrates a general overview of a traditional RNN model. An example with four layers is provided, but this application does not limit the number of layers in an RNN model. This traditional RNN model can be understood as a time-series RNN model.

[0033] Depend on Figure 1 It can be seen that each layer of an RNN model has a fixed input matrix x. i and transition matrix h i Here, i can take the value of an integer such as 1, 2, 3, or 4. For example, the first-layer RNN model includes an input matrix x1 and a transition matrix h1, where h0 represents the transition matrix of the previous layer of the first-layer RNN model, and y iThis represents the output of each layer of the RNN model. It can be seen that the output of the previous layer of the RNN model can serve as part of the input to the next layer, influencing its output; this output is what is referred to as sequence information. In this application, the sequence information can, for example, be a user-defined pattern.

[0034] Wherein, the input matrix x i Used to input the temperature and humidity of the room where the indoor unit is located, as well as the airflow of the indoor unit.

[0035] Combination Figure 1 The RNN model shown is Figure 2 This diagram illustrates the input and output of each layer in a traditional RNN model.

[0036] Depend on Figure 2 It can be seen that the input of each layer in the RNN model includes the input matrix x. i+1 and transition matrix h i Input matrix x i+1 With input matrix x i+1 The matrix parameter W xi+1 The first product obtained by multiplication, and the transition matrix h i With the transition matrix h i The matrix parameter W hi The output h of each layer is obtained by adding the second product obtained from the multiplication. i+1 This is used as the input to the next layer of the RNN model. For example, when i = 0, the result of adding the input matrix x1 to the transition matrix h0 is input to the transition matrix h2. Where x... i+1 h i and h i+1 It is a vector. Through h i+1 A definite value y is obtained by transformation through the activation function. i+1 .

[0037] based on Figure 2 According to the introduction, machine learning algorithms (RNN models) are now widely used in the air conditioning field for preset user settings.

[0038] As users' demands for indoor environmental comfort continue to increase, air conditioners, as a common indoor temperature regulation device, are becoming increasingly important in intelligent air conditioning control systems for accurately predicting users' settings in order to achieve a comfortable indoor environment.

[0039] However, when traditional RNN models are applied to the air conditioning field, they cannot accurately predict user-defined modes by inputting environmental parameters and indoor unit operating parameters into the input matrix x4 of the last layer of the RNN model. This is due to the inherent limitations of traditional RNN models.

[0040] In view of this, this application obtains an optimized RNN model by introducing a context-aware input matrix and a context-aware transition matrix into the traditional RNN model. The context-aware input matrix is ​​obtained through the interaction of input context information and the input matrix. The context-aware transition matrix is ​​obtained through the interaction of transition context and the transition matrix. The output of the optimized RNN model varies with different input and transition context information, thereby making the predicted user-defined patterns more accurate. The input context information can be understood as the weekday, the ten-day period of the month, and the weather type when the user-defined pattern is set. The transition context can represent the correlation between the previous predicted user-defined pattern and the current predicted user-defined pattern.

[0041] like Figure 3 The diagram shown is a schematic representation of each layer of an optimized RNN model provided in this application. Figure 3 It can be seen that the optimized RNN model in Figure 2 Based on this, a context-aware input matrix and a context-aware transition matrix are introduced.

[0042] Wherein, the input matrix x i+1 First with matrix parameter W xi+1 Perform multiplication, then matrix multiplication with the input context matrix to obtain the context-aware input matrix. The transition matrix h... i First with matrix parameter W hi Perform multiplication, then matrix multiplication with the transition context matrix to obtain the context-aware transition matrix. Finally, add the context-aware input matrix to the context-aware transition matrix to obtain the output, which is then input into the transition matrix h. i+1 The input matrix x in the optimized RNN model i+1 The matrix parameter W xi+1 and transition matrix h i The matrix parameter W hi The prediction of user-defined patterns varies depending on the input context and the transition context. Embodiments of this application will be described below.

[0043] like Figure 4The diagram shown is an architectural schematic of an air conditioning system 400 provided in this application. The air conditioning system 400 includes a cloud server 401 and a multi-split central air conditioning unit 402. The multi-split central air conditioning unit 402 includes a central controller 4020, an outdoor unit 4021, and multiple indoor units 4022. The cloud server 401 and the central controller 4020 are coupled together. The outdoor unit 4021 and each of the multiple indoor units 4022 are coupled via a bus. The central controller 4020 is coupled to the bus on which the outdoor unit 4021 and each of the multiple indoor units 4022 are coupled.

[0044] Cloud server 401 is used to store historical environmental data (e.g., temperature and humidity of the room where indoor unit 4022 is located), historical operating parameters (components of indoor unit), historical input context information (time information and weather parameters), historical transfer context information (time stamps), and historical user setting modes. The historical input context information is obtained by cloud server 401 from the weather system of the relevant organization.

[0045] In this embodiment, the optimized RNN model is deployed on a cloud server 401 by staff. The cloud server 401 uses the optimized RNN model to predict the user-defined mode and sends the predicted user-defined mode to a central controller 4020. The central controller 4020 controls the indoor unit 4022 to adjust the user-defined mode according to the predicted user-defined mode.

[0046] A multi-split central air conditioning system 102 refers to a system where one outdoor unit 4021 is connected to two or more indoor units 4022 via piping, commonly known as a "one-to-many" system. It is a single-stage refrigerant air conditioning system 100 that uses air-cooled heat exchange on the outdoor side and direct evaporation heat exchange on the indoor side. The multi-split central air conditioning system 102 is a high-efficiency refrigerant air conditioning system 100 that meets indoor heating and cooling load requirements by controlling the refrigerant circulation volume of the compressor and the refrigerant flow rate into the indoor heat exchanger. The multi-split central air conditioning system 102 requires a variable frequency compressor, multi-stage compressor, unloading compressor, or a combination of multiple compressors to achieve compressor capacity control. An electronic expansion valve or other auxiliary circuit is required in the refrigeration system to regulate the refrigerant flow rate into the indoor units 4022, and the capacity of the heat exchangers is adjusted by controlling the fan speed of the indoor and outdoor heat exchangers. The multi-split central air conditioning system 102 can operate in heating or cooling mode to regulate the indoor ambient temperature.

[0047] based on Figure 4 In addition to the description of the air conditioning system 400, this application also provides a control method for the air conditioning system 400.

[0048] like Figure 5The diagram shown is a flowchart of an air conditioning system 400 control method provided in this application, which includes the following steps.

[0049] 501. The cloud server 401 obtains the environmental parameters, operating parameters, input context, and transfer context of the multi-split central air conditioning system. The input context includes the time information and weather parameters when the user-defined mode is set for the current forecast. The transfer context includes the time difference between the current user-defined mode and the previous user-defined mode.

[0050] The transition context here can also be understood as including the timestamp of the predicted user-defined pattern.

[0051] In some embodiments, environmental parameters include at least one of the temperature and humidity of the room where the indoor unit is located during the current user-defined mode prediction. Operating parameters include the airflow of the indoor unit during the current user-defined mode prediction. Time information for the current user-defined mode prediction includes at least one weekday and ten-day period of the month during the current user-defined mode prediction. Weather parameters include the weather type during the current user-defined mode prediction. User-defined modes include heating mode, cooling mode, and dehumidification mode.

[0052] The weather type here can be, for example, sunny, cloudy, or overcast, etc., and this application does not limit it.

[0053] For example, when the cloud server 401 makes a user mode prediction, the indoor unit 4022 collects the temperature of the room it is in using a temperature sensor, the humidity of the room it is in using a humidity sensor, and the airflow of the room it is in using an airflow sensor. The indoor unit 4022 sends the collected temperature, humidity, and airflow data to the cloud server. The cloud server obtains the time information and weather type for this user mode prediction from the weather system of the relevant organization. The cloud server 401 calculates the time difference between the current user mode prediction and the previous user mode prediction. For example, if the user mode prediction is made at 14:00, and the previous user mode prediction was at 12:00, the time difference is 2 hours.

[0054] 502. The cloud server 401 determines the user setting mode for this prediction based on environmental parameters, operating parameters, input context, transition context, and the first neural network model.

[0055] The first neural network model here can be understood as... Figure 3 The optimized RNN model.

[0056] In some embodiments, the first neural network model includes multiple sets of correspondences, each set of correspondences including a set of historical environmental parameters, historical operating parameters, historical input context, historical transition context, and a historical user-defined mode.

[0057] For example, the indoor unit 4022 is located in a room with a temperature of 26 degrees Celsius, a humidity of 60% RH, an indoor unit 4022 with a low-speed fan, a time of Wednesday, mid-week, and a sunny weather type, corresponding to the heating mode.

[0058] Among them, historical environmental parameters, historical operating parameters, historical input contexts, historical transfer contexts, and historical user setting modes are stored in the database of cloud server 401.

[0059] In some embodiments, the cloud server 401 determines the user setting mode for this prediction based on the environmental parameters, operating parameters, input context, transfer context, and multiple sets of correspondence obtained when predicting the user setting mode locally.

[0060] The environmental parameters, runtime parameters, input context, and transfer context have already been introduced above and will not be repeated here.

[0061] For example, let's take an optimized RNN model with 4 layers as an example, where i is an integer less than or equal to 4. The cloud server 401 inputs the acquired environment parameters, runtime parameters, and input context into the context-aware input matrix of the last layer of the optimized RNN model. The environment parameters and runtime parameters are input into the input matrix x4 (i.e., i = 3), and the input context is input into the input context matrix. The acquired transition context is then input into the transition context matrix of the last layer of the optimized RNN model. The context-aware input matrix is ​​then added to the context-aware transition matrix, and the result, h4, is transformed by an activation function to obtain a specific value y4, which is the user-defined pattern predicted in this case.

[0062] Therefore, this application determines the user setting mode for this prediction by acquiring the environmental parameters, operating parameters, input context, and transition context of the multi-split central air conditioning unit 402, as well as the first neural network model. The input context includes the time information and weather parameters at the time of this user setting mode prediction, and the transition context includes the time difference between the current user setting mode prediction and the previous user setting mode prediction. Compared to existing technologies that predict user setting modes by inputting indoor environmental parameters into the input matrix of the RNN model, this application adds input context and transition context to predict user setting modes, making the prediction results more accurate.

[0063] In some embodiments, when the time difference between the current predicted user setting mode and the previous predicted user setting mode is less than or equal to a preset threshold, the user setting mode for the current prediction is determined based on the previous predicted user setting mode, environmental parameters, operating parameters, input context, transition context, and the first neural network model obtained during the current predicted user setting mode. When the time difference between the current predicted user setting mode and the previous predicted user setting mode is greater than the preset threshold, the user setting mode for the current prediction is determined based on the environmental parameters, operating parameters, input context, transition context, and the first neural network model obtained during the current predicted user setting mode.

[0064] In other words, when cloud server 401 performs user pattern prediction, if the time difference between the current predicted user setting pattern and the previous predicted user setting pattern is small, the optimized RNN model will also refer to the previous predicted user setting pattern when making the current prediction. If the time difference between the current predicted user setting pattern and the previous predicted user setting pattern is large, the optimized RNN model will determine the current predicted user setting pattern based on the environmental parameters, running parameters, input context, transition context, and multiple sets of correspondences obtained when predicting user setting patterns locally.

[0065] based on Figure 5 In step 502, before the cloud server 401 determines the user setting mode for this prediction based on environmental parameters, running parameters, input context, transition context and the first neural network model, the cloud server 401 can also train the first neural network model based on multiple sets of training data.

[0066] like Figure 6 The diagram shown illustrates a model training structure provided in this application. This structure includes a database 601, a model training module 602, and a model output module 603.

[0067] Database 601 is used to store multiple sets of training data for model training 602, wherein each set of training data includes historical environment parameters, historical running parameters, historical input context, historical transition context, and historical user setting mode.

[0068] Model training 602 is used to train the model based on multiple sets of training data stored in database 601 to obtain the first neural network model.

[0069] Model output 603 is used to output the first neural network model obtained from training 602.

[0070] Based on Figure 6 An introduction to the structure diagram of model training, such as... Figure 7 The diagram shown is a flowchart of a model training process provided in this application. The model training includes the following steps.

[0071] 701. Cloud server 401 acquires multiple sets of training data. Each set of training data includes sampling data of multiple variables at different set temperatures and historical user setting modes.

[0072] In some embodiments, the variables include historical environment parameters, historical operating parameters, historical input context, and historical transfer context.

[0073] For example, cloud server 401 obtains multiple sets of historical environmental parameters, historical operating parameters, historical input contexts, historical transfer contexts, and historical user setting modes from its database 601. Historical environmental parameters can be understood as parameters such as temperature and humidity of the room where indoor unit 4022 is located, obtained over a period of time. Historical operating parameters can be understood as the airflow of indoor unit 4022, obtained over a period of time. Historical input contexts can be understood as the weekday, the ten-day period of the month, and the corresponding weather type for the historical user setting modes obtained over a period of time. Historical transfer contexts can be understood as the time difference between two historical predicted user setting modes obtained over a period of time. Historical user setting modes can be understood as the user-set modes obtained over a period of time, such as heating or cooling modes.

[0074] 702. The cloud server 401 uses multiple sets of training data as training data for the recurrent neural network (RNN) model to obtain the first neural network model.

[0075] The recurrent neural network (RNN) model here can be understood as... Figure 1 or Figure 2 The RNN model in [the context of the text].

[0076] For example, cloud server 401 uses 70% of the acquired training data from multiple sets of training data for model training. For instance, if there are 100 sets of training data, cloud server 401 selects 70 sets for model training. Cloud server 401 continuously samples these 70 sets of training data, inputting historical environment parameters, historical running parameters, historical input context, and historical transition context into a multi-layered continuous RNN model for training. The historical environment parameters, historical running parameters, and historical input context form a context-aware input matrix, and the historical transition context forms a context-aware transition matrix. Cloud server 401 adds the context-aware input matrix and the context-aware transition matrix, and then inputs the result into... Figure 1The first neural network model is obtained by training a multi-layer RNN model.

[0077] Then, cloud server 401 calculates the loss function value based on the predicted user setting mode value (e.g., heating is 0, cooling is 1, etc., and the data type of the setting mode is integer (INT)) and the historical user setting mode values. Based on the calculated loss function value, the matrix parameters W of the RNN model are adjusted using the stochastic gradient descent (SGD) algorithm. xi+1 and matrix parameters W hi Update the RNN model until it converges.

[0078] The following section will introduce this application in conjunction with specific application scenarios.

[0079] like Figure 8 The diagram shown is a flowchart illustrating a control method for an air conditioning system 400 provided in this application.

[0080] Data is collected from cloud server 401 and 801.

[0081] For example, cloud server 401 collects historical environmental parameters, historical operating parameters, historical input context, historical transition context and historical user setting mode from cloud server 401 database 601 for model training and testing to obtain the first neural network model.

[0082] 802. The cloud server 401 performs data preprocessing on the collected data.

[0083] For example, the cloud server 401 cleans and processes the parameters collected in step 801, such as handling missing values, outliers, and normalizing the collected parameters. For instance, let's take the indoor unit 4022 collecting the temperature of the room it's in via a temperature sensor as an example. As shown in Table 1, one set of temperature data is missing.

[0084] Table 1

[0085] 1 26 2 3 28

[0086] As shown in Table 1, the temperature data of the second set of training data (ID2) is missing among the three sets of training data collected.

[0087] The cloud server 401 first uses the average value μ in ID1-ID3 to fill ID2, and we know that μ=(26). After filling, it can be shown in Table 2.

[0088] Table 2

[0089] 1 26 2 28 3 28

[0090] Next, cloud server 401 normalizes the data in Table 2 using the formula Y = (X - μ) / σ. Here, σ represents the standard deviation, Y represents the temperature data obtained after standard deviation normalization, and X represents the temperature data in Table 2. The data obtained from standard deviation normalization in Table 2 is shown in Table 3.

[0091] Table 3

[0092] 1 -1.225 2 0 3 1.225

[0093] Model training is performed on cloud server 401 and 803.

[0094] For details on the model training process on cloud server 401, please refer to steps 701-702, which will not be repeated here.

[0095] Model testing is performed on cloud server 401 and 804.

[0096] For example, the cloud server 401 inputs the acquired multiple sets of historical environment parameters, historical running parameters, historical input context, historical transition context, and the remaining 30% of the parameters in the historical user setting mode into the converged RNN model for testing, and evaluates the prediction of the RNN model.

[0097] 805. Cloud server 401 performs user pattern prediction.

[0098] For user pattern prediction on cloud server 401, please refer to steps 501-502, which will not be repeated here.

[0099] 806. Central controller 4020 adjusts the air conditioner user setting mode.

[0100] For example, the cloud server 401 sends the predicted user setting mode to the central controller 4020, and the central controller 4020 automatically adjusts the setting mode of the air conditioning system 400 according to the predicted user setting mode.

[0101] Alternatively, when a user sets a user-defined mode through the central controller 4020, if the user-defined mode predicted by the cloud server 401 differs significantly from the actual user-defined mode, the central controller 4020 will automatically adjust the air conditioning system 400 to the actual user-defined mode.

[0102] based on Figure 8 This application provides a control method for an air conditioning system 400, and also provides an optimization scheme.

[0103] like Figure 9The diagram shown is a flowchart of another air conditioning system 400 control method provided in this application, which includes the following process.

[0104] 901 and cloud server 401 collect data and import multimodal data.

[0105] The multimodal data here can be, for example, the number of people indoors and the carbon dioxide concentration, etc., which are not limited in this application.

[0106] For example, in Figure 8 Based on step 801, the cloud server 401 also collects the number of people in the room where the indoor unit 4022 is located and the carbon dioxide concentration.

[0107] It is evident that, since the number of people indoors and the carbon dioxide concentration are closely related to user-defined patterns, the diversity and accuracy of the RNN model's output can be improved, thereby enhancing the prediction results of the RNN model. For example, when predicting user-defined patterns in an office, inputting the number of people indoors and the carbon dioxide concentration into the RNN model can better reflect the actual situation and improve the prediction accuracy.

[0108] 902. Cloud server 401 performs data preprocessing on the collected data.

[0109] For information on how to preprocess the collected data, please refer to [reference needed]. Figure 8 Step 802 will not be repeated here.

[0110] 903. Cloud server 401 introduces an attention mechanism for model training.

[0111] For example, in Figure 8 Building upon step 803, an attention mechanism is introduced. This mechanism enables RNN models to better capture long-range mutual information. For distant information, RNNs need to compute sequentially, requiring several time steps to accumulate information before connecting the two points. Furthermore, the greater the distance, the lower the probability of effective capture, potentially resulting in the loss of some long-range information. Introducing the attention mechanism allows RNN models to more easily capture long-range mutual information, helping the model converge faster.

[0112] Furthermore, during training, the cloud server 401 also fine-tunes the parameters of the RNN model to improve prediction accuracy. For example, it selects appropriate hyperparameters for the RNN model, such as the learning rate, number of epochs, number of hidden layers, number of hidden layer units, activation function, and batch size. Following the data processing procedures during model training, it tries different combinations of hyperparameters multiple times, selecting the best set based on the predicted loss function value.

[0113] 904. Cloud server 401 introduces evaluation indicators for model testing.

[0114] The evaluation metrics here can be, for example, the root-mean-square error (RMSE) and the mean absolute error (MAE), which are not limited in this application.

[0115] For example, in Figure 8 Based on step 804, the root-mean-square error (RMSE) and mean absolute error (MAE) metrics were used to test the model.

[0116] 905. Cloud server 401 performs user pattern prediction.

[0117] For instructions on how to perform user pattern prediction, please refer to step 805; it will not be repeated here.

[0118] 906. Central controller 4020 adjusts the air conditioner user setting mode.

[0119] For instructions on how to adjust the air conditioner user settings mode, please refer to step 806. It will not be repeated here.

[0120] It should be noted that the central controller 4020 here can also be replaced with a human-computer interaction interface or other smart devices.

[0121] Based on the above description of the control method for the air conditioning system 400, this application also provides a cloud server.

[0122] like Figure 10The diagram shown is a structural schematic of a cloud server 1000 provided in this application. The cloud server 1000 includes a parameter acquisition unit 1001, a mode determination unit 1002, and a model training unit 1003. It should be noted that the cloud server 1000 here can also be a cloud server 401.

[0123] The parameter acquisition unit 1001 is configured to acquire the environmental parameters, operating parameters, input context, and transfer context of the multi-split central air conditioning system. The input context includes the time information and weather parameters of the current user-defined mode, and the transfer context includes the time difference between the current user-defined mode and the previous user-defined mode.

[0124] The pattern determination unit 1002 is configured to determine the user-defined pattern for this prediction based on environmental parameters, operating parameters, input context, transition context, and the first neural network model.

[0125] In some embodiments, the pattern determination unit 1002 is further configured to determine the user setting mode for the current prediction based on the previous user setting mode, environmental parameters, operating parameters, input context, transition context, and the first neural network model when the time difference between the current predicted user setting mode and the previous predicted user setting mode is less than or equal to a preset threshold. When the time difference between the current predicted user setting mode and the previous predicted user setting mode is greater than the preset threshold, the user setting mode for the current prediction is determined based on the environmental parameters, operating parameters, input context, transition context, and the first neural network model obtained when the current predicted user setting mode is used.

[0126] In some embodiments, environmental parameters include at least one of the temperature and humidity of the room where the indoor unit is located during the current user-defined mode prediction. Operating parameters include the airflow of the indoor unit during the current user-defined mode prediction. Time information for the current user-defined mode prediction includes at least one weekday and ten-day period of the month during the current user-defined mode prediction. Weather parameters include the weather type during the current user-defined mode prediction. User-defined modes include heating mode, cooling mode, and dehumidification mode.

[0127] In some embodiments, the first neural network model includes multiple sets of correspondences, each set including a set of historical environmental parameters, historical operating parameters, historical input context, historical transition context, and a historical user-defined pattern correspondence. The pattern determination unit is further configured to determine the user-defined pattern for the current prediction based on the environmental parameters, operating parameters, input context, transition context, and multiple sets of correspondences obtained when predicting the user-defined pattern locally.

[0128] In some embodiments, the parameter acquisition unit 1001 is further configured to acquire multiple sets of training data, each set of training data including sampled data of multiple variables at different set temperatures and historical user-defined modes. The multiple variables include historical environmental parameters, historical operating parameters, historical input context, and historical transfer context.

[0129] The model training unit 1003 is configured to train multiple sets of training data as training data for a recurrent neural network (RNN) model to obtain a first neural network model.

[0130] Therefore, this application determines the user setting mode for this prediction by acquiring the environmental parameters, operating parameters, input context, and transition context of the multi-split central air conditioning unit 402, as well as the first neural network model. The input context includes the time information and weather parameters at the time of this user setting mode prediction, and the transition context includes the time difference between the current user setting mode prediction and the previous user setting mode prediction. Compared to existing technologies that predict user setting modes by inputting indoor environmental parameters into the input matrix of the RNN model, this application adds input context and transition context to predict user setting modes, making the prediction results more accurate.

[0131] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, the disclosure, and the appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0132] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

[0133] The above figures are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A cloud server, characterized in that, The cloud server communicates with the central controller of the multi-split central air conditioning system, which includes multiple indoor units. The cloud server includes: The parameter acquisition unit is configured to acquire the environmental parameters, operating parameters, input context, and transfer context of the multi-split central air conditioning system. The input context includes the time information and weather parameters when the user setting mode is predicted this time. The transfer context includes the time difference between the user setting mode prediction this time and the previous user setting mode prediction. The pattern determination unit is configured to determine the user-defined pattern for this prediction based on the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model. The mode determination unit is further configured to determine the user setting mode predicted this time based on the previous user setting mode, the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model when the time difference between the current predicted user setting mode and the previous predicted user setting mode is less than or equal to a preset threshold. When the time difference between the current predicted user setting mode and the previous predicted user setting mode is greater than the preset threshold, the predicted user setting mode is determined based on the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model obtained during the current prediction of the user setting mode.

2. The cloud server according to claim 1, characterized in that, The environmental parameters include at least one of the temperature and humidity of the room where the indoor unit is located when the user sets the mode for this prediction. The operating parameters include the air volume of the indoor unit when the user sets the mode for this prediction; The time information for the user setting mode in this prediction includes at least one weekday and one ten-day period of the month in which the user setting mode is predicted. The weather parameters include the weather type when the user set the mode for this forecast. The user-defined modes include heating mode, cooling mode, and dehumidification mode.

3. The cloud server according to claim 1, characterized in that, The first neural network model includes multiple sets of correspondences. Each set of correspondences includes a set of historical environmental parameters, historical operating parameters, historical input context, historical transition context, and a historical user-defined mode. The mode determination unit is further configured to determine the user setting mode for this prediction based on the environmental parameters, the operating parameters, the input context, the transition context, and the multiple sets of correspondences obtained when predicting the user setting mode locally.

4. The cloud server according to claim 1, characterized in that, The parameter acquisition unit is also configured to acquire multiple sets of training data, each set of training data including sampled data of multiple variables at different set temperatures and historical user setting modes; The variables include historical environment parameters, historical operating parameters, historical input context, and historical transfer context; The cloud server further includes a model training unit, configured to use the multiple sets of training data as training data for a recurrent neural network (RNN) model to obtain the first neural network model.

5. A control method for an air conditioning system, characterized in that, The method is applied to a cloud server, which communicates with a central controller in a multi-split central air conditioning system, the multi-split central air conditioning system comprising multiple indoor units, and includes: The environmental parameters, operating parameters, input context, and transfer context of the multi-split central air conditioning system are obtained. The input context includes the time information and weather parameters when the user-set mode is predicted this time. The transfer context includes the time difference between the current user-set mode prediction and the previous user-set mode prediction. The user setting mode for this prediction is determined based on the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model. The step of determining the user setting mode for this prediction based on the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model includes: When the time difference between the current predicted user setting mode and the previous predicted user setting mode is less than or equal to a preset threshold, the current predicted user setting mode is determined based on the previous predicted user setting mode, the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model. When the time difference between the current predicted user setting mode and the previous predicted user setting mode is greater than the preset threshold, the predicted user setting mode is determined based on the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model obtained during the current prediction of the user setting mode.

6. The method according to claim 5, characterized in that, The environmental parameters include at least one of the temperature and humidity of the room where the indoor unit is located when the user sets the mode for this prediction. The operating parameters include the air volume of the indoor unit when the user sets the mode for this prediction; The time information for the user setting mode in this prediction includes at least one weekday and one ten-day period of the month in which the user setting mode is predicted. The weather parameters include the weather type when the user set the mode for this forecast. The user-defined modes include heating mode, cooling mode, and dehumidification mode.

7. The method according to claim 5, characterized in that, The first neural network model includes multiple sets of correspondences. Each set of correspondences includes a set of historical environmental parameters, historical operating parameters, historical input context, historical transition context, and a historical user-defined mode. The step of determining the user setting mode for this prediction based on the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model includes: The user setting mode for this prediction is determined based on the environmental parameters, operating parameters, input context, transfer context, and multiple sets of corresponding relationships obtained when predicting the user setting mode locally.

8. The method according to claim 5, characterized in that, Before determining the user-defined pattern for this prediction based on the environmental parameters, the operating parameters, the input context, the transition context, and the first neural network model, the method further includes: Acquire multiple sets of training data, each set including sampled data of multiple variables at different set temperatures and historical user setting modes; The variables include historical environment parameters, historical operating parameters, historical input context, and historical transfer context; The multiple sets of training data are used as training data for a recurrent neural network (RNN) model to obtain the first neural network model.

Images