Air conditioning load prediction method and device, electronic equipment, air conditioner and storage medium

Abstract

The application provides an air conditioner load prediction method and device, electronic equipment, an air conditioner and a storage medium, and relates to the field of air conditioners. The method comprises the following steps: acquiring actual operation environment data of an air conditioner; wherein the actual operation environment data comprises a space occupancy parameter and a door and window operation parameter, the space occupancy parameter is used for indicating information that an object exists in a space where the air conditioner is located, and the door and window operation parameter is used for indicating information that a door and window in the space is opened or closed; encoding the actual operation environment data to obtain a first feature representation; and performing load prediction according to the first feature representation to obtain a predicted load of the air conditioner. Therefore, by capturing the main disturbance factors in the real operation scene of the air conditioner, including the space occupancy parameter reflecting the heat source distribution of objects such as personnel, equipment and pets in the space where the air conditioner is located, and the door and window operation parameter representing the opening and closing behavior or the ventilation intensity of the door and window, and performing load prediction after encoding the above parameters, the accuracy of air conditioner load prediction can be significantly improved.

Description

Technical Field

[0001] This application relates to the field of air conditioning technology, and in particular to an air conditioning load prediction method, device, electronic equipment, air conditioner, and storage medium. Background Technology

[0002] In related technologies, physical simulation prediction methods are mainly used to simulate and predict air conditioning load. However, physical simulation prediction methods rely on accurate modeling of physical parameters, but these physical parameters are often difficult to obtain accurately, resulting in low accuracy of air conditioning load prediction. Summary of the Invention

[0003] This application proposes an air conditioning load prediction method, apparatus, electronic device, air conditioner, and storage medium to at least partially solve one of the technical problems in the related art.

[0004] One embodiment of this application proposes an air conditioning load forecasting method, including:

[0005] Acquire actual operating environment data of the air conditioner; wherein, the actual operating environment data includes space occupancy parameters and door and window operation parameters, the space occupancy parameters are used to indicate the information of objects existing in the space where the air conditioner is located, and the door and window operation parameters are used to indicate the information of opening and closing of doors and windows in the space; The actual operating environment data is encoded to obtain a first feature representation; Load prediction is performed based on the first feature representation to obtain the predicted load of the air conditioner.

[0006] Another embodiment of this application proposes an air conditioning load prediction device, comprising: The first acquisition module is used to acquire actual operating environment data of the air conditioner; wherein, the actual operating environment data includes space occupancy parameters and door and window operation parameters, the space occupancy parameters are used to indicate the information of objects existing in the space where the air conditioner is located, and the door and window operation parameters are used to indicate the information of opening and closing of doors and windows in the space; The first encoding module is used to encode the actual operating environment data to obtain a first feature representation; Load prediction is performed based on the first feature representation to obtain the predicted load of the air conditioner.

[0007] In another aspect of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the air conditioning load forecasting method as described in the foregoing aspect.

[0008] Another embodiment of this application provides an air conditioner, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to: implement the air conditioner load prediction method as described in the foregoing aspect.

[0009] Another aspect of this application proposes a non-transitory computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the air conditioning load forecasting method as described in the foregoing aspect.

[0010] In another aspect, this application provides a computer program product having a computer program stored thereon, which, when executed by a processor, implements the air conditioning load prediction method as described in the foregoing aspect.

[0011] The air conditioning load prediction method, device, electronic equipment, air conditioner, and storage medium proposed in this application collect key dynamic information from the actual operating environment of the air conditioner—including space occupancy parameters reflecting the distribution of heat sources such as people, equipment, and pets in the space where the air conditioner is located, and door and window operation parameters characterizing the ventilation intensity—and encode the above parameters to generate a first feature representation. Then, load prediction is performed based on the first feature representation. This can effectively avoid the dependence of physical modeling on complex and difficult-to-obtain physical parameters. In other words, in this application, the main disturbance factors in the real operating scenario of the air conditioner are captured in a data-driven manner, which significantly improves the accuracy and real-time adaptability of air conditioning load prediction.

[0012] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0013] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A schematic flowchart of an air conditioning load forecasting method provided for an exemplary embodiment of this application; Figure 2 A schematic flowchart of another air conditioning load forecasting method provided for an exemplary embodiment of this application; Figure 3 A schematic flowchart illustrating a training method for a sparse encoder provided for an exemplary embodiment of this application; Figure 4 A schematic flowchart of another air conditioning load forecasting method provided for an exemplary embodiment of this application; Figure 5 A schematic flowchart of another air conditioning load forecasting method provided for an exemplary embodiment of this application; Figure 6 A flowchart illustrating an air conditioning load forecasting method based on data fusion and decision tree algorithm provided for an exemplary embodiment of this application; Figure 7 A schematic diagram of a data fusion mechanism provided for an exemplary embodiment of this application; Figure 8 A schematic diagram illustrating the training principle of a sparse encoder provided for an exemplary embodiment of this application; Figure 9 A schematic diagram of the structure of an air conditioning load prediction device provided for an exemplary embodiment of this application; Figure 10 This is a schematic diagram of the structure of an electronic device provided for an exemplary embodiment of this application. Detailed Implementation

[0014] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0015] It should be noted that the acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations and do not violate public order and good morals.

[0016] It should also be noted that all data processed in this application is data that has been explicitly authorized by the user or relevant parties, and has been de-identified or anonymized before collection and use, and does not contain any personally identifiable information or user privacy content; all data is used only for the purpose of air conditioning load prediction, ensuring that data security and user privacy rights are fully protected while achieving technical effects.

[0017] In related technologies, air conditioning load forecasting methods are mainly divided into two categories: physical simulation forecasting methods and data-driven model forecasting methods. Physical simulation forecasting methods rely on accurate modeling of physical parameters such as building thermodynamic characteristics and equipment performance. However, these physical parameters are often difficult to obtain accurately, resulting in limited forecasting accuracy. While data-driven model forecasting methods can learn load change patterns from historical operating data, they generally lack sufficient modeling of dynamic factors in the space where the air conditioning is located. Furthermore, the dimensions of the collected monitoring data are limited, and the information coverage is incomplete, which restricts the accuracy and generalization ability of the forecasting results.

[0018] Therefore, in view of at least one of the problems existing in the above-mentioned related technologies, this application proposes an air conditioning load prediction method, apparatus, electronic device, air conditioner and storage medium.

[0019] The following description, with reference to the accompanying drawings, describes an air conditioning load prediction method, apparatus, electronic device, air conditioner, and storage medium according to embodiments of this application. Before specifically describing the embodiments of this application, for ease of understanding, commonly used technical terms are first introduced: Decision tree: A tree-based supervised learning model specifically designed for classification tasks. It partitions the feature space using a recursive binary search strategy, constructing a top-down tree structure: each internal node corresponds to a conditional test for a feature attribute (e.g., "outdoor temperature > 25℃"), each branch represents the Boolean output (yes / no) of the conditional test, and leaf nodes store discrete class labels (e.g., "high load" / "low load"). Through this hierarchical partitioning mechanism, the decision tree model effectively captures nonlinear decision boundaries in the data.

[0020] Regression Tree: A regression extension variant of the decision tree that achieves continuous value prediction by modifying the output layer of the leaf nodes. For example, in the task of predicting air conditioning load, the regression tree model can use a recursive bisection method to divide the feature space, with each leaf node representing a feature subspace, and outputting the statistics of the air conditioning load of all samples in that feature subspace (e.g., a mean of 12.5 kW).

[0021] Figure 1 This is a flowchart illustrating an air conditioning load forecasting method provided for an exemplary embodiment of this application.

[0022] It should be noted that the air conditioning load forecasting method of this application embodiment can be applied to an air conditioning load forecasting device. In some possible embodiments, the air conditioning load forecasting device can be configured in an electronic device or an air conditioner so that the electronic device or air conditioner can perform the air conditioning load forecasting function. In addition, in some possible embodiments, the air conditioning load forecasting device can also be software in an electronic device or an air conditioner.

[0023] The electronic devices include, but are not limited to, terminals and servers that communicate with the air conditioner. The terminals can be automobiles with communication capabilities, smart cars, mobile phones, wearable devices, tablets, computers with wireless transceiver capabilities, virtual reality (VR) terminals, augmented reality (AR) terminals, wireless terminals in industrial control, wireless terminals in assisted driving, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, and so on. The embodiments of this application do not limit the specific technology or device form used in the terminals.

[0024] Among them, assisted driving refers to the technology that uses sensors, algorithms and artificial intelligence to perceive the environment, make decisions, plan and execute control commands of the vehicle in order to assist the driver to drive more safely and efficiently.

[0025] For ease of explanation, the following description will use air conditioning as the subject of this air conditioning load forecasting method.

[0026] like Figure 1 As shown, the air conditioning load forecasting method may include the following steps S101 to S103: Step S101: Obtain the actual operating environment data of the air conditioner; wherein, the actual operating environment data includes space occupancy parameters and door and window operation parameters. The space occupancy parameters are used to indicate the information of objects existing in the space where the air conditioner is located, and the door and window operation parameters are used to indicate the information of opening and closing of doors and windows in the space.

[0027] The actual operating environment data can be obtained through relevant sensors in the space where the air conditioner is located.

[0028] The objects include, but are not limited to: static objects (such as heat dissipation equipment) and dynamic objects (such as people, pets, etc.).

[0029] The actual operating environment data includes, but is not limited to, space occupancy parameters and door and window operation parameters. Among them, space occupancy parameters can be used to quantify the distribution characteristics or object density of objects in the space where the air conditioner is located, reflecting the changing pattern of heat source intensity, including but not limited to: the number of people, the number of equipment, and the number of pets indoors.

[0030] Among them, the door and window operation parameters are used to characterize the dynamic opening and closing behavior of doors and windows or the intensity of ventilation, so as to quantify the impact of air infiltration on the characteristics of air conditioning ventilation, including but not limited to: the number of times doors and windows are opened and closed, the duration of opening and closing, and the opening and closing angle.

[0031] In other words, this application takes into account that the different distribution of heat sources in the space where the air conditioner is located will have different effects on the air conditioning load, and that different ventilation volumes will also have different effects on the air conditioning load. Therefore, this application can simultaneously monitor the distribution of heat sources in the space where the air conditioner is located, the number of times doors and windows are opened and closed, the duration of opening and closing, and the opening and closing angle, and use this as the actual operating environment data of the air conditioner to improve the accuracy and adaptability of load prediction.

[0032] In any embodiment of this application, the actual operating environment data may further include: outdoor meteorological parameters and / or indoor environmental parameters; wherein, the outdoor meteorological parameters include, but are not limited to: outdoor temperature T1, outdoor humidity RH1, solar radiation intensity DNI1, etc.; the indoor environmental parameters include, but are not limited to: indoor temperature T2, indoor humidity RH2, indoor wind speed WS, etc.

[0033] In summary, since outdoor meteorological parameters and indoor environmental parameters directly affect the heat transfer characteristics of building envelopes, changes in indoor heat load, and human thermal comfort needs, incorporating them into the actual operating environment data of air conditioning systems helps improve the perception and response accuracy to complex environmental changes, thereby significantly enhancing the accuracy, robustness, and engineering practicality of subsequent load forecasting.

[0034] Step S102: Encode the actual operating environment data to obtain the first feature representation.

[0035] In this embodiment of the application, an encoding algorithm (or feature extraction algorithm) can be used to encode the actual operating environment data of the air conditioner to obtain a first feature representation.

[0036] It should be noted that encoding the actual operating environment data can extract key features of load impact, reduce noise and redundant information, and not only improve the forecasting efficiency of air conditioning load, but also improve the forecasting accuracy of air conditioning load.

[0037] Step S103: Perform load prediction based on the first feature representation to obtain the predicted load of the air conditioner.

[0038] In this embodiment of the application, load prediction can be performed based on the first feature representation to obtain the predicted load of the air conditioner.

[0039] For example, deep learning models, reinforcement learning models, or large language models in the field of artificial intelligence can be used to predict the load on the first feature representation to obtain the predicted load of the air conditioner.

[0040] For example, a decision tree model or a regression tree model can be used to predict the load based on the first feature representation in order to obtain the predicted load of the air conditioner.

[0041] The air conditioning load prediction method of this application collects key dynamic information in the actual operating environment of the air conditioner, including space occupancy parameters reflecting the distribution of heat sources such as people, equipment, and pets in the space where the air conditioner is located, and door and window operation parameters characterizing the opening and closing behavior or ventilation intensity. The above parameters are encoded to generate a first feature representation, and load prediction is then performed based on the first feature representation. This method can effectively avoid the dependence of physical modeling on complex and difficult-to-obtain physical parameters. In other words, in this application, the main disturbance factors in the real operating scenario of the air conditioner are captured in a data-driven manner, which significantly improves the accuracy and real-time adaptability of air conditioning load prediction.

[0042] As one possible implementation method, Figure 2 A flowchart illustrating another air conditioning load forecasting method provided for an exemplary embodiment of this application.

[0043] It should be noted that the air conditioning load forecasting method can be executed alone, or it can be executed together with any embodiment or possible implementation in the embodiment of this application, or it can be executed together with any technical solution in the related technology. The embodiments of this application do not limit this.

[0044] like Figure 2 As shown, the air conditioning load forecasting method may include the following steps S201 to S205: Step S201: Based on the heat dissipation of objects in the space where the air conditioner is located, classify the objects to obtain at least one type of object; wherein, objects of the same type have similar heat dissipation.

[0045] Among them, similarity means that the difference in heat dissipation is less than a set threshold.

[0046] The heat dissipation of different objects can be determined according to a preset heat dissipation reference table (i.e., the lookup table method), or it can be dynamically predicted based on the physical characteristics (such as body shape, size, etc.) and behavioral characteristics (such as activity intensity) of different objects. This application embodiment does not limit this.

[0047] The heat dissipation comparison table records the heat dissipation of different objects.

[0048] In this embodiment of the application, the presence of objects in the space where the air conditioner is located can be detected by relevant sensors. If objects are present, the objects in the space where the air conditioner is located can be classified based on the heat dissipation of the detected different objects to obtain at least one type of object; wherein, objects of the same type have similar heat dissipation.

[0049] Step S202: Quantity monitoring is performed on at least one type of object to obtain space occupancy parameters.

[0050] The space occupancy parameter indicates the number of objects of different types.

[0051] In this embodiment of the application, the quantity of at least one type of object can be monitored using relevant sensors to obtain space occupancy parameters.

[0052] Step S203: Monitor at least one of the following: number of times the doors and windows in the space are opened and closed, opening and closing duration, and opening and closing angle, in order to obtain the operating parameters of the doors and windows.

[0053] Among them, the door and window operation parameters can be used to indicate at least one of the following: the number of times the doors and windows in the space where the air conditioner is located are opened and closed within a set time period, the opening and closing duration, and the opening and closing angle.

[0054] In this embodiment of the application, at least one of the following can be monitored by relevant sensors: the number of times doors and windows are opened and closed, the duration of opening and closing, and the opening and closing angle, so as to obtain the operating parameters of doors and windows.

[0055] Step S204: Encode the actual operating environment data of the air conditioner to obtain the first feature representation; wherein, the actual operating environment data includes space occupancy parameters and door and window operation parameters.

[0056] It should be noted that the explanation of step S204 can be found in the relevant description in any embodiment of this application, and will not be repeated here.

[0057] In any embodiment of this application, the actual operating environment data is obtained, for example, by acquiring sensor data collected by multiple sensors in the space where the air conditioner is located, and preprocessing the sensor data of the multiple sensors to obtain the actual operating environment data.

[0058] The preprocessing includes at least one of the following: time alignment, smoothing, normalization, and noise reduction.

[0059] Among them, various sensors include, but are not limited to: temperature sensors, humidity sensors, wind speed sensors, thermal imaging sensors, solar radiation sensors, and door magnetic sensors (door magnetic sensors can output switch status signals in real time and upload them to the air conditioning controller or intelligent central control system).

[0060] The actual operating environment data includes space occupancy parameters and door and window operation parameters. Optionally, the actual operating environment data may also include outdoor meteorological parameters and / or indoor environmental parameters.

[0061] As an example, the nearest neighbor interpolation method is used to perform time-series alignment processing on sensor data collected by multiple sensors to achieve synchronous mapping of parameters in the sensor data at any time. Whenever a sensor in the space where the air conditioner is located uploads its collected sensor data, the acquisition time corresponding to the sensor data can be used as a new reference time point. In the historical data sequence collected by other sensors, the two sampling times closest to the reference time point are found and used as upper and lower bounds, respectively. Based on the nearest neighbor interpolation method, the equivalent observation values ​​of other sensors at the reference time point are calculated according to the actual observation values ​​corresponding to the upper and lower bounds.

[0062] For example, for a reference time point t, and a historical data sequence (t0) of a certain sensor x, i x i The estimated equivalent observations are: (1) Where t is the reference time point, i.e., the time point at which the equivalent observations from other sensors need to be estimated. i These are the various sampling times of data collected by other sensors, where i represents the i-th sampling time, and x... i For sampling time t i The actual observations collected represent the values ​​of other sensors at t i The actual reading; , representing the number of the sampling time closest to t; The observation corresponding to the most recent sampling time t; This is the equivalent observation value of the sensor at reference time point t, calculated using the nearest neighbor interpolation method. For example, the equivalent indoor temperature value of outdoor temperature T1 at reference time point t is: (2) By repeating the above steps, the timing alignment of sensor data collected by multiple sensors can be achieved.

[0063] As an example, let's consider using a moving average method to smooth sensor data collected by multiple sensors. For each moment's sensor data, the average of the original observations at that moment and several previous moments can be used to replace the current original observation. Assume a historical data sequence collected by a certain sensor is {x1, x2, ..., x...}. n If the window length is N1, then the moving average at time t is: (3) in, x is the moving average (i.e., smoothed value) of the sensor at time t. t-i For the sensor at time t The original observation value of i, and N1 is the length of the sliding window. For example, for indoor temperature T2, For the indoor temperature sensor at time t The original observed value of i, and the temperature value after smoothing the indoor temperature T2. .

[0064] As an example, the following formula can be used to normalize sensor data from multiple sensors: (4) in, For normalized sensor data (values ​​ranging from -1 to 1), x min x is the minimum value in the sensor data. max This represents the maximum value in the sensor data. For example, for the normalization of indoor wind speed WS, the normalized WS is: .

[0065] in, For the normalized WS, WS min For minimum indoor wind speed, WS max This represents the maximum indoor wind speed.

[0066] In summary, by acquiring raw sensor data from multiple sensors deployed in the space where the air conditioner is located, and performing preprocessing operations including time alignment, smoothing, normalization, and noise reduction, interference caused by inconsistent timestamps, measurement noise, dimensional differences, and signal fluctuations in multi-source heterogeneous sensor data can be effectively eliminated. This generates high-quality, highly consistent actual operating environment data, thereby improving the input reliability of subsequent feature coding and load prediction.

[0067] In any embodiment of this application, the first feature representation may be obtained by sparsely encoding the actual operating environment data of the air conditioner using a trained sparse encoder. The sparse encoding can automatically extract the most discriminative key features from the high-dimensional and redundant actual operating environment data, suppress the interference of irrelevant or weakly correlated variables, thereby improving the information density and semantic focus of the first feature representation and improving the efficiency and accuracy of subsequent load prediction. In addition, the sparse first feature representation has stronger generalization ability and noise robustness, and can better adapt to non-stationary dynamic changes such as fluctuations in the number of objects and sudden changes in the state of doors and windows in the air conditioner operating environment.

[0068] In summary, the first feature representation obtained through sparse coding not only enhances the model's perception accuracy of real thermal disturbance factors, but also takes into account computational efficiency and prediction stability, providing strong support for high-precision and low-cost prediction of air conditioning load.

[0069] Step S205: Perform load forecasting based on the first feature representation to obtain the predicted load of the air conditioner.

[0070] It should be noted that the explanation of step S205 can be found in the relevant description in any embodiment of this application, and will not be repeated here.

[0071] The air conditioning load prediction method in this application classifies objects within the space where the air conditioner is located based on their heat dissipation and dynamically monitors the quantity of each type of object. This allows for a more accurate quantification of the heat source distribution characteristics within the space, generating more physically meaningful space occupancy parameters. Simultaneously, fine-grained monitoring of key ventilation behaviors such as the number of times doors and windows are opened and closed, the duration of opening and closing, and the opening and closing angles effectively characterizes the impact of external air infiltration on the indoor thermal environment, thereby obtaining high-resolution door and window operation parameters. Thus, the physical mechanisms of heat source characteristics and ventilation behavior can be integrated into the data acquisition process, significantly improving the representativeness of actual operating environment data and the correlation with load prediction. This provides more discriminative input features for subsequent high-precision air conditioning load prediction, thereby enhancing the accuracy and scenario adaptability of the prediction results.

[0072] The above embodiments are application methods of sparse encoders. This application also provides a training method for sparse encoders.

[0073] Figure 3 This is a flowchart illustrating a training method for a sparse encoder provided for an exemplary embodiment of this application.

[0074] It should be noted that the training method of the sparse encoder can be executed alone, or it can be executed together with any embodiment of this application or any possible implementation in the embodiment, or it can be executed together with any technical solution in the related technology. The embodiments of this application do not limit this.

[0075] like Figure 3 As shown, the training method for this sparse encoder may include the following steps S301 to S304: Step S301: Obtain at least one first sample; wherein the first sample includes the first operating environment data of the sample air conditioner.

[0076] The sample air conditioner and the air conditioner for which the load needs to be predicted in the above embodiments may be the same air conditioner or different air conditioners. This application does not limit this.

[0077] The first operating environment data can be collected by various sensors in the space where the sample air conditioner is located, including but not limited to the following: space occupancy parameters of the space where the sample air conditioner is located, door and window operation parameters, outdoor meteorological parameters, and indoor environmental parameters.

[0078] Step S302: Use a sparse encoder to sparsely encode the first operating environment data to obtain the second feature representation.

[0079] In this embodiment of the application, a sparse encoder can be used to sparsely encode the first operating environment data to obtain a sparse feature representation, which is referred to as the second feature representation in this application.

[0080] For example, the encoding formula of a sparse encoder is as follows: (5) Where x is the input vector (i.e., the first runtime environment data). For the weights of the sparse encoder, Let f be the activation function in the sparse encoder (e.g., the Rectified Linear Unit (ReLU) activation function), h be the sparse feature representation, and f be the sparse feature representation. enc It is the encoding function of the sparse encoder.

[0081] Step S303: The second feature representation is decoded using a decoder to obtain the reconstructed second operating environment data.

[0082] In this embodiment of the application, a decoder can be used to decode the dimensionality-reduced second feature representation to obtain the reconstructed sample air conditioner operating environment data, which is referred to as the second operating environment data in this application.

[0083] For example, the decoding formula of the decoder is as follows: (6) Among them, W d Here, h represents the weights of the decoder, and b represents the sparse feature representation of the influence on air conditioning load after dimensionality reduction. d For the decoder bias, y is the activation function in the decoder. out For the reconstructed runtime environment data, f dec It is the decoding function of the decoder.

[0084] Step S304: Jointly train the sparse encoder and decoder based on the first operating environment data, the second operating environment data, and the second feature representation.

[0085] In the embodiments of this application, the value of the loss function can be determined by combining the first operating environment data, the second operating environment data, and the second feature representation. In this application, this value is denoted as the target loss value. Based on the target loss value, the sparse encoder and decoder can be jointly trained to minimize the target loss value.

[0086] It should be noted that the above example only uses the termination condition of joint training as the objective of minimizing the loss value, but this application is not limited to this. For example, the termination condition may also include: the training time reaches a set time, the training rounds reach a set number of rounds, etc. The embodiments of this application do not impose specific limitations on this.

[0087] In any embodiment of this application, the target loss value is calculated, for example, by determining a reconstruction loss term based on the difference between the second operating environment data and the first operating environment data, and determining a sparsity penalty term based on the second feature representation. Thus, in this application, the target loss value can be determined based on the sum of the reconstruction loss term and the sparsity penalty term.

[0088] Among them, the reconstruction loss term is positively correlated with the difference between the second operating environment data and the first operating environment data. That is, the greater the difference, the larger the value of the reconstruction loss term, and vice versa.

[0089] As an example, the sparsity penalty term is determined as follows: based on the second feature representation, the average activation value of each neuron in the sparse encoder on at least one first sample is determined, and based on the average activation value of each neuron and the set expected sparsity, the sparsity regularization term of each neuron is determined, wherein the sparsity regularization term is used to indicate the degree of sparsity of neuron activation. Thus, in this application, the sparsity penalty term can be determined based on the sum of the sparsity regularization terms of each neuron.

[0090] For example, the target loss value It can be calculated using the following formula: (7) Where N2 is the number of the first sample, m is the number of hidden layer neurons (referred to as hidden neurons) of the sparse encoder, and x (j) This is the first runtime environment data of the j-th first sample. It is the second runtime environment data output by the decoder, β is the sparse regularization weight, and ρ is the expected sparsity.

[0091] in, To reconstruct the loss term, For sparse penalty terms, Let be the sparse regularization term for the i-th hidden neuron, which can be calculated as follows: (8) in, , where is the average activation value of the i-th hidden neuron across all the first samples, and KL is short for Kullback-Leibler divergence.

[0092] The optimization objective of joint training can be to minimize the target loss value: (9) in, Represents all trainable parameters, including W. e b e W d b d .

[0093] In summary, by jointly optimizing the reconstruction loss term and the sparsity penalty term, and using the minimization of their weighted sum as the optimization objective, end-to-end joint training of the sparse encoder and decoder not only ensures the high-fidelity reproduction of the original operating environment data from the reconstructed operating environment data, but also forces the hidden layer features (i.e., the second feature representation) to maintain sparsity, thereby effectively suppressing redundant information and highlighting key thermal disturbance features. The first feature representation learned in this way has stronger discriminative power, robustness, and generalization ability, and can more accurately characterize the intrinsic relationship between air conditioning load and the actual operating environment, significantly improving the accuracy of subsequent load forecasting.

[0094] The sparse encoder training method of this application employs a self-encoding structure consisting of a sparse encoder and a decoder, enabling efficient, low-dimensional, and sparse representations of actual air conditioning operating environment data in an unsupervised manner. During the joint training phase, the model not only minimizes the reconstruction error (i.e., the difference between the first operating environment data and the reconstructed second operating environment data), but also uses sparsity constraints to ensure that the second feature representation retains only the key information most explanatory to load changes. The resulting sparse encoder effectively filters redundancy and noise from the actual operating environment data, extracting key features highly correlated with air conditioning load, thereby significantly improving the accuracy and generalization ability of subsequent load prediction.

[0095] As one possible implementation method, Figure 4 This is a flowchart illustrating yet another air conditioning load forecasting method provided for an exemplary embodiment of this application.

[0096] It should be noted that the air conditioning load forecasting method can be executed alone, or it can be executed together with any embodiment or possible implementation in the embodiment of this application, or it can be executed together with any technical solution in the related technology. The embodiments of this application do not limit this.

[0097] like Figure 4 As shown, the air conditioning load forecasting method may include the following steps S401 to S404: Step S401: Obtain the actual operating environment data of the air conditioner and encode the actual operating environment data to obtain the first feature representation.

[0098] The actual operating environment data includes space occupancy parameters and door and window operation parameters. The space occupancy parameters are used to indicate the information of objects in the space where the air conditioner is located, and the door and window operation parameters are used to indicate the information of opening and closing of doors and windows in the space.

[0099] It should be noted that the explanation of step S401 can be found in the relevant description in any embodiment of this application, and will not be repeated here.

[0100] Step S402: Query the target regression tree model according to the first feature representation; wherein, the target regression tree model is constructed based on multiple second samples in the sample set, and the leaf nodes in the target regression tree model have corresponding loads, which are used to indicate the actual load of the sample air conditioner under the operating environment data of each second sample contained in the leaf node.

[0101] The target regression tree model can be constructed based on multiple second samples in the sample set. Each second sample includes a sparse feature representation of the operating environment data of a sample air conditioner, and the actual load y of that sample air conditioner under that operating environment data.

[0102] The sparse feature representation can be obtained by sparsely encoding the sample air conditioner's operating environment data using the trained sparse encoder described above. For example, the sparse feature representation is labeled as follows: , where d refers to the number of feature dimensions.

[0103] In the target regression model, each leaf node has a corresponding load, which indicates the actual load y of the sample air conditioner under the operating environment data of each second sample contained in that leaf node. For example, the mean and median of the actual load y of the sample air conditioner under the operating environment data of each second sample contained in each leaf node can be used as the load of that leaf node.

[0104] As an example, multiple second samples in the sample set can be recursively partitioned to construct the target regression tree model. Exemplarily, the multidimensional sparse feature representation can be partitioned layer by layer, automatically identifying the partition intervals of key features affecting the predicted load. At each node, all features and candidate split points are traversed, the residual sum of squares (RSS) after partitioning is calculated, and the partitioning method that minimizes the RSS is selected. For each encoded feature z... j And the split point s divides it into two subsets R1 and R2: (10) The sum of squared residuals is: (11) in, It is the actual load of the i-th second sample in subset R1 / R2. Let y be the average value of the actual load y corresponding to R1. R2 represents the average value of the actual load y. and The calculation formula is as follows: (12) The relatively better split point The choice is: (13) The target regression tree model can be constructed using the above splitting method. Taking the load of each leaf node in the target regression tree model as an example, which indicates the mean of the actual load y of the air conditioners in the sample air conditioning data of each second sample contained in that leaf node, the load of the leaf node... It can be calculated using the following formula: (14) Among them, S L It is the set of indices of the second sample that falls into the same leaf node.

[0105] As an example, the construction method of the target regression tree model is as follows: multiple second samples in the sample set are grouped to obtain k disjoint subsets, and k candidate regression tree models are constructed based on the k subsets using the k-fold cross-validation algorithm (for example, each candidate regression tree model can be constructed using the above formula (10-13)). The evaluation index of the kth candidate regression tree model is determined. Thus, in this application, the target regression tree model can be determined from the k candidate regression tree models based on the evaluation index of the k candidate regression tree models.

[0106] The evaluation indicators include positive evaluation indicators and / or negative evaluation indicators.

[0107] For example, the target regression tree model can be determined by selecting the candidate regression tree model with the largest positive evaluation index as the target regression tree model, or by selecting the candidate regression tree model with the smallest negative evaluation index as the target regression tree model.

[0108] For example, the construction method of the candidate regression tree model and the determination method of the corresponding evaluation index are as follows: select the i-th subset of the k subsets as the test set, and take the k-1 subsets other than the test set as the training set. Then, the multiple second samples in the training set can be recursively divided to construct the i-th candidate regression tree model. For example, the i-th candidate regression tree model can be constructed using the above formula (10-13). Then, the i-th candidate regression tree model can be used to test the test set to determine the evaluation index of the i-th candidate regression tree model based on the test results.

[0109] Where k is a natural number greater than 1; i is a positive integer not greater than k.

[0110] For example, using a negative evaluation metric such as Mean Absolute Percentage Error (MAPE) as an example, its calculation method is as follows: (15) Among them, y i To test the actual load of the i-th second sample in the test set, Let n be the predicted load of the i-th second sample in the test set, and n be the number of second samples in the test set.

[0111] In summary, by dividing the sample set into k disjoint subsets and constructing and evaluating k candidate regression tree models using a k-fold cross-validation algorithm, we can fully utilize the limited training data and effectively avoid overfitting while evaluating the model's generalization performance under different data distributions. Based on the evaluation metrics of the k candidate regression tree models, the target regression tree model is selected optimally, ensuring that the selected target regression tree model achieves the best balance in predictive stability, accuracy, and robustness. Therefore, this not only improves the target regression tree model's ability to model the complex nonlinear relationship between air conditioning load and operating environment but also enhances its adaptability and reliability in practical deployments facing diverse usage scenarios and data fluctuations.

[0112] Step S403: From each leaf node of the target regression tree model, determine the target leaf node that matches the first feature representation.

[0113] In this embodiment, the target regression tree model can be queried based on the first feature representation to determine the leaf node that matches the first feature representation from each leaf node of the target regression tree model. In this application, the leaf node is referred to as the target leaf node.

[0114] Step S404: Determine the predicted load of the air conditioner based on the load of the target leaf node.

[0115] For example, the load of the target leaf node can be used as the predicted load of the air conditioner.

[0116] The air conditioning load prediction method in this application embodiment has the ability to model nonlinear relationships and high-dimensional features by means of regression tree models, and its leaf nodes aggregate actual load statistics under similar operating environments. Therefore, it can effectively capture the complex mapping law between dynamic environmental factors such as space occupancy and door and window status and air conditioning load, and improve the model transparency while ensuring prediction accuracy.

[0117] As one possible implementation method, Figure 5 A schematic flowchart of another air conditioning load forecasting method provided for an exemplary embodiment of this application.

[0118] It should be noted that the air conditioning load forecasting method can be executed alone, or it can be executed together with any embodiment or possible implementation in the embodiment of this application, or it can be executed together with any technical solution in the related technology. The embodiments of this application do not limit this.

[0119] like Figure 5 As shown, the air conditioning load forecasting method may include the following steps S501 to S505: Step S501: Obtain the actual operating environment data of the air conditioner and encode the actual operating environment data to obtain the first feature representation.

[0120] The actual operating environment data includes space occupancy parameters and door and window operation parameters. The space occupancy parameters are used to indicate the information of objects in the space where the air conditioner is located, and the door and window operation parameters are used to indicate the information of opening and closing of doors and windows in the space.

[0121] Step S502: Perform load prediction based on the first feature representation to obtain the predicted load of the air conditioner.

[0122] It should be noted that the explanations of steps S501 to S502 can be found in the relevant descriptions in any embodiment of this application, and will not be repeated here.

[0123] Step S503: From a variety of preset operating modes, determine the target operating mode that matches the predicted load, and control the air conditioner to switch to the target operating mode.

[0124] The operating modes include, but are not limited to: energy-saving mode, comfort mode, and silent mode.

[0125] For example, the correlation between different operating modes and load value ranges can be pre-configured. In this application, the above correlation can be queried according to the predicted load of the air conditioner to determine the target operating mode and control the air conditioner to switch to the target operating mode; wherein the predicted load of the air conditioner is located within the load value range associated with the target operating mode.

[0126] For example, when the predicted load is relatively low, the air conditioner can be switched to energy-saving mode to reduce energy consumption; when the predicted load is relatively high and may continue to rise, the air conditioner can be switched to non-energy-saving mode or high-performance mode to ensure that the load demand is met.

[0127] Step S504: Adjust the operating parameters of the air conditioner in the current operating mode according to the predicted load.

[0128] The operating parameters include, but are not limited to, at least one of the following: temperature setpoint, humidity setpoint, start / stop time, wind speed, and wind direction.

[0129] For example, different load value ranges and the correlation between operating parameters can be pre-configured. In this application, the above correlation can be queried according to the predicted load of the air conditioner to determine the target operating parameters associated with the load value range in which the predicted load is located. Based on the target operating parameters, the operating parameters of the air conditioner in the current operating mode can be adjusted.

[0130] In step S505, in response to the air conditioner including multiple indoor units, the operating strategies of the multiple indoor units are adjusted according to the predicted load.

[0131] The operating strategies include, but are not limited to, at least one of the following: load distribution, start-stop scheduling, and air supply parameter coordination. Load distribution strategy determines the rationality and balance of the load borne by each indoor unit; a reasonable load distribution strategy can prevent some indoor units from operating under overload conditions and extend the equipment's lifespan. Start-stop scheduling strategy determines the on and off states of different indoor units at different times, helping to reduce equipment energy consumption. Air supply parameter coordination strategy involves the unified adjustment of parameters such as wind speed, wind direction, and temperature, directly affecting the comfort of the indoor environment.

[0132] For example, at least one strategy among load distribution, start-stop scheduling, and air supply parameter coordination of multiple indoor units can be adjusted based on the performance characteristics of multiple indoor units and the predicted load distribution status, in order to improve the overall operating efficiency and reliability of the air conditioning system.

[0133] For example, when the predicted load on the air conditioning system shows an increasing trend (e.g., rising outdoor temperatures or increased indoor heat dissipation), the load allocation strategy can be adjusted first: based on the cooling / heating capacity of each indoor unit, the increased load should be rationally allocated to the higher-performance indoor units to ensure their efficient operation, while preventing the lower-performance indoor units from malfunctioning due to overload. For the start-stop scheduling strategy, some standby indoor units can be turned on in advance to meet the upcoming increased load demand and prevent excessive system pressure due to a sudden increase in load. For the air supply parameter coordination strategy, the air supply velocity can be appropriately increased to accelerate indoor air circulation, allowing heat or cooling to be transferred to all corners of the room more quickly, thereby improving the indoor cooling / heating effect in a short time to cope with the increased predicted load. Through the coordinated adjustment of these strategies, the air conditioning system's ability to cope with load changes can be effectively improved, ensuring its stable and efficient operation.

[0134] It should be noted that any one of steps S503, S504, and S505 can be executed, or multiple steps can be executed simultaneously; this embodiment of the application does not impose any restrictions on this. When multiple steps in S503, S504, and S505 are executed simultaneously, their execution order is not restricted. Figure 5 The description is provided for illustrative purposes only, and the steps are executed sequentially from S503 to S505.

[0135] The air conditioning load prediction method of this application actively applies the predicted load to the intelligent control of the air conditioning system. This "prediction-decision-control" closed-loop mechanism not only gives full play to the pre-guidance role of load prediction and significantly improves the energy efficiency and comfort of the air conditioning system, but also realizes flexible control from single-unit fine adjustment to multi-unit collaborative optimization, effectively reducing energy consumption, extending equipment life, and enhancing the adaptability of the air conditioning system to complex usage scenarios.

[0136] In any embodiment of this application, the operating environment data affecting the air conditioning load collected by this application includes, but is not limited to: outdoor meteorological parameters (outdoor temperature T1, outdoor humidity RH1, solar radiation intensity DNI1), indoor environmental parameters (indoor temperature T2, indoor humidity RH2, indoor wind speed WS), space occupancy parameters (number of objects with different heat dissipation, including indoor occupants PN, number of heat dissipation devices, number of pets), and door and window operation parameters (such as number of times doors and windows are opened and closed DN (times / hour), duration of door and window opening and closing, and door and window opening and closing angle, etc.). Based on the above operating environment data, the air conditioning load can be predicted quickly and accurately.

[0137] For example, the principle of air conditioning load forecasting can be as follows: Figure 6 As shown, it mainly includes the following parts: Part 1: Based on IoT-based multi-source sensors and real-time meteorological information, realize the collection of multi-source operating environment data of air conditioning system.

[0138] This application can obtain multi-source operating environment data that affects air conditioning load, including indoor environmental parameters, outdoor meteorological parameters, space occupancy parameters, and door and window operation parameters.

[0139] Leveraging IoT technology, sensor data collected from multiple sources can be efficiently scheduled and managed, improving the sensing capabilities and response efficiency of the air conditioning system. To achieve more comprehensive and accurate environmental perception, real-time outdoor meteorological parameters also need to be collected simultaneously, providing more sufficient input data support for the load forecasting phase.

[0140] In this application, multi-source sensors can be used to collect and transmit different parameters affecting air conditioning load. The controller then transmits these parameters back to the gateway. Finally, the collected sensor data undergoes normalization, smoothing, data filling, feature extraction, and load prediction modeling. The multi-source sensors may include, for example... Figure 7 The temperature sensor, humidity sensor, wind speed sensor, thermal imaging sensor (or infrared people sensor), solar radiation sensor, and door magnetic sensor shown are all included. (The door magnetic sensor outputs the switch status signal in real time and uploads it to the air conditioning controller or intelligent central control system.)

[0141] The following sensor data collected from multiple sources include: outdoor temperature T1, outdoor humidity RH1, indoor temperature T2, indoor humidity RH2, solar radiation intensity DNI1, indoor wind speed WS, number of people indoors PN, and number of times doors are opened and closed DN (times / hour) as an example, and are denoted as dataset DS1.

[0142] Part Two: Add a unified timestamp to the collected operating environment data to complete the time synchronization process.

[0143] In a multi-sensor system for air conditioning load prediction, due to differences in performance and different tasks performed by the sensors, the measurement data obtained by each sensor from observing the same target may not be synchronized. Therefore, it is necessary to convert the measurement data obtained by different sensors at different times to a unified fusion time.

[0144] For example, the nearest neighbor interpolation method shown in formulas (1) and (2) can be used to achieve synchronous mapping of measurement data observed by each sensor at any time. The dataset after time synchronization processing is denoted as dataset DS2.

[0145] Part Three: Perform preprocessing operations such as smoothing, normalization, and noise reduction on the operating environment data to improve data quality and usability.

[0146] Monitoring data from air conditioning systems often fluctuates and contains noise due to sensor errors and environmental disturbances. To improve data quality and the accuracy of subsequent modeling, sensor data collected by various sensors can be smoothed to remove abnormal fluctuations and highlight the main trend characteristics. For example, a moving average method can be used to smooth the sensor data, resulting in dataset DS3.

[0147] To eliminate the influence of different dimensions between sensor data, dataset DS3 can be normalized to ensure comparability between data indicators and bring them to the same order of magnitude. For example, formula (4) can be used to normalize the sensor data in dataset DS3 to obtain dataset DS4.

[0148] Part 4: Use a sparse encoder (such as a multilayer autoencoder) to perform sparse coding (feature extraction) on the preprocessed operating environment data to obtain fused features or sparse feature representations that affect air conditioning load.

[0149] The training method for sparse encoders can be as follows: Figure 8 As shown, the main steps include: using a sparse encoder to learn features from the DS4 dataset to obtain a dimensionality-reduced sparse feature representation of the impact on air conditioning load; using a decoder to reconstruct the dimensionality-reduced sparse feature representation to obtain reconstructed operating environment data affecting air conditioning load; calculating a loss function based on the reconstructed operating environment data and the actual observed operating environment data, and jointly training the sparse encoder and decoder based on the loss function. The joint training process stops when the loss function converges to its minimum value, resulting in the trained sparse encoder.

[0150] For example, the loss function can be calculated using the above formula (7), and the optimization objective can be shown in formula (9). That is, the sparse encoder can make the hidden layer features efficiently represent the input by minimizing the weighted sum of the reconstruction error term and the sparse penalty term.

[0151] The trained sparse encoder is used to perform sparse encoding or feature extraction on the DS4 dataset to obtain a sparse feature representation of the impact on air conditioning load after dimensionality reduction, denoted as the DS5 dataset. , where d is the encoded feature dimension, d < 8.

[0152] Part 5: Air Conditioning Load Forecasting Based on Decision Tree Algorithm. Decision trees can adapt to multivariate interactions and complex data distributions, and have a fast response speed, enabling efficient and real-time forecasting of air conditioning loads, thereby improving the intelligence level and operating efficiency of air conditioning systems.

[0153] 1. Divide the DS5 dataset into training and testing sets according to a ratio (e.g., 80% training, 20% testing), and use a k (e.g., 5) fold cross-validation mechanism to evaluate the model's generalization ability.

[0154] 2. By constructing an air conditioning load prediction model based on a regression tree structure (referred to as the regression tree model in this application), the multidimensional sparse feature representation of the input is partitioned layer by layer, automatically identifying the feature partitioning intervals that are key to the predicted load. At each node, the model traverses all features and candidate splitting points, calculates the residual sum of squares (RSS) after splitting, and selects the partitioning method that minimizes the RSS. For each encoded feature z... j The split point s is divided into two subsets R1 and R2 as shown in formula (10). The sum of squared residuals can be shown in formula (11). Formula (13) is used to determine the relatively better split point. Finally, the predicted load of the leaf node can be shown in formula (14).

[0155] 3. After generating the regression tree model, the accuracy of the regression tree model can be tested using the test set. For example, the evaluation index can be the mean absolute percentage error shown in formula (15).

[0156] 4. Input the sparse feature representation of the actual operating environment data of the air conditioner into the regression tree model in real time to quickly map the input sparse feature representation and output the predicted load value of the air conditioner, providing data support for subsequent system scheduling and energy-saving optimization strategies.

[0157] In summary, the solution provided in this application has at least the following advantages: In the field of air conditioning load forecasting, multi-sensor data fusion technology can comprehensively process sensor data from different sources and types, giving full play to their respective advantages and characteristics, thereby extracting more scientific, effective, and practically valuable information than single data. Sparse coding technology can fuse multi-dimensional data affecting air conditioning load, extract key information, reduce redundancy, and improve forecasting efficiency. Decision tree algorithms have strong nonlinear modeling capabilities, can automatically analyze and filter the main factors affecting air conditioning load, and effectively characterize the complex variable relationships in building energy consumption data; among them, the hierarchical branching structure of the decision tree not only gives the model good interpretability and transparency, making it easy for engineers to understand and apply, but also automatically selects relatively optimal splitting features and thresholds (such as s in formula (10-13)). In the process of air conditioning load forecasting, decision trees can adapt to multivariate interactions and complex data distributions, and have a fast response speed, enabling efficient and real-time forecasting of air conditioning load, further improving the intelligence level and operating efficiency of the air conditioning system.

[0158] To achieve the above embodiments, this application also proposes an air conditioning load prediction device.

[0159] Figure 9 This is a schematic diagram of the structure of an air conditioning load prediction device provided for an exemplary embodiment of this application.

[0160] like Figure 9 As shown, the air conditioning load prediction device 900 may include: a first acquisition module 910, a first encoding module 920, and a prediction module 930.

[0161] The first acquisition module 910 is used to acquire the actual operating environment data of the air conditioner. The actual operating environment data includes space occupancy parameters and door and window operation parameters. The space occupancy parameters are used to indicate the information of objects in the space where the air conditioner is located, and the door and window operation parameters are used to indicate the information of opening and closing of doors and windows in the space. The first encoding module 920 is used to encode the actual operating environment data to obtain the first feature representation; The prediction module 930 is used to perform load prediction based on the first feature representation to obtain the predicted load of the air conditioner.

[0162] In one implementation of this application, the first acquisition module 910 is used to: classify objects based on the heat dissipation of objects existing in the space to obtain at least one type of object; wherein objects of the same type have similar heat dissipation; monitor the number of objects of at least one type to obtain space occupancy parameters; and monitor at least one of the number of times doors and windows are opened and closed, the duration of opening and closing, and the opening and closing angle to obtain door and window operation parameters.

[0163] In one implementation of this application, the first encoding module 920 is used to: use a sparse encoder to sparsely encode the actual operating environment data to obtain a first feature representation; The sparse encoder is trained using the following modules: The second acquisition module is used to acquire at least one first sample; wherein, the first sample includes the first operating environment data of the sample air conditioner; The second encoding module is used to sparsely encode the first operating environment data using a sparse encoder to obtain a second feature representation. The decoding module is used to decode the second feature representation using a decoder to obtain the reconstructed second runtime environment data; The joint training module is used to jointly train the sparse encoder and decoder based on the first runtime environment data, the second runtime environment data, and the second feature representation.

[0164] In one implementation of this application, the joint training module is used to: determine a reconstruction loss term based on the difference between the second operating environment data and the first operating environment data; determine a sparse penalty term based on the second feature representation; determine a target loss value based on the sum of the reconstruction loss term and the sparse penalty term; and perform joint training on the sparse encoder and decoder using the target loss value.

[0165] In one implementation of this application, the joint training module is configured to: determine the average activation value of each neuron in the sparse encoder on at least one first sample based on the second feature representation; determine the sparse regularization term of each neuron based on the average activation value of each neuron and the set expected sparsity; wherein the sparse regularization term is used to indicate the degree of sparsity of neuron activation; and determine the sparse penalty term based on the sum of the sparse regularization terms of each neuron.

[0166] In one implementation of this application, the prediction module 930 is configured to: query a target regression tree model based on a first feature representation; wherein the target regression tree model is constructed based on multiple second samples in a sample set, and the leaf nodes in the target regression tree model have corresponding loads, which are used to indicate the actual load of the sample air conditioner under the operating environment data of each second sample contained in the leaf node; determine the target leaf node that matches the first feature representation from each leaf node of the target regression tree model; and determine the predicted load of the air conditioner based on the load of the target leaf node.

[0167] In one implementation of this application, the target regression tree model is constructed using the following modules: The grouping module is used to group multiple second samples in the sample set into k disjoint subsets; where k is a natural number greater than 1. The processing module is used to construct k candidate regression tree models based on k subsets using a k-fold cross-validation algorithm, and to determine the evaluation index of the k-th candidate regression tree model. The determination module is used to determine the target regression tree model from the k candidate regression tree models based on the evaluation indicators of the k candidate regression tree models.

[0168] In one implementation of this application, the processing module is configured to: select the i-th subset of k subsets as the test set, and select the k-1 subsets other than the test set as the training set; where i is a positive integer not greater than k; recursively partition multiple second samples in the training set to construct the i-th candidate regression tree model; and use the i-th candidate regression tree model to test the test set to determine the evaluation index of the i-th candidate regression tree model based on the test results.

[0169] In one implementation of this application, the first acquisition module 910 is used to: acquire sensor data collected by multiple sensors in the space where the air conditioner is located; preprocess the sensor data of the multiple sensors to obtain actual operating environment data; wherein, the preprocessing includes at least one of time alignment processing, smoothing processing, normalization processing, and noise reduction processing.

[0170] In one implementation of this application embodiment, the air conditioning load prediction device 900 may further include: The control module is configured to perform at least one of the following: determine a target operating mode matching the predicted load from a plurality of preset operating modes, and control the air conditioner to switch to the target operating mode; adjust the operating parameters of the air conditioner in the current operating mode according to the predicted load; wherein the operating parameters include at least one of the following: temperature setpoint, humidity setpoint, start-stop time, fan speed, and fan direction; and adjust the operating strategies of the multiple indoor units according to the predicted load in response to the air conditioner including multiple indoor units; wherein the operating strategies include at least one of load distribution, start-stop scheduling, and air supply parameter coordination.

[0171] In one implementation of this application, the actual operating environment data further includes: outdoor meteorological parameters and / or indoor environmental parameters; wherein, the outdoor meteorological parameters include at least one of outdoor temperature, outdoor humidity, and solar radiation intensity; wherein, the indoor environmental parameters include at least one of indoor temperature, indoor humidity, and indoor wind speed.

[0172] It should be noted that the foregoing explanation of any air conditioning load prediction method or sparse encoder training method embodiment also applies to the air conditioning load prediction device of this embodiment, and will not be repeated here.

[0173] In the air conditioning load prediction device of this application embodiment, key dynamic information in the actual air conditioning operating environment is collected, including space occupancy parameters reflecting the distribution of heat sources such as people, equipment, and pets in the space where the air conditioner is located, and door and window operation parameters characterizing the opening and closing behavior or ventilation intensity. The above parameters are encoded to generate a first feature representation, and load prediction is performed based on the first feature representation. This can effectively avoid the dependence of physical modeling on complex and difficult-to-obtain physical parameters. That is, in this application, the main disturbance factors in the real operating scenario of air conditioning are captured in a data-driven manner, which significantly improves the accuracy and real-time adaptability of air conditioning load prediction.

[0174] To implement the above embodiments, this application also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the air conditioning load prediction method or the sparse encoder training method as described in any of the foregoing embodiments.

[0175] To implement the above embodiments, this application also proposes an air conditioner, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to: implement the air conditioner load prediction method or the sparse encoder training method as described in any of the foregoing embodiments.

[0176] Figure 10 This is a schematic diagram of the structure of an electronic device provided for an exemplary embodiment of this application. For example, the electronic device 1000 may be a server, a terminal, a vehicle, etc.

[0177] Reference Figure 10 The electronic device 1000 may include one or more of the following components: a processing component 1002, a memory 1004, a power component 1006, a multimedia component 1008, an audio component 1010, an input / output (I / O) interface 1012, a sensor component 1014, and a communication component 1016.

[0178] Processing component 1002 typically controls the overall operation of electronic device 1000, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 1002 may include one or more processors 1020 to execute instructions to perform all or part of the steps of the methods described above. Furthermore, processing component 1002 may include one or more modules to facilitate interaction between processing component 1002 and other components. For example, processing component 1002 may include a multimedia module to facilitate interaction between multimedia component 1008 and processing component 1002.

[0179] The memory 1004 is configured to store various types of data to support the operation of the electronic device 1000. The memory 1004 can be implemented by any type of volatile or non-volatile storage device or a combination thereof.

[0180] Power component 1006 provides power to various components of electronic device 1000. Power component 1006 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 1000.

[0181] The multimedia component 1008 includes a screen that provides an output interface between the electronic device 1000 and the user. In some embodiments, the multimedia component 1008 includes a front-facing camera and / or a rear-facing camera. Each front-facing and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0182] Audio component 1010 is configured to output and / or input audio signals. For example, audio component 1010 includes a microphone (MIC) configured to receive external audio signals when electronic device 1000 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 1004 or transmitted via communication component 1016. In some embodiments, audio component 1010 also includes a speaker for outputting audio signals.

[0183] I / O interface 1012 provides an interface between processing component 1002 and peripheral interface modules, such as keyboards, click wheels, buttons, etc.

[0184] Sensor assembly 1014 includes one or more sensors for providing state assessments of various aspects of electronic device 1000. For example, sensor assembly 1014 can detect the on / off state of electronic device 1000, the relative positioning of components, changes in the position of electronic device 1000 or a component of electronic device 1000, the presence or absence of user contact with electronic device 1000, the orientation or acceleration / deceleration of electronic device 1000, temperature changes of electronic device 1000, the presence of nearby objects, etc.

[0185] Communication component 1016 is configured to facilitate wired or wireless communication between electronic device 1000 and other devices. Electronic device 1000 can access wireless networks based on communication standards, such as WiFi, 4G, or 5G, or combinations thereof. In one exemplary embodiment, communication component 1016 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 1016 further includes a Near Field Communication (NFC) module to facilitate short-range communication.

[0186] In an exemplary embodiment, the electronic device 1000 may be implemented by one or more chips for performing the above-described method.

[0187] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 1004 including instructions, which can be executed by a processor 1020 of an electronic device 1000 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0188] To implement the above embodiments, this application also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the air conditioning load prediction method or the sparse encoder training method as described in any of the foregoing method embodiments.

[0189] To implement the above embodiments, this application also proposes a computer program product having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the air conditioning load prediction method or the sparse encoder training method as described in any of the foregoing method embodiments.

[0190] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0191] Furthermore, the terms "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 technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0192] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0193] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and compact disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0194] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0195] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0196] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0197] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for predicting air conditioning load, characterized in that, include: Acquire actual operating environment data of the air conditioner; wherein, the actual operating environment data includes space occupancy parameters and door and window operation parameters, the space occupancy parameters are used to indicate the information of objects existing in the space where the air conditioner is located, and the door and window operation parameters are used to indicate the information of opening and closing of doors and windows in the space; The actual operating environment data is encoded to obtain a first feature representation; Load prediction is performed based on the first feature representation to obtain the predicted load of the air conditioner.

2. The method according to claim 1, characterized in that, The acquisition of actual operating environment data for the air conditioner includes: Based on the heat dissipation of objects existing in the space, the objects are classified to obtain at least one type of object; wherein, objects of the same type have similar heat dissipation. The quantity of at least one type of object is monitored to obtain the space occupancy parameter; The number of times the doors and windows in the space are opened and closed, the duration of opening and closing, and the opening and closing angle are monitored to obtain the operating parameters of the doors and windows.

3. The method according to claim 1, characterized in that, The first feature represents that is The data is obtained by sparsely encoding the actual operating environment data using a sparse encoder. The training method of the sparse encoder includes: Obtain at least one first sample; wherein the first sample includes first operating environment data of the sample air conditioner; The first operating environment data is sparsely encoded using a sparse encoder to obtain a second feature representation; The second feature representation is decoded using a decoder to obtain the reconstructed second runtime environment data; The sparse encoder and the decoder are jointly trained based on the first operating environment data, the second operating environment data, and the second feature representation.

4. The method according to claim 3, characterized in that, The step of jointly training the sparse encoder and the decoder based on the first operating environment data, the second operating environment data, and the second feature representation includes: Based on the differences between the second operating environment data and the first operating environment data, the reconstruction loss item is determined; Based on the second feature representation, determine the sparsity penalty term; The target loss value is determined based on the sum of the reconstruction loss term and the sparse penalty term; The sparse encoder and the decoder are jointly trained using the target loss value.

5. The method according to claim 4, characterized in that, The step of determining the sparsity penalty term based on the second feature representation includes: Based on the second feature representation, determine the average activation value of each neuron in the sparse encoder on the at least one first sample; Based on the average activation value of each neuron and the set desired sparsity, a sparsity regularization term is determined for each neuron; wherein, the sparsity regularization term is used to indicate the degree of sparsity of the neuron's activation. The sparse penalty term is determined based on the sum of the sparse regularization terms of each neuron.

6. The method according to claim 1, characterized in that, The step of performing load forecasting based on the first feature representation to obtain the predicted load of the air conditioner includes: Based on the first feature representation, query the target regression tree model; wherein, the target regression tree model is constructed based on multiple second samples in the sample set, and the leaf nodes in the target regression tree model have corresponding loads, which are used to indicate the actual load of the sample air conditioner under the operating environment data of each second sample contained in the leaf node; From each leaf node of the target regression tree model, determine the target leaf node that matches the first feature representation; The predicted load of the air conditioner is determined based on the load of the target leaf node.

7. The method according to claim 6, characterized in that, The construction method of the target regression tree model includes: The multiple second samples in the sample set are grouped to obtain k disjoint subsets; where k is a natural number greater than 1. Based on the k subsets, a k-fold cross-validation algorithm is used to construct k candidate regression tree models, and the evaluation index of the kth candidate regression tree model is determined. The target regression tree model is determined from the k candidate regression tree models based on the evaluation index of the k candidate regression tree models.

8. The method according to claim 7, characterized in that, The step of constructing k candidate regression tree models based on the k subsets using a k-fold cross-validation algorithm, and determining the evaluation metric for the k-th candidate regression tree model, includes: Select the i-th subset from the k subsets as the test set, and select the k-1 subsets other than the test set as the training set; where i is a positive integer not greater than k; The training set is recursively divided into multiple second samples to construct the i-th candidate regression tree model; The i-th candidate regression tree model is used to test the test set, and the evaluation index of the i-th candidate regression tree model is determined based on the test results.

9. The method according to any one of claims 1-8, characterized in that, The acquisition of actual operating environment data for the air conditioner includes: Acquire sensor data collected by various sensors in the space where the air conditioner is located; The sensor data from the various sensors are preprocessed to obtain the actual operating environment data; The preprocessing includes at least one of time alignment processing, smoothing processing, normalization processing, and noise reduction processing.

10. The method according to any one of claims 1-8, characterized in that, After performing load forecasting based on the first feature representation to obtain the predicted load of the air conditioner, the method further includes at least one of the following: From a variety of preset operating modes, a target operating mode that matches the predicted load is determined, and the air conditioner is controlled to switch to the target operating mode. Based on the predicted load, the operating parameters of the air conditioner in the current operating mode are adjusted; wherein the operating parameters include at least one of the following: temperature setpoint, humidity setpoint, start / stop time, fan speed, and fan direction; In response to the air conditioner including multiple indoor units, the operating strategies of the multiple indoor units are adjusted according to the predicted load; wherein the operating strategies include at least one of load distribution, start-stop scheduling, and air supply parameter coordination.

11. The method according to any one of claims 1-8, characterized in that, The actual operating environment data also includes: outdoor meteorological parameters and / or indoor environmental parameters; The outdoor meteorological parameters include at least one of outdoor temperature, outdoor humidity, and solar radiation intensity. The indoor environmental parameters include at least one of indoor temperature, indoor humidity, and indoor wind speed.

12. An air conditioning load prediction device, characterized in that, include: The first acquisition module is used to acquire actual operating environment data of the air conditioner; wherein, the actual operating environment data includes space occupancy parameters and door and window operation parameters, the space occupancy parameters are used to indicate the information of objects existing in the space where the air conditioner is located, and the door and window operation parameters are used to indicate the information of opening and closing of doors and windows in the space; The first encoding module is used to encode the actual operating environment data to obtain a first feature representation; The prediction module is used to perform load prediction based on the first feature representation to obtain the predicted load of the air conditioner.

13. The apparatus according to claim 12, characterized in that, The first acquisition module is used for: Based on the heat dissipation of objects existing in the space, the objects are classified to obtain at least one type of object; wherein, objects of the same type have similar heat dissipation. The quantity of at least one type of object is monitored to obtain the space occupancy parameter; The number of times the doors and windows in the space are opened and closed, the duration of opening and closing, and the opening and closing angle are monitored to obtain the operating parameters of the doors and windows.

14. The apparatus according to claim 12, characterized in that, The first encoding module is configured to: use a sparse encoder to sparsely encode the actual operating environment data to obtain the first feature representation; The sparse encoder is trained using the following modules: The second acquisition module is used to acquire at least one first sample; wherein, the first sample includes first operating environment data of the sample air conditioner; The second encoding module is used to perform sparse encoding on the first operating environment data using a sparse encoder to obtain a second feature representation; The decoding module is used to decode the second feature representation using a decoder to obtain the reconstructed second runtime environment data; The joint training module is used to jointly train the sparse encoder and the decoder based on the first operating environment data, the second operating environment data, and the second feature representation.

15. The apparatus according to claim 14, characterized in that, The joint training module is used for: Based on the differences between the second operating environment data and the first operating environment data, the reconstruction loss item is determined; Based on the second feature representation, determine the sparsity penalty term; The target loss value is determined based on the sum of the reconstruction loss term and the sparse penalty term; The sparse encoder and the decoder are jointly trained using the target loss value.

16. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 11.

17. An air conditioner, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured as follows: The steps of implementing the method as described in any one of claims 1 to 11.

18. A non-transitory computer-readable storage medium having computer program instructions stored thereon, characterized in that, When executed by a processor, the program instructions implement the steps of the method described in any one of claims 1 to 11.

19. A computer program product, characterized in that, It includes a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 11.

Images