Air conditioner load data control method and system

Abstract

The invention relates to the technical field of air conditioner control, and discloses an air conditioner load data control method and system.The method comprises the steps that firstly, meteorological parameters and load parameters of an air conditioner are collected, and data cleaning, timestamp definition and standardization processing are conducted; then, defining a feature vector according to the processed data, generating a feature data set, and constructing a load prediction model by using a neural network algorithm; and pre-training the model by using the source domain data set through a transfer learning strategy, and enhancing the target domain data to generate a new sample. Next, parameters of the pre-training model are migrated to the new task model, and partial network layer parameters are frozen to form an initialized reinforcement learning model; and finally, training unfrozen layer parameters by using the target domain data set, and finely adjusting the model to realize real-time and accurate optimal setting of the air conditioner temperature, thereby achieving the purpose of load optimization control. Through combination of transfer learning and reinforcement learning, the generalization ability, robustness and adaptability of the model are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning control, and in particular to an air conditioning load data control method and system. Background Art

[0002] In the current field of air-conditioning load data control, with the rapid development of intelligent buildings and energy management systems, efficient and accurate load forecasting and optimal control of air-conditioning systems have become the key to improving energy efficiency and reducing operating costs.

[0003] In recent years, with the rise of data science and machine learning technology, some data-driven air conditioning load forecasting and control methods have gradually been applied. However, these methods still face many challenges in practical applications.

[0004] Traditional machine learning models often rely on a large amount of labeled data for training, but in the field of air conditioning load control, high-quality historical data is often difficult to obtain, and there are large differences and incompleteness between the data, which seriously restricts the learning effect and generalization ability of the model. On the other hand, even if enough data can be obtained, due to the complexity and dynamics of air conditioning system operation, a single model is difficult to adapt to the load control requirements in different scenarios, resulting in poor robustness and adaptability of the model in practical applications. Summary of the invention

[0005] The object of the present invention is to provide an air conditioning load data control method and system to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: an air conditioning load data control method, the method comprising:

[0007] S1: Collect meteorological parameters and load parameters of air conditioners, and clean, timestamp and standardize the collected data;

[0008] S2: Define feature vectors based on the collected data and generate several sets of combined features to form a feature data set;

[0009] S3: Use a neural network algorithm to build a load forecasting model, use the feature data set obtained in step S2 to generate a source domain data set and a target domain data set, and use the source domain data set to pre-train the load forecasting model;

[0010] S4: Enhance the data in the target domain and generate new data samples;

[0011] S5: Migrating the parameters of the load forecasting model to the model of the new task, freezing the parameters of some network layers, and obtaining an initialized reinforcement learning model; the reinforcement learning model is used to predict the optimal setting of the air conditioning temperature;

[0012] S6: Use the target domain dataset to train the parameters of the unfrozen layers in the reinforcement learning model and fine-tune the model;

[0013] S7: Obtain weather data in real time and use the trained reinforcement learning model to give the optimal air conditioning temperature setting to achieve load optimization control.

[0014] Preferably, in step S1, the data cleaning method includes data missing value processing and data denoising:

[0015] For missing values, linear interpolation algorithm is used to fill them;

[0016] For noisy data, the sliding average filtering algorithm is used for data denoising, and the algorithm formula is:

[0017]

[0018] Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.

[0019] Preferably, in step S1, the Z-score method is used to standardize the data, and the mean of each numerical data is subtracted and divided by its standard deviation, so that the processed data conforms to the standard normal distribution. The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0020] Preferably, in step S2, the defined feature vector includes the maximum load, minimum load, load standard deviation, load average, first timestamp and second timestamp of the air conditioner, and the temperature and humidity of the outdoor environment.

[0021] Preferably, a long short-term memory network (LSTM) model is used to build a load forecasting model, and the specific steps are as follows:

[0022] Initialize the parameters of the model. The initialization parameters include: initializing the structural parameters of the LSTM model, including the feature dimension of the input layer, the number of hidden layers and the number of hidden units in each layer, and the number of nodes in the output layer. The feature dimension of the input layer matches the number of extracted air conditioning load-related features, and the number of nodes in the output layer is equal to the predicted target number, that is, the predicted value of all air conditioning load states; initialize the model weights and biases, and use small random numbers for initialization; set the activation function of the LSTM model, including setting the Sigmoid function for the gating mechanism and setting the Tanh function for the calculation of candidate memory units and hidden states; configure the optimization algorithm, and set the learning rate parameters;

[0023] Iteratively train the selected model using the source domain dataset;

[0024] During the training process, the validation set is used regularly to evaluate the model performance until the performance of the model satisfies. After the training is completed, the weights and biases of the model are saved.

[0025] Preferably, during the model training process, the mean square error MSE algorithm is used to evaluate the performance of the load forecasting model. The specific algorithm is:

[0026]

[0027] Where n is the number of data points, Y pred,i is the load value predicted by the model, Y actual,i is the actual load value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

[0028] Preferably, in step S4, the data enhancement method includes data generation, data transformation and data mixing, and the implementation method includes:

[0029] Data generation: For each sample x in the air conditioning load dataset old , generate new data x by adding random numbers within a certain range new ;

[0030] Data transformation: For the label value y in the air conditioning load dataset old , generate a new label value y through mathematical transformation new , where y new =f(y old ), f represents the data transformation operation, including logarithmic transformation, exponential transformation and linear transformation, which is used to increase the adaptability of the model to different label value distributions;

[0031] Data mixing: Randomly select two samples x1 and x2 of different categories from the air conditioning load data set, and generate a new mixed sample z according to the set mixing ratio α. The calculation formula of z is:

[0032] z=αx1+(1-α)x2

[0033] The value of α ranges from 0 to 1.

[0034] Preferably, step S6 is implemented as follows:

[0035] Import the target domain dataset after data enhancement into the reinforcement learning model;

[0036] Initialize the parameters of the unfrozen layers in the reinforcement learning model, set the learning rate and number of iterations;

[0037] During the training process, the back-propagation algorithm is used to calculate the loss function value, and the parameters of the unfrozen layers are adjusted through the optimization algorithm to minimize the loss function;

[0038] Iterate the training until the preset stop condition is reached, including reaching the maximum number of iterations or the loss function value converges to the preset threshold;

[0039] Save the trained reinforcement learning model parameters for subsequent prediction of the optimal air conditioning temperature setting.

[0040] Preferably, an air conditioning load data control system comprises a data preprocessing layer, a feature extraction layer, a model training layer and a control execution layer;

[0041] The data preprocessing layer includes a data cleaning module and a timestamp definition and standardization module; the data cleaning module is used to clean the collected air conditioning meteorological parameters and load parameters; the timestamp definition and standardization module is used to timestamp the cleaned data and perform standardization;

[0042] The feature extraction layer includes a feature vector definition module and a combined feature generation module; the feature vector definition module is used to define a feature vector according to the preprocessed data; the combined feature generation module is used to generate several groups of combined features to form a feature data set;

[0043] The model training layer includes a load forecasting model construction module, a data enhancement module, a parameter migration module and a reinforcement learning module; the load forecasting model construction module uses a neural network algorithm to build a load forecasting model, and uses a feature data set to generate a source domain data set and a target domain data set, and uses the source domain data set to pre-train the load forecasting model; the data enhancement module is used to enhance the target domain data set to generate new data samples; the parameter migration module migrates the parameters of the load forecasting model to the model of the new task and freezes the parameters of some network layers; the reinforcement learning module obtains the initialized reinforcement learning model, uses the target domain data set to train the parameters of the unfrozen layers in the reinforcement learning model, and fine-tunes the model;

[0044] The control execution layer includes a real-time data acquisition module and an optimal setting output module; the real-time data acquisition module is used to acquire weather data in real time; the optimal setting output module uses the trained reinforcement learning model to give the optimal air-conditioning temperature setting to achieve load optimization control.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] Traditional machine learning models are limited by the scarcity of high-quality historical data. However, this invention uses a transfer learning strategy to pre-train the load forecasting model using the source domain data set, effectively alleviating the problem of insufficient data. Even if the initial data is limited, the model can acquire knowledge from related tasks through transfer learning, improve generalization ability, and enable the model to show good prediction performance in different scenarios.

[0047] The complexity and dynamics of air-conditioning system operation require the model to be highly robust and adaptable. The present invention enhances the target domain data through reinforcement learning to generate new data samples, which not only enriches the training set, but also improves the model's resistance to noise and outliers. In addition, the parameter migration and partial freezing strategies in transfer learning enable the model of new tasks to quickly adapt to specific scenarios, improving the practicality and flexibility of the model. The present invention uses the trained reinforcement learning model to give the optimal air-conditioning temperature setting according to weather data in real time, thereby achieving precise control of the load. This not only improves the user's comfort experience, but also significantly reduces energy consumption and operating costs, which is in line with the development trend of intelligent buildings and energy management systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a step diagram of the air conditioning load data control method of the present invention;

[0049] Figure 2 The flowchart for building and training the load forecasting model;

[0050] Figure 3 Flowchart for building and training a reinforcement learning model. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] See also Figure 1-3 The present invention provides a technical solution: an air conditioning load data control method, the method comprising the steps of:

[0053] S1. Data collection and preprocessing:

[0054] First, the meteorological parameters (such as outdoor temperature, humidity, solar radiation intensity, etc.) and load parameters (such as indoor temperature, humidity, air conditioning energy consumption, etc.) of the air conditioning system are collected through the intelligent building management system or a dedicated sensor network. The collected data needs to be cleaned to remove outliers and missing values ​​to ensure the accuracy and completeness of the data. Then, the data is timestamped for subsequent time series analysis. Finally, the data is standardized to eliminate the dimensional differences between different parameters and improve the training efficiency of the model.

[0055] S2. Feature Engineering:

[0056] Based on the preprocessed data, the feature vector is defined according to the actual needs of air conditioning load control. The feature vector should contain meteorological parameters and load parameters that have a significant impact on the air conditioning load. In order to capture more data features, several sets of combined features are generated, including the maximum load, minimum load, load standard deviation, load average, first timestamp and second timestamp of the air conditioner, as well as the temperature and humidity of the outdoor environment, to form a feature data set.

[0057] S3. Load forecasting model construction and pre-training:

[0058] A load forecasting model is constructed using a neural network algorithm. Using the feature data set generated in step S2, a source domain data set and a target domain data set are generated respectively. The source domain data set should contain historical data or data in similar scenarios for pre-training the load forecasting model. Through pre-training, the model can learn basic load forecasting rules, laying the foundation for subsequent transfer learning and reinforcement learning.

[0059] S4. Data enhancement:

[0060] In order to solve the problem of insufficient or low-quality data in the target domain, data enhancement technology is used to generate new data samples. Data enhancement can be achieved by simulating changes in meteorological parameters, adding noise, time shifting, etc. The enhanced data samples should be able to cover a wider range of meteorological conditions and load conditions, and improve the robustness and adaptability of the model.

[0061] S5. Model migration and initialization:

[0062] Transferring the parameters of the pre-trained load forecasting model to the model of the new task is called transfer learning. Transfer learning can retain useful knowledge in the source domain and reduce the reliance on a large amount of labeled data for the new task. In the model of the new task, the parameters of some network layers are frozen and kept unchanged to obtain the initialized reinforcement learning model. The reinforcement learning model is used to predict the optimal setting of the air conditioning temperature to achieve load optimization control.

[0063] S6. Model training and fine-tuning:

[0064] Use the target domain dataset to train the parameters of the unfrozen layers in the reinforcement learning model. Since the target domain data may be different from the source domain data, the model needs to be fine-tuned to adapt to the characteristics of the target domain. During the fine-tuning process, hyperparameters such as learning rate and number of iterations can be adjusted according to actual conditions to obtain the best model performance.

[0065] S7, real-time control and optimization:

[0066] After the model training is completed, weather data is obtained in real time (such as through weather stations, Internet weather services, etc.). The weather data is input into the trained reinforcement learning model, and the model will give the optimal air conditioning temperature setting. According to the output results of the model, the operating parameters of the air conditioning system are adjusted to achieve load optimization control. Through real-time control and optimization, the energy utilization efficiency of the air conditioning system can be further improved and the operating cost can be reduced.

[0067] Embodiment 1:

[0068] In step S1, data cleaning includes data missing value processing and data denoising:

[0069] Data missing value processing: When collecting meteorological parameters and load parameters of air conditioners, some data may be missing due to equipment failure, data transmission errors, etc. In order to maintain the integrity and continuity of the data, a linear interpolation algorithm is used to fill in the missing values.

[0070] The specific implementation is as follows:

[0071] First, we iterate through the entire dataset and identify the locations of all missing values.

[0072] For each missing value, determine the two nearest known values ​​before and after it, denoted as x1 and x2, and their corresponding timestamps t1 and t2. According to the linear interpolation formula:

[0073]

[0074] Where y is the fill value of the missing value position, and t is the timestamp corresponding to the missing value. Use the above formula to calculate the fill value of the missing value and insert it into the corresponding position in the data set.

[0075] Assume that in the data of a certain day, the outdoor temperature is missing at a certain time, and the known outdoor temperatures before and after are 25℃ and 27℃, respectively, and the corresponding timestamps are 14:00 and 16:00, respectively. If the timestamp corresponding to the missing value is 15:00, then according to the linear interpolation formula, the filling value of the missing value can be calculated as 26℃.

[0076] Data denoising processing:

[0077] The collected data may contain noise due to the influence of sensor accuracy, environmental factors, etc. In order to improve the accuracy and reliability of the data, a sliding average filter algorithm is used to denoise the data.

[0078] The specific implementation is as follows:

[0079] Set the size of the sliding window N, and choose it according to the fluctuation of the data and the denoising requirements. For example, for relatively stable data, you can choose a smaller N value; for data with large fluctuations, you can choose a larger N value.

[0080] For each data point x[n], calculate the average value of the N / 2 data points before and after it as the denoised data y[n]. The specific algorithm formula is:

[0081]

[0082] When n is less than N / 2 or greater than the total length of the data minus N / 2, the sliding window needs to be adjusted accordingly to ensure that the window always contains N data points.

[0083] Traverse the entire data set and perform the above calculations on each data point to obtain the denoised data set.

[0084] Assume that the size of the sliding window N is 5. For a data point x[n], the two data points before and after it are x[n-2], x[n-1], x[n+1], and x[n+2]. According to the sliding average filter algorithm formula, the denoised data y[n] can be calculated as the average value of these five data points.

[0085] In step S1, the Z-score method is used to standardize the data. This method aims to subtract the mean of each numerical data and divide it by its standard deviation, so that the processed data conforms to the standard normal distribution (mean is 0 and standard deviation is 1). The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0086] Embodiment 2:

[0087] This embodiment describes in detail the specific steps of building a load prediction model using a long short-term memory network (LSTM) model, aiming to accurately predict the air-conditioning load through the model.

[0088] Step 1: Initialize model parameters:

[0089] Initialization of structural parameters: First, according to the requirements of air conditioning load prediction, the structural parameters of the LSTM model are initialized. The feature dimension of the input layer is set to N, which matches the number of load-related features extracted from the air conditioning system. For example, it can include outdoor temperature, indoor temperature, humidity, personnel flow and other features. The number of hidden layers is set to L, and the number of hidden units in each layer is specified to M. The number of nodes in the output layer is set to P, which is equal to the target number of predictions, that is, the predicted value of all air conditioning load states, for example, it can be the prediction of air conditioning load in the future.

[0090] Weight and bias initialization: Use small random numbers to initialize the weights and biases of the LSTM model to prevent the model from falling into a local optimal solution at the beginning of training. Specifically, you can use Gaussian distribution or uniform distribution to generate small random numbers and assign these random numbers to the weights and biases of the model.

[0091] Activation function setting: Set the activation function of the LSTM model. For the gating mechanism, select the Sigmoid function, which can map the input value to between 0 and 1 to control the flow and blocking of information. For the calculation of candidate memory units and hidden states, select the Tanh function, which can map the input value to between -1 and 1 to enhance the nonlinear expression ability of the model.

[0092] Optimization algorithm configuration: Configure optimization algorithms, such as Adam, RMSprop, etc., to update the weights and biases of the model during training. At the same time, set appropriate learning rate parameters to control the update step size of the model in each iteration.

[0093] Step 2: Model training:

[0094] Dataset preparation: Prepare a source domain dataset, which should contain enough historical air conditioning load data and related feature data. Divide the dataset into training set, validation set, and test set for model training, validation, and testing.

[0095] Iterative training: The LSTM model is iteratively trained using the training set. In each iteration, the training data is input into the model, the error between the predicted value and the true value is calculated, and the weights and biases of the model are updated according to the error. This process is repeated until the set number of iterations is reached or other stopping conditions are met.

[0096] Performance evaluation: During the training process, the validation set is used regularly to evaluate the performance of the model. The mean square error (MSE) indicator can be used to measure the prediction accuracy of the model. The algorithm for the mean square error MSE is:

[0097]

[0098] Where n is the number of data points, Ypred,i is the load value predicted by the model, Y actual,i is the actual load value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

[0099] If the performance of the model meets the preset requirements, stop training; otherwise, continue to adjust the model parameters and training strategy.

[0100] Step 3: Save the model:

[0101] Save weights and biases: After training is complete, save the weights and biases of the LSTM model. These parameters are the knowledge learned by the model during training and are crucial for subsequent load forecasting. They can be saved to a file or database so that they can be loaded and used when needed.

[0102] The LSTM model can effectively capture long-term dependencies in sequence data through its unique gating mechanism (input gate, forget gate, output gate) and memory units. This is particularly important for load forecasting, because load data is often affected by multiple factors that may change at different time scales. LSTM can remember key information from the past and combine it with current input to make more accurate predictions.

[0103] Embodiment 3:

[0104] The present invention relates to a data enhancement step in an air conditioning load data control method. Specific embodiments of data generation, data transformation and data mixing will be described in detail below with reference to specific examples.

[0105] Data generation: Data generation refers to adding random noise to numerical data, performing linear transformation, etc., to generate new data samples from the original data. Taking air conditioning load data as an example, suppose there is a set of numerical data including outdoor temperature, humidity, and air conditioning load. In order to increase the diversity of the data, a small range of random numbers can be added to each data point to generate new data samples.

[0106] Example:

[0107] Original data sample: x old =[25.0,60.0,3.0] (outdoor temperature 25.0℃, humidity 60.0%, air conditioning load 3.0kW).

[0108] Add random number: Add a random number between -0.5 and 0.5 to each element.

[0109] Generated new data sample: x new =[24.8,60.2,3.1] or x new=[25.3,59.7,2.9], etc.

[0110] Data transformation: Data transformation refers to the mathematical transformation of the label value of the original data (if any), such as logarithmic transformation, exponential transformation, linear transformation, etc., to generate new label values. This helps the model adapt to different label value distributions and improves the accuracy of prediction.

[0111] Example:

[0112] Original data samples and their labels: x old =[25.0,60.0,3.0], label y old =1 (indicating a certain load state).

[0113] Logarithmic transformation of labels: y new =log(y old +1)=log(2)≈0.693.

[0114] Generated new data samples and their transformed labels: x new =[25.0,60.0,3.0], label y new =0.693.

[0115] Data mixing: Data mixing refers to randomly selecting samples of two different categories from the data set and generating new mixed samples according to the set mixing ratio. This helps to simulate complex data situations that may arise in actual scenarios and improve the generalization ability of the model.

[0116] Example:

[0117] Select two data samples of different categories: x1 = [25.0, 60.0, 3.0] (category 1) and x2 = [30.0, 70.0, 5.0] (category 2).

[0118] The mixing ratio α was set to 0.6.

[0119] Generate mixed samples: z = αx1 + (1-α)x2

[0120] =0.6×[25.0,60.0,3.0]+0.4×[30.0,70.0,5.0]=[27.0,64.0,3.8].

[0121] The value of α ranges from 0 to 1.

[0122] Step S6 imports the target domain data set after data enhancement processing into the reinforcement learning model to train the model and optimize the parameters to achieve the prediction of the optimal setting of the air conditioning temperature. The specific implementation steps include:

[0123] Data import and preprocessing: Import the target domain dataset after data enhancement into the reinforcement learning model. The dataset should contain enough historical data of air conditioning temperature settings and related feature data, such as outdoor temperature, indoor temperature, humidity, personnel flow, time, etc. Before importing the data, the data needs to be preprocessed, including data cleaning, format conversion, normalization and other operations to ensure the accuracy and consistency of the data.

[0124] Model initialization: Initialize the parameters of the unfrozen layers in the reinforcement learning model. In transfer learning, it is usually necessary to freeze the parameters of some layers to retain the useful information in the pre-trained model, and only train the parameters of the unfrozen layers to adapt to the new task. Therefore, it is necessary to set the initial parameters of the unfrozen layers, which can be randomly initialized or fine-tuned based on the parameters of the pre-trained model. At the same time, set the learning rate and number of iterations to control the training process and update step size of the model.

[0125] Model training: During the training process, the back propagation algorithm is used to calculate the loss function value. The back propagation algorithm is a classic algorithm for training neural networks. It calculates the error between the output value and the true value and propagates the error back to the network to update the network weights and biases. In this embodiment, a suitable loss function (such as mean square error, cross entropy, etc.) is selected to measure the prediction performance of the model, and the parameters of the unfrozen layers are adjusted through optimization algorithms (such as gradient descent, Adam, etc.) to minimize the loss function.

[0126] Stop condition judgment: During the iterative training process, it is necessary to constantly judge whether the preset stop condition is reached. These conditions may include reaching the maximum number of iterations or the loss function value converging to the preset threshold. When one of the conditions is met, training is stopped to avoid overfitting of the model or wasting computing resources.

[0127] Model saving: Save the parameters of the trained reinforcement learning model. These parameters are the knowledge learned by the model during the training process, which are crucial for the subsequent prediction of the optimal air conditioning temperature setting. They can be saved to a file or database so that they can be loaded and used when needed.

[0128] In order to realize the above-mentioned air conditioning load data control method, an air conditioning load data control system is designed, including a data preprocessing layer, a feature extraction layer, a model training layer and a control execution layer;

[0129] The data preprocessing layer includes a data cleaning module and a timestamp definition and standardization module; the data cleaning module is used to clean the collected air conditioning meteorological parameters and load parameters; the timestamp definition and standardization module is used to timestamp the cleaned data and perform standardization;

[0130] The feature extraction layer includes a feature vector definition module and a combined feature generation module; the feature vector definition module is used to define a feature vector according to the preprocessed data; the combined feature generation module is used to generate several groups of combined features to form a feature data set;

[0131] The model training layer includes a load forecasting model construction module, a data enhancement module, a parameter migration module and a reinforcement learning module; the load forecasting model construction module uses a neural network algorithm to build a load forecasting model, and uses a feature data set to generate a source domain data set and a target domain data set, and uses the source domain data set to pre-train the load forecasting model; the data enhancement module is used to enhance the target domain data set to generate new data samples; the parameter migration module migrates the parameters of the load forecasting model to the model of the new task and freezes the parameters of some network layers; the reinforcement learning module obtains the initialized reinforcement learning model, uses the target domain data set to train the parameters of the unfrozen layers in the reinforcement learning model, and fine-tunes the model;

[0132] The control execution layer includes a real-time data acquisition module and an optimal setting output module; the real-time data acquisition module is used to acquire weather data in real time; the optimal setting output module uses the trained reinforcement learning model to give the optimal air-conditioning temperature setting to achieve load optimization control.

[0133] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0134] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The air conditioning load data control method is characterized in that: The method comprises: S1: Collect meteorological parameters and load parameters of air conditioners, and clean, timestamp and standardize the collected data; S2: Define feature vectors based on the collected data and generate several sets of combined features to form a feature data set; S3: Use a neural network algorithm to build a load forecasting model, use the feature data set obtained in step S2 to generate a source domain data set and a target domain data set, and use the source domain data set to pre-train the load forecasting model; S4: Enhance the data in the target domain and generate new data samples; S5: Migrating the parameters of the load forecasting model to the model of the new task, freezing the parameters of some network layers, and obtaining an initialized reinforcement learning model; the reinforcement learning model is used to predict the optimal setting of the air conditioning temperature; S6: Use the target domain dataset to train the parameters of the unfrozen layers in the reinforcement learning model and fine-tune the model; S7: Obtain weather data in real time and use the trained reinforcement learning model to give the optimal air conditioning temperature setting to achieve load optimization control.

2. The air conditioning load data control method according to claim 1, characterized in that: In step S1, data cleaning methods include data missing value processing and data denoising: For missing values, linear interpolation algorithm is used to fill them; For noisy data, the sliding average filtering algorithm is used for data denoising, and the algorithm formula is: Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.

3. The air conditioning load data control method according to claim 1, characterized in that: In step S1, the Z-score method is used to standardize the data. The mean of each numerical data is subtracted and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Among them, X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

4. The air conditioning load data control method according to claim 1, characterized in that: In step S2, the defined feature vector includes the maximum load, minimum load, load standard deviation, load average, first timestamp and second timestamp of the air conditioner, and the temperature and humidity of the outdoor environment.

5. The air conditioning load data control method according to claim 1, characterized in that: The long short-term memory network LSTM model is used to build a load forecasting model. The specific steps are as follows: Initialize the parameters of the model. The initialization parameters include: initializing the structural parameters of the LSTM model, including the feature dimension of the input layer, the number of hidden layers and the number of hidden units in each layer, and the number of nodes in the output layer. The feature dimension of the input layer matches the number of extracted air conditioning load-related features, and the number of nodes in the output layer is equal to the predicted target number, that is, the predicted value of all air conditioning load states; initialize the model weights and biases, and use small random numbers for initialization; set the activation function of the LSTM model, including setting the Sigmoid function for the gating mechanism and setting the Tanh function for the calculation of candidate memory units and hidden states; configure the optimization algorithm, and set the learning rate parameters; Iteratively train the selected model using the source domain dataset; During the training process, the validation set is used regularly to evaluate the model performance until the performance of the model satisfies. After the training is completed, the weights and biases of the model are saved.

6. The air conditioning load data control method according to claim 5, characterized in that: During the model training process, the mean square error (MSE) algorithm is used to evaluate the performance of the load forecasting model. The specific algorithm is as follows: Where n is the number of data points, Y pred，i is the load value predicted by the model, Y actual，i is the actual load value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

7. The air conditioning load data control method according to claim 1, characterized in that: In step S4, the data enhancement methods include data generation, data transformation and data mixing, and the implementation methods include: Data generation: For each sample x in the air conditioning load dataset old , generate new data x by adding random numbers within a certain range new ; Data transformation: For the label value y in the air conditioning load dataset old , generate a new label value y through mathematical transformation new , where y new =f(y old ), f represents the data transformation operation, including logarithmic transformation, exponential transformation and linear transformation, which is used to increase the adaptability of the model to different label value distributions; Data mixing: Randomly select two samples x1 and x2 of different categories from the air conditioning load data set, and generate a new mixed sample z according to the set mixing ratio α. The calculation formula of z is: z=αx1+(1-α)x2 The value of α ranges from 0 to 1.

8. The air conditioning load data control method according to claim 7, characterized in that: Step S6 is implemented as follows: Import the target domain dataset after data enhancement into the reinforcement learning model; Initialize the parameters of the unfrozen layers in the reinforcement learning model, set the learning rate and number of iterations; During the training process, the back-propagation algorithm is used to calculate the loss function value, and the parameters of the unfrozen layers are adjusted through the optimization algorithm to minimize the loss function; Iterate the training until the preset stop condition is reached, including reaching the maximum number of iterations or the loss function value converges to the preset threshold; Save the trained reinforcement learning model parameters for subsequent prediction of the optimal air conditioning temperature setting.

9. An air conditioning load data control system, characterized in that: It includes data preprocessing layer, feature extraction layer, model training layer and control execution layer; The data preprocessing layer includes a data cleaning module and a timestamp definition and standardization module; the data cleaning module is used to clean the collected air conditioning meteorological parameters and load parameters; the timestamp definition and standardization module is used to timestamp the cleaned data and perform standardization; The feature extraction layer It includes a feature vector definition module and a combined feature generation module; The feature vector definition module is used to define feature vectors according to the preprocessed data; the combined feature generation module is used to generate several groups of combined features to form a feature data set; The model training layer includes a load forecasting model building module, a data enhancement module, a parameter migration module and an enhanced learning module; The load forecasting model construction module uses a neural network algorithm to build a load forecasting model, and uses a feature data set to generate a source domain data set and a target domain data set, and uses the source domain data set to pre-train the load forecasting model; the data enhancement module is used to enhance the target domain data set to generate new data samples; the parameter migration module migrates the parameters of the load forecasting model to the model of the new task and freezes the parameters of some network layers; The reinforcement learning module obtains the initialized reinforcement learning model, trains the parameters of the unfrozen layers in the reinforcement learning model using the target domain dataset, and fine-tunes the model; The control execution layer includes a real-time data acquisition module and an optimal setting output module; The real-time data acquisition module is used to acquire weather data in real time; the optimal setting output module uses the trained reinforcement learning model to give the optimal air-conditioning temperature setting to achieve load optimization control.

Images