Method and system for predicting industrial cooling load
Through deep learning-based methods, the prediction of industrial cooling loads is solved by using LSTM or Transformer models, and the problems of insufficient prediction accuracy and long calculation time in the prior art are solved, and accurate prediction and rapid calculation of cooling loads are achieved.
Patent Information
- Application Number
- CN202411566165.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-05
- Publication Date
- 2025-02-14
AI Technical Summary
The existing cold load prediction methods have limitations when dealing with complex nonlinear relationships, resulting in insufficient prediction accuracy, difficulty in adapting to variable environmental parameters, and high requirements for data volume and calculation time, making it difficult to meet the actual needs of the HVAC industry.
Using a deep learning-based method, the LSTM or Transformer model is trained by collecting data on dry bulb temperature, wet bulb temperature and relative humidity to predict the cooling load size and distribution of the next day. This method reduces sensitivity to data volume and data types through data standardization, feature engineering and optimization of training parameters, and improves calculation speed and generalization capabilities.
Accurate prediction of industrial cooling loads is realized, sensitivity to data volume and data types is reduced, computing speed and generalization ability is improved, and problems of insufficient accuracy and long calculation time in traditional prediction methods are solved.
Smart Images
Figure CN119443696A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of refrigeration control, and in particular relates to a method and a system for predicting industrial cooling load. Background Art
[0002] In modern building energy management systems, accurate prediction of air conditioning cooling load is of great significance for achieving energy conservation and consumption reduction and optimizing resource allocation. Accurate cooling load prediction can obtain more accurate energy efficiency reports at the beginning of the design, and make targeted in-depth optimization possible. It can further adjust the pipeline structure, select equipment that matches the high-efficiency range, and have more accurate control over redundancy, thereby improving design quality and reducing initial investment.
[0003] The existing load forecasting methods can be roughly divided into two types. One is the built-in empirical formula of the load simulation software, which calculates the wall heat transfer, weather, sunshine and other conditions to obtain the cooling load, which is called the "white box model". The other is predicted through the historical data of the main parameters. The first white box model requires detailed building data. When the building data is insufficient, it is mainly supplemented by the experience of engineers, which is not enough to provide an accurate reference for the actual situation. Especially in renovation projects, China is in the deep waters of industrial transformation and upgrading. Renovation projects occupy a considerable scale in the HVAC industry. However, after more than ten years of use, a lot of building information has been lost and may be significantly different from the original design. At this time, the load forecasting model obtained by the white box model is not credible. Another approach in the white box model is to use algorithms such as particle swarm neural network algorithms and rough set neural network algorithms. These algorithms have different requirements for the size of data, calculation speed, and are highly dependent on parameter adjustment and engineering experience. In the second mode, a variety of algorithms appear. For example, according to the extreme values of historical data, the prediction results are obtained by interpolation in the form of months and times. This approach is more empirical and generally adaptable to future loads.
[0004] It can be seen that the existing cooling load prediction methods are mainly based on traditional statistical models, such as multiple regression analysis, time series analysis, etc. However, these models have significant limitations when dealing with complex nonlinear relationships, resulting in insufficient prediction accuracy, difficulty in adapting to changing environmental parameters, and prediction distortion. In addition, these current methods usually require massive historical data and extremely long calculation time. The valid data collected in the HVAC industry usually appears in the form of 8760 hours. The hundreds of thousands of orders of magnitude of data required by traditional models cannot be met at all. The use of accumulated data from many years ago cannot truly reflect the actual situation due to problems such as equipment aging. Summary of the invention
[0005] The present invention aims to provide a method and system for predicting industrial cooling load, so as to reduce the sensitivity to data volume and data type in load prediction through the method for predicting industrial cooling load, and ensure faster calculation speed and stronger generalization ability in HVAC cooling load prediction.
[0006] To achieve the above objectives, the present invention adopts the following technical solutions:
[0007] A method for predicting industrial cooling load is provided, comprising:
[0008] Collect data related to three environmental characteristics: dry-bulb temperature, wet-bulb temperature and relative humidity over a period of time;
[0009] The collected data is preprocessed and feature engineered, and then the deep learning model is trained using the preprocessed and feature engineered data. The trained deep learning model is used to predict the size and distribution of the cooling load for the next day.
[0010] Furthermore, the deep learning model is an LSTM model or a Transformer model.
[0011] Furthermore, the preprocessing and feature engineering of the collected data includes:
[0012] Data standardization: The collected data were normalized using the standard deviation standardization method, and the sliding time window technology was applied to retain the data of the first 30 time steps;
[0013] Dataset division and enhancement: The data set is divided reasonably, 20% of the data is reserved as the test set, and the data set is randomly shuffled. The diversity of the data set is further expanded through data enhancement technology.
[0014] Furthermore, the training of the deep learning model using the preprocessed and feature engineered data includes:
[0015] Training parameter configuration: Learning rate: Set the initial learning rate to 0.001; Number of training rounds: Set the maximum number of model training rounds to 10,000, and set an early stopping strategy to stop training when the model is no longer optimized in 200 consecutive rounds of training; Batch size: The batch size for training and testing is 1,000; Optimizer selection: Use the Adam optimizer for model training;
[0016] Set model checkpoints and log storage: Set model storage checkpoints. When the loss function optimization of the model on the validation set exceeds the set threshold, the current optimal model is automatically stored; and the training log is saved.
[0017] The present invention also provides a system for predicting industrial cooling load, comprising:
[0018] Data acquisition module: used to collect data related to the three environmental characteristics of dry bulb temperature, wet bulb temperature and relative humidity over a period of time;
[0019] Data processing and model training prediction module: used to preprocess and feature engineer the collected data, then use the preprocessed and feature engineered data to train the deep learning model, and use the trained deep learning model to predict the size and distribution of the cooling load for the next day.
[0020] Furthermore, the deep learning model is an LSTM model or a Transformer model.
[0021] Furthermore, the data processing and model training prediction module includes:
[0022] Data standardization processing unit: used to normalize the collected data using the standard deviation standardization method, and apply the sliding time window technology to retain the data of the first 30 time steps;
[0023] Dataset division and enhancement unit: used to reasonably divide the data set, retain 20% of the data as the test set, randomly shuffle the data set, and further expand the diversity of the data set through data enhancement technology;
[0024] The training parameter configuration unit is used to configure the following parameters: learning rate: set the initial learning rate to 0.001; number of training rounds: set the maximum number of model training rounds to 10000, and set the early stopping strategy to stop training when the model is no longer optimized in 200 consecutive rounds of training; batch size: the batch size for training and testing is 1000; optimizer selection: use the Adam optimizer for model training;
[0025] Model checkpoint and log saving setting unit: used to set model storage checkpoints. When the loss function optimization of the model on the validation set exceeds the set threshold, the current optimal model is automatically stored; and the training log is saved.
[0026] The present invention also provides the application of any of the above-mentioned methods for predicting industrial cooling load in energy efficiency control and selection of building refrigeration equipment and preparation of energy efficiency scheme for building refrigeration equipment.
[0027] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the method for predicting industrial cooling load according to any one of claims 1 to 4 when executing the computer program.
[0028] The present invention also provides a computer-readable storage medium, in which a computer program is stored. The computer program is executed by a processor to implement the method for predicting industrial cooling load according to any one of claims 1 to 4.
[0029] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The method for predicting industrial cooling load uses the three key environmental features of dry bulb temperature, wet bulb temperature and relative humidity to predict the cooling load of refrigeration equipment based on LSTM or Transformer model. The method can accurately predict the cooling load of buildings by designing a highly customized deep learning model and optimizing training and feature engineering. The method is not sensitive to the amount and type of data, and has a fast calculation speed and strong generalization ability in HVAC cooling load prediction, and solves the problems of insufficient accuracy and difficulty in handling complex environmental parameters in traditional prediction methods. 2. In the method for predicting industrial cooling load, the LSTM model can be used, and the Transformer model can also be used. Different models provide different implementation solutions for cooling load prediction to meet different application requirements. In particular, the LSTM model is applied to the prediction of air conditioning cooling load, and the LSTM model has good effectiveness and accuracy in air conditioning load prediction. 3. Compared with the existing cooling load prediction technology, all the data required for this method of predicting industrial cooling load can be collected by sensors, without the need for project building information and manual experience estimation. As most projects can only provide 8760 hours of cooling load data, this method does not require a high amount of data. In actual measurements, the calculation speed is fast and the generalization ability and prediction accuracy are high. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0031] Figure 1 The flowchart of an embodiment of the method for predicting industrial cooling load according to the present invention.
[0032] Figure 2 This is a flow chart of preprocessing and feature engineering of collected data in one embodiment of the method for predicting industrial cooling load of the present invention.
[0033] Figure 3 This is a flowchart of training a deep learning model using preprocessed and feature engineered data in one embodiment of a method for predicting industrial cooling load according to the present invention.
[0034] Figure 4 This is a chart of data collected for prediction in an embodiment of a method for predicting industrial cooling load according to the present invention.
[0035] Figure 5 This is a statistical chart of predicted values and actual values in an embodiment of a method for predicting industrial cooling load according to the present invention.
[0036] Figure 6 This is a structural block diagram of an embodiment of a system for predicting industrial cooling load according to the present invention.
[0037] Figure 7 This is a structural block diagram of a data processing and model training prediction module in an embodiment of a system for predicting industrial cooling load according to the present invention.
[0038] Figure 8 It is a structural schematic diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0039] 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.
[0040] In one embodiment, a method for predicting industrial cooling load is provided. Figure 1 As shown, the method for predicting industrial cooling load includes the following steps:
[0041] Step S100: collecting data related to three environmental characteristics: dry-bulb temperature, wet-bulb temperature and relative humidity over a period of time.
[0042] Step S200: preprocess and feature engineer the collected data, then use the preprocessed and feature engineered data to train a deep learning model, and use the trained deep learning model to predict the size and distribution of the cooling load for the next day.
[0043] The dry-bulb temperature, wet-bulb temperature and relative humidity collected in step S100 are parameters collected by sensors of the refrigeration equipment, and these parameters are directly related to the cooling load of the refrigeration equipment.
[0044] In step S200, the deep learning model used in the method for predicting industrial cooling load is the LSTM model, which is a long short-term memory artificial neural network model. The model has a strong ability to process time series data. The network architecture of the LSTM model has significant advantages in processing time series problems and can capture the long-term dependencies between input features. The specific configuration of the LSTM model is as follows:
[0045] Input layer: The input layer receives three key environmental features, namely dry-bulb temperature, wet-bulb temperature, and relative humidity (input_size=3).
[0046] Hidden layer: The hidden layer consists of an LSTM layer with 64 hidden units (hidden_size=64), which can effectively capture the feature changes in the time series. The number of LSTM layers is 1 (num_layers=1). By controlling the number of layers, the model can strike a balance between complexity and computational efficiency.
[0047] Output layer: The output of the hidden layer is converted into the final cooling load prediction value through a fully connected layer. The multi-layer output of LSTM is flattened and passed to the linear layer for the final regression calculation.
[0048] The LSTM model design architecture can make full use of the advantages of the LSTM network in processing sequence data, effectively solve the problems of complex nonlinear relationships and strong interdependence between features in cooling load prediction, and can be used to accurately predict cooling loads to optimize the operating efficiency of building energy management systems.
[0049] In other embodiments, in step S200, the deep learning model used in the method for predicting industrial cooling load may also adopt a Transformer model. The Transformer model better captures global dependencies through a multi-head attention mechanism and is suitable for capturing long-term dependencies in time series data. Among them, by setting a multi-head attention mechanism (nhead=3), the model can process information in parallel in different subspaces and enhance the ability to capture key features; the encoder layer (num_layers=2) of the Transformer model can efficiently process sequence data and output the cooling load prediction results through a fully connected layer. The Transformer model provides a different prediction perspective from the LSTM model, and is particularly suitable for data sets with long-term dependency features. Therefore, the Transformer model is suitable for application in cooling load prediction.
[0050] Further, in another embodiment, Figure 2 As shown, in step S200, preprocessing and feature engineering of the collected data includes:
[0051] Step S101, data standardization: normalize the collected data using the standard deviation standardization method, and apply the sliding time window technology to retain the data of the first 30 time steps.
[0052] Step S102, data set division and enhancement: divide the data set reasonably, retain 20% of the data as a test set, randomly shuffle the data set, and further expand the diversity of the data set through data enhancement technology.
[0053] The purpose of fine-grained preprocessing and feature engineering of input data is to ensure that the input data of the model can effectively reflect the environmental characteristics. In the data preprocessing stage, the standard deviation normalization method (SCALER_TYPE='std') is used to normalize the input features. The normalization process can eliminate the differences between different feature dimensions, making the model training more stable and converging faster; the sliding time window technology (WINDOW_SIZE=30) used in the preprocessing process retains the data of the first 30 time steps, so that the model can make full use of historical information for prediction. The data feature engineering here is to divide and enhance the data set. Before model training, the data set is reasonably divided, and 20% of the data is retained as a test set (TEST_SIZE=0.2) to ensure the generalization ability of the model. In addition, in order to prevent overfitting, the data set has been randomly shuffled (SHUFFLE=True), and the diversity of the data set has been further expanded through data enhancement technology.
[0054] Further, in another embodiment, Figure 3 As shown, in the step S200, training the deep learning model with the preprocessed and feature engineered data includes:
[0055] Step S203, training parameter configuration: learning rate: set the initial learning rate to 0.001; number of training rounds: the maximum number of model training rounds is set to 10000, and an early stopping strategy is set to stop training when the model is no longer optimized in 200 consecutive rounds of training; batch size: the batch size during training and testing is 1000; optimizer selection: use Adam optimizer for model training.
[0056] Step S204, setting model checkpoint and log saving: setting model storage checkpoint, when the loss function optimization of the model on the validation set exceeds the set threshold, automatically storing the current optimal model; and saving the training log.
[0057] Through model training, the hyperparameters are carefully tuned and optimized to ensure the efficiency and accuracy of the model in the cold load prediction task. In the training parameter configuration, the initial learning rate is set to 0.001 (LR = 1e-3) to ensure that the model can converge robustly during the training process. The maximum number of training rounds of the model is set to 10000 (EPOCHS = 10000), and the early stopping strategy (when the model is no longer optimized in 200 consecutive rounds of training) is used to prevent overfitting. The batch size during training and testing is set to 1000 (TRAIN_BATCH_SIZE = 1000, TEST_BATCH_SIZE = 1000), which balances the training speed and memory consumption. The batch size here refers to the number of samples used in each iteration during the training model. The Adam optimizer is used for model training. The Adam optimizer has the characteristics of adaptive learning rate, which can accelerate the convergence of the model and avoid falling into the local optimal solution.
[0058] To ensure the effectiveness of model training, a model storage checkpoint is set here. When the loss function optimization of the model on the validation set exceeds the set threshold (DELTA=0), the current optimal model will be automatically stored. In addition, by saving the training log (LOG_SAVE_PATH='logs'), users can review and analyze the model training process at any time to further optimize the model performance.
[0059] It can be seen from the above embodiments that the method for predicting industrial cooling load proposes a prediction method that is not sensitive to data volume and data type, has a fast calculation speed, and has strong generalization ability in HVAC cooling load prediction. Among them, the method for predicting industrial cooling load is an air conditioning cooling load prediction method based on LSTM (Long Short-Term Memory Network). This method accurately predicts the cooling load of a building by using three key environmental characteristics: dry bulb temperature, wet bulb temperature, and relative humidity. This method solves the problems of insufficient accuracy, long response time, and difficulty in handling complex environmental parameters in traditional prediction methods by designing a highly customized deep learning model and optimizing training and feature engineering.
[0060] In this method of predicting industrial cooling load, the LSTM model or the Transformer model can be used. Different models provide different implementation schemes for cooling load prediction to meet different application requirements. In particular, the LSTM model is applied to the prediction of air conditioning cooling load. The research results show that the LSTM model has achieved excellent results with an average relative error of 0.03 in cooling load prediction, which shows the effectiveness and accuracy of the model in air conditioning load prediction.
[0061] Compared with the existing cooling load prediction technology, the data required for this method of industrial cooling load prediction can all be collected by sensors, without the need for project building information and manual experience estimation. It adopts a long-term and short-term artificial "black box model". For most projects that can only provide 8760 hours of cooling load data, this method does not require a high amount of data. Figure 4 and Figure 5 As shown, Figure 4 The data in this paper are the actual operating data of the refrigeration equipment in an electronic factory in Shenzhen. After the sensor data is recorded in the background, it is learned by the algorithm compiled based on the method of predicting industrial cooling load. Figure 5 As shown, Figure 5 For the cooling load prediction results, the results after learning prediction are compared with the actual operation results. The prediction error is 3.56%, and this error includes the irregular start and stop operation of the equipment. In the actual measurement, the calculation speed only takes 190 seconds, which shows that its generalization ability and prediction accuracy are high.
[0062] In some embodiments, the method for predicting industrial cooling load in the previous embodiment is provided for application in the selection of building refrigeration equipment and the preparation of building refrigeration equipment energy efficiency plan. The specific application scenarios are as follows:
[0063] 1. Application of the method for predicting industrial cooling load in energy efficiency control of building refrigeration equipment:
[0064] This method for predicting industrial cooling load can be applied to the application scenario of water storage / ice storage energy efficiency control. It can predict the size and distribution of the cooling load of the next day by collecting the data of the water storage / ice storage equipment in the past week, and help the upper computer energy-saving system to control the storage time or adjust the storage rate during nighttime storage to achieve the purpose of efficient operation of the equipment. The effect of peak load shaving and valley filling can be achieved throughout the year, reducing operating electricity costs. Especially in the transition season and winter, less cold storage can be used to ensure sufficient cooling capacity and low comprehensive electricity prices for the next day without waste, thereby improving energy efficiency throughout the year. According to evaluation, the use of a cold storage algorithm based on this method for predicting industrial cooling load can save up to millions of electricity bills each year in high-energy consumption industries (such as lithium batteries, semiconductor chip factories, large chemical cooling facilities, smelting plants, large commercial centers, etc.).
[0065] 2. Application of this method for predicting industrial cooling load in the selection of building refrigeration equipment:
[0066] This application scenario is mainly aimed at the early equipment selection in the building refrigeration equipment renovation project. This method can use the data of the old refrigeration equipment used in the renovation project to provide accurate cooling load data in the building refrigeration equipment renovation design stage, helping users and engineers understand the load distribution throughout the year, determine the equipment high efficiency range according to different load frequencies, and reasonably select the corresponding chillers and water pumps to determine the pipe diameter, improve energy efficiency, reduce initial investment, and deeply arrange various equipment and pipelines, optimize the use area of refrigeration equipment, and reduce space waste. According to evaluation, for large-scale renovation projects, it can save tens of millions of labor and material costs while ensuring energy efficiency.
[0067] 3. Application of this method for predicting industrial cooling load in the preparation of energy efficiency plans for building refrigeration equipment:
[0068] This method can predict the energy efficiency plan of the refrigeration equipment of the renovation project. Specifically, accurate load information can be obtained based on the 8760 hours of historical data of the old refrigeration equipment in the past year. Based on this load information and the upper computer optimization algorithm model, efficient instructions for future control can be obtained to determine the expected energy efficiency of the refrigeration room throughout the year. It does not rely on manual experience and is of great help to reasonable bidding assessment thresholds and acceptance methods.
[0069] This method for predicting industrial cooling load has shown significant advantages in multiple practical application scenarios, especially in the water storage / ice storage transition season scenario, early equipment selection and energy efficiency plan production. According to the algorithm prediction of this method for predicting industrial cooling load, the daily demand cooling load can be quickly determined to avoid waste in the storage of cooling capacity, and truly achieve "peak shaving and valley filling" and "making the best use of resources" for electric energy. This method for predicting industrial cooling load significantly improves the generalization ability and prediction accuracy of the model by adopting advanced feature engineering technologies such as time window sliding, standard deviation standardization, and data enhancement. By setting reasonable training parameters and optimization strategies, the performance of the model in high-precision cooling load prediction is ensured, which is extremely practical. In addition, this method can match the scale of industry data and has strong application value.
[0070] In one embodiment, a system for predicting industrial cooling load is provided. Figure 4As shown, the system includes a data acquisition module 300 and a data processing and model training prediction module 400, wherein the data acquisition module 300 is used to collect three environmental characteristic-related data of the building's dry-bulb temperature, wet-bulb temperature and relative humidity over a period of time in the past, as well as the corresponding cooling load data; the data processing and model training prediction module 400 is used to preprocess and feature engineer the collected data, and then use the preprocessed and feature engineered data to train the deep learning model, and use the trained deep learning model to predict the size and distribution of the cooling load for the next day. The deep learning model used in the data processing and model training prediction module 400 here is an LSTM model or a Transformer model.
[0071] Furthermore, in one embodiment, the data processing and model training prediction module 400 includes a data standardization processing unit 401, a data set division and enhancement unit 402, a training parameter configuration unit 403, and a model checkpoint and log saving setting unit 404, wherein the data standardization processing unit 401 is used to normalize the collected data using a standard deviation normalization method, and apply a sliding time window technology to retain the data of the first 30 time steps; the data set division and enhancement unit 402 is used to reasonably divide the data set, retain 20% of the data as a test set, and randomly shuffle the data set, and further use data enhancement technology to further step to expand the diversity of the data set; the training parameter configuration unit 403 is used to configure the following parameters: learning rate: set the initial learning rate to 0.001; number of training rounds: the maximum number of model training rounds is set to 10000, and an early stopping strategy is set to stop training when the model is no longer optimized in 200 consecutive rounds of training; batch size: the batch size during training and testing is 1000; optimizer selection: the Adam optimizer is used for model training; the model checkpoint and log saving setting unit 404 is used to set the model storage checkpoint. When the loss function optimization of the model on the validation set exceeds the set threshold, the current optimal model is automatically stored; and the training log is saved.
[0072] In one embodiment, an electronic device is provided, Figure 8 As shown, it includes a processor 10 and a memory 20, and is connected via a bus, wherein the memory 20 stores a computer program, including at least one instruction, at least one program, a code set or an instruction set, and the computer program is loaded and executed by the processor to implement the method for predicting industrial cooling load as in the previous embodiment.
[0073] In one embodiment, a computer storage medium is provided, in which computer executable instructions are stored, and the computer executable instructions are used to execute the method for predicting industrial cooling load in the previous embodiment. The storage medium here includes: a mobile storage device, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
[0074] It should be noted that, in this article, terms such as "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes 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 apparatus.
[0075] 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. A method for predicting industrial cooling load, characterized in that: include: Collect data related to three environmental characteristics: dry-bulb temperature, wet-bulb temperature and relative humidity over a period of time; The collected data is preprocessed and feature engineered, and then the deep learning model is trained using the preprocessed and feature engineered data. The trained deep learning model is used to predict the size and distribution of the cooling load for the next day.
2. The method for predicting industrial cooling load according to claim 1, characterized in that: The deep learning model is an LSTM model or a Transformer model.
3. The method for predicting industrial cooling load according to claim 2, characterized in that: The preprocessing and feature engineering design of the collected data includes: Data standardization: The collected data were normalized using the standard deviation standardization method, and the sliding time window technology was applied to retain the data of the first 30 time steps; Dataset division and enhancement: The data set is divided reasonably, 20% of the data is reserved as the test set, and the data set is randomly shuffled. The diversity of the data set is further expanded through data enhancement technology.
4. The method for predicting industrial cooling load according to claim 3, characterized in that: The method of training a deep learning model using preprocessed and feature engineered data includes: Training parameter configuration: Learning rate: Set the initial learning rate to 0.001; Number of training rounds: Set the maximum number of model training rounds to 10,000, and set an early stopping strategy to stop training when the model is no longer optimized in 200 consecutive rounds of training; Batch size: The batch size for training and testing is 1,000; Optimizer selection: Use the Adam optimizer for model training; Set model checkpoints and log storage: Set model storage checkpoints. When the loss function optimization of the model on the validation set exceeds the set threshold, the current optimal model is automatically stored; and the training log is saved.
5. A system for predicting industrial cooling load, characterized in that: include: Data acquisition module: used to collect data related to the three environmental characteristics of dry bulb temperature, wet bulb temperature and relative humidity over a period of time; Data processing and model training prediction module: used to preprocess and feature engineer the collected data, then use the preprocessed and feature engineered data to train the deep learning model, and use the trained deep learning model to predict the size and distribution of the cooling load for the next day.
6. The system for predicting industrial cooling load according to claim 5, characterized in that: The deep learning model is an LSTM model or a Transformer model.
7. The system for predicting industrial cooling load according to claim 6, characterized in that: The data processing and model training prediction module includes: Data standardization processing unit: used to normalize the collected data using the standard deviation standardization method, and apply the sliding time window technology to retain the data of the first 30 time steps; Dataset division and enhancement unit: used to reasonably divide the data set, retain 20% of the data as the test set, randomly shuffle the data set, and further expand the diversity of the data set through data enhancement technology; The training parameter configuration unit is used to configure the following parameters: learning rate: set the initial learning rate to 0.001; number of training rounds: set the maximum number of model training rounds to 10000, and set the early stopping strategy to stop training when the model is no longer optimized in 200 consecutive rounds of training; batch size: the batch size for training and testing is 1000; optimizer selection: use the Adam optimizer for model training; Model checkpoint and log saving setting unit: used to set model storage checkpoints. When the loss function optimization of the model on the validation set exceeds the set threshold, the current optimal model is automatically stored; and the training log is saved.
8. Application of the method for predicting industrial cooling load according to any one of claims 1 to 4 in energy efficiency control and selection of building refrigeration equipment and preparation of energy efficiency plan for building refrigeration equipment.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method for predicting industrial cooling load according to any one of claims 1 to 4 is implemented.
10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program is executed by a processor to implement the method for predicting industrial cooling load according to any one of claims 1 to 4.