Machine room temperature prediction method and device, equipment, storage medium and computer program product

Abstract

The invention relates to the technical field of machine room environment evaluation, in particular to a machine room temperature prediction method, device and equipment, a storage medium and a computer program product. According to the method, LightGBM, LSTM and NeuralProphet serve as initial base learners, the initial base learners are trained through five-fold cross validation on the basis of machine room temperature characteristic data, hyper-parameters of the initial base learners are optimized through an FA algorithm, and a target base learner is obtained; weighting the prediction result of the target base learner based on a preset IEM algorithm; taking the weighted prediction result as an input feature, and constructing multiple linear regression based on the input feature to obtain a meta-learner; based on the target base learner and the meta learner, the Stacking machine room temperature prediction model is constructed, and the machine room temperature prediction precision is improved.

Description

Technical Field

[0001] This application relates to the technical field of computer room environment assessment, and particularly to a method, device, equipment, storage medium and computer program product for predicting the temperature of a computer room. Background Art

[0002] The purpose of computer room temperature prediction is to predict the change trend of the computer room temperature within a certain period in the future based on the historical temperature data and related features in the computer room. Accurate temperature prediction is crucial for maintaining the efficient operation of the computer room, which can optimize the air conditioning operation strategy, reduce energy consumption, and ensure the stable operation of equipment at the same time. However, since the computer room temperature is affected by various complex factors, such as the external environmental temperature, the operation status of the air conditioner, the change of equipment load, etc., this makes the prediction model need to have strong non-linear modeling ability and robustness. At present, the common computer room temperature prediction methods mainly rely on single machine learning or deep learning models, such as Support Vector Machine (SVM), Long Short-Term Memory Network (LSTM), etc. These models can capture the regularity of data to a certain extent, but when dealing with complex non-linear relationships and diverse influencing factors, the prediction accuracy and generalization ability are often limited. In addition, a single model is easily restricted by the data distribution, and its adaptability to different computer room environments and characteristic data is poor, resulting in unstable model performance and poor accuracy of computer room temperature prediction. Therefore, how to improve the accuracy of computer room temperature prediction has become an urgent technical problem to be solved. Summary of the Invention

[0003] The main purpose of this application is to provide a method, device, equipment, storage medium and computer program product for predicting the temperature of a computer room, aiming to solve the technical problem of how to improve the accuracy of computer room temperature prediction.

[0004] To achieve the above purpose, this application provides a method for predicting the temperature of a computer room, and the method includes the following steps:

[0005] Taking LightGBM, LSTM, and NeuralProphet as initial base learners, training the initial base learners based on the computer room temperature feature data through five-fold cross-validation, and optimizing the hyperparameters of the initial base learners by using the FA algorithm to obtain target base learners;

[0006] Based on the preset IEM algorithm, performing weighted processing on the prediction results of the target base learners;

[0007] Taking the weighted prediction results as input features, and constructing a multiple linear regression based on the input features to obtain a meta-learner;

[0008] Based on the target base learners and the meta-learner, constructing a Stacking computer room temperature prediction model.

[0009] In one embodiment, before the step of using LightGBM, LSTM, and NeuralProphet as initial base learners, training the initial base learners based on the computer room temperature feature data through five-fold cross-validation, and optimizing the hyperparameters of the initial base learners using the FA algorithm to obtain the target base learners, the following steps are further included:

[0010] Obtain the historical temperature data and environmental data of the computer room;

[0011] Preprocess the historical temperature data and environmental data of the computer room to obtain preprocessed data, where the preprocessing includes one or more of data cleaning, timestamp processing, and data normalization;

[0012] Extract time series features from the preprocessed data to obtain the computer room temperature feature data.

[0013] In one embodiment, the step of using LightGBM, LSTM, and NeuralProphet as initial base learners, training the initial base learners based on the computer room temperature feature data through five-fold cross-validation, and optimizing the hyperparameters of the initial base learners using the FA algorithm to obtain the target base learners includes:

[0014] Use LightGBM, LSTM, and NeuralProphet as initial base learners, and based on the five-fold cross-validation, divide the computer room temperature feature data into a validation set and a training set;

[0015] Train the initial base learners based on the validation set and the training set;

[0016] Based on the results of the training, perform position update and iteration using the FA algorithm;

[0017] Determine the optimal hyperparameters according to the results of the position update and iteration;

[0018] Retrain the initial base learners based on the optimal hyperparameters to obtain the target base learners.

[0019] In one embodiment, the step of weighting the prediction results of the target base learners based on the preset IEM algorithm includes:

[0020] Based on a preset weight allocation strategy, assign initial weights to the target base learners;

[0021] Obtain the prediction errors of the target base learners on the validation set;

[0022] Based on the prediction error, the preset IEM algorithm is used to adjust the weights of the target base learner corresponding to the prediction error, and the adjusted weights are subjected to the weighted processing.

[0023] In one embodiment, the step of using the weighted prediction result as an input feature and constructing a multiple linear regression based on the input feature to obtain a meta-learner includes:

[0024] Using the weighted prediction result as an input feature, and arranging the input feature into a feature matrix based on the sample order, where each column of the feature matrix corresponds to the prediction output of a target base learner;

[0025] Using the actual observed value of the initial base learner as the target variable, and pairing the feature matrix with the target variable;

[0026] Based on the pairing result, using a preset multiple linear regression method to fit the feature matrix with the target variable to obtain the meta-learner.

[0027] In one embodiment, the step of constructing a Stacking computer room temperature prediction model based on the target base learner and the meta-learner includes:

[0028] Training the meta-learner based on the prediction result and the actual observed value of the target base learner;

[0029] Learning the mapping relationship between the prediction result of the target base learner and the actual observed value based on the training result;

[0030] Combining the target base learner and the meta-learner into the Stacking computer room temperature prediction model based on the mapping relationship.

[0031] In addition, to achieve the above object, the present application also proposes a computer room temperature prediction device, and the computer room temperature prediction device includes:

[0032] A base learner module, configured to use LightGBM, LSTM, and NeuralProphet as initial base learners, train the initial base learners based on computer room temperature feature data through five-fold cross-validation, and optimize the hyperparameters of the initial base learners using the FA algorithm to obtain target base learners;

[0033] A weighting module, configured to perform weighted processing on the prediction results of the target base learners based on a preset IEM algorithm;

[0034] A meta-learner module, configured to use the weighted prediction result as an input feature and construct a multiple linear regression based on the input feature to obtain a meta-learner;

[0035] A target module, configured to construct a Stacking computer room temperature prediction model based on the target base learner and the meta-learner.

[0036] In addition, to achieve the above object, the present application further provides a computer room temperature prediction device, which includes: a memory, a processor, and a computer room temperature prediction program stored on the memory and executable on the processor, and the computer room temperature prediction program is configured to implement the steps of the computer room temperature prediction method as described above.

[0037] In addition, to achieve the above object, the present application further provides a storage medium, on which a computer room temperature prediction program is stored, and when the computer room temperature prediction program is executed by a processor, the steps of the computer room temperature prediction method as described above are implemented.

[0038] In addition, to achieve the above object, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the computer room temperature prediction method as described above are implemented.

[0039] In the present application, LightGBM, LSTM, and NeuralProphet are used as initial base learners. Based on the computer room temperature feature data, the initial base learners are trained through five-fold cross-validation, and the FA algorithm is used to optimize the hyperparameters of the initial base learners to obtain target base learners; based on the preset IEM algorithm, the prediction results of the target base learners are weighted; the weighted prediction results are used as input features, and based on the input features, a multiple linear regression is constructed to obtain a meta-learner; based on the target base learners and the meta-learner, a Stacking computer room temperature prediction model is constructed. In the present application, by using LightGBM, LSTM, and NeuralProphet as base learners, training the base learners by five-fold cross-validation, and using the FA algorithm to optimize their hyperparameters, the prediction accuracy and generalization ability of the base learners are improved; the prediction results of the target base learners are weighted based on the preset IEM algorithm to give full play to the advantages of each base learner and further reduce the prediction error; the weighted prediction results are used as input features, a multiple linear regression is constructed as the meta-learner, and finally a Stacking model is constructed by combining the target base learners and the meta-learner, effectively integrating the prediction capabilities of multiple models and improving the accuracy of computer room temperature prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic flowchart of the first embodiment of the computer room temperature prediction method of the present application;

[0041] Figure 2It is a schematic diagram of a sub - process in the second embodiment of the computer room temperature prediction method of this application;

[0042] Figure 3 It is a schematic diagram of a sub - process in the third embodiment of the computer room temperature prediction method of this application;

[0043] Figure 4 It is a schematic diagram of the module structure of the computer room temperature prediction device in the embodiment of this application;

[0044] Figure 5 It is a schematic diagram of the device structure of the hardware operating environment involved in the computer room temperature prediction method in the embodiment of this application.

[0045] The realization, functional characteristics, and advantages of the purpose of this application will be further described in combination with the embodiments and with reference to the accompanying drawings. Specific Embodiments

[0046] It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0047] In order to better understand the technical solution of this application, the following will be described in detail in combination with the accompanying drawings of the specification and specific embodiments.

[0048] It should be noted that the purpose of computer room temperature prediction is to predict the change trend of the computer room temperature within a certain period in the future based on the historical temperature data and related features in the computer room. Accurate temperature prediction is crucial for maintaining the efficient operation of the computer room, which can optimize the air - conditioning operation strategy, reduce energy consumption, and ensure the stable operation of equipment at the same time. However, since the computer room temperature is affected by various complex factors, such as the external environmental temperature, the operation status of the air - conditioner, the change of equipment load, etc., this makes the prediction model need to have strong non - linear modeling ability and robustness. At present, common computer room temperature prediction methods mainly rely on single machine learning or deep learning models, such as support vector machine (SVM), long short - term memory network (LSTM), etc. These models can capture the regularity of data to a certain extent, but when dealing with complex non - linear relationships and diverse influencing factors, the prediction accuracy and generalization ability are often limited. In addition, single models are easily restricted by the data distribution, and their adaptability to different computer room environments and characteristic data is poor, resulting in unstable model performance and poor accuracy of computer room temperature prediction. Therefore, how to improve the accuracy of computer room temperature prediction has become an urgent technical problem to be solved.

[0049] The main solution of this application is as follows: By using LightGBM, LSTM, and NeuralProphet as the initial base learners, based on the computer room temperature feature data, the initial base learners are trained through five-fold cross-validation, and the FA algorithm is used to optimize the hyperparameters of the initial base learners to obtain the target base learners; based on the preset IEM algorithm, the prediction results of the target base learners are weighted; the weighted prediction results are used as input features, and based on the input features, multiple linear regression is constructed to obtain the meta-learner; based on the target base learners and the meta-learner, a Stacking computer room temperature prediction model is constructed.

[0050] In this application, by using LightGBM, LSTM, and NeuralProphet as the base learners, training the base learners through five-fold cross-validation, and using the FA algorithm to optimize their hyperparameters, the prediction accuracy and generalization ability of the base learners are improved; based on the preset IEM algorithm, the prediction results of the target base learners are weighted to give full play to the advantages of each base learner and further reduce the prediction error; the weighted prediction results are used as input features, multiple linear regression is constructed as the meta-learner, and finally, a Stacking model is constructed by combining the target base learners and the meta-learner, effectively integrating the prediction capabilities of multiple models and improving the accuracy of computer room temperature prediction.

[0051] It should be noted that the execution subject of the method in this embodiment can be a computing service device with data processing, network communication, and program running functions, or the above-mentioned computer room temperature prediction device with the same or similar functions. This embodiment and the following embodiments will be described by taking the computer room temperature prediction device as an example.

[0052] Based on this, the first embodiment of the computer room temperature prediction method of this application is proposed. Please refer to Figure 1 , Figure 1 which is the flowchart of the first embodiment of the computer room temperature prediction method of this application.

[0053] In this embodiment, the computer room temperature prediction method includes the following steps:

[0054] S1: Use LightGBM, LSTM, and NeuralProphet as the initial base learners, based on the computer room temperature feature data, train the initial base learners through five-fold cross-validation, and use the FA algorithm to optimize the hyperparameters of the initial base learners to obtain the target base learners;

[0055] It should be noted that LightGBM (Light Gradient Boosting Machine) is an efficient gradient boosting framework that focuses on quickly processing large-scale data and high-dimensional features. It is commonly used in tasks such as regression and classification and is particularly suitable for processing data with non-linear relationships. LSTM refers to Long Short-Term Memory, a special type of recurrent neural network that is good at handling long-term dependencies in time series data and can capture the time trends and dynamic patterns of temperature changes. NeuralProphet is a time series prediction model based on Facebook Prophet that integrates deep learning techniques and can handle time series data with strong trends, seasonality, and periodicity well. Five-fold cross-validation is a model validation method that divides the training data into five subsets. Each time, one of the subsets is selected as the validation set, and the other four subsets are used as the training set. After repeating this five times, the average value is taken to evaluate the model performance, aiming to improve the generalization ability of the model. The FA algorithm refers to the Firefly Algorithm, which is a population-based optimization meta-heuristic algorithm that simulates the behavior of fireflies attracting each other through light intensity and has global optimization capabilities. It is used to efficiently optimize the hyperparameters of the base learner. The target base learner refers to the base learner that is trained through five-fold cross-validation and optimized by the FA algorithm, which has higher prediction accuracy and robustness and is used as the input of the first-layer model of the Stacking model.

[0056] Specifically, LightGBM, LSTM, and NeuralProphet are used as the initial base learners, and input samples are constructed based on the historical data and feature data of the computer room temperature. Each base learner is trained through five-fold cross-validation. The training data is divided into five subsets. Each time, one of the subsets is selected as the validation set, and the other four subsets are used as the training set. Training and validation are carried out alternately to ensure that all data points are used for validation. In this way, during the training process, the performance fluctuations of the model caused by the randomness of data division are effectively avoided, and the stability and generalization ability of the base learner are improved. After the training is completed, the prediction results and performance indicators of each base learner are saved as the basis for subsequent optimization and model integration.

[0057] Furthermore, after the base learners are trained, the FA algorithm is used to optimize their hyperparameters. First, the initial parameters of the FA algorithm (such as population size, maximum number of iterations) and the hyperparameter search space of the base learners (such as the learning rate of LightGBM, the number of hidden units of LSTM, the trend flexibility parameter of NeuralProphet, etc.) are defined. By evaluating the performance of the base learners (such as mean squared error or mean absolute error on the validation set) as the fitness function, the FA algorithm searches for and optimizes the hyperparameter combinations in the initial population. In each iteration, the fireflies will adjust their positions (i.e., hyperparameter combinations) according to the fitness values and move towards a direction with better performance, and finally output the globally optimal hyperparameter configuration. The optimized base learners have better performance, and the generated target base learners have higher prediction accuracy.

[0058] Through five-fold cross-validation training, the problem of model bias that may be brought by a single data set division is avoided, enabling the base learners to fully learn the characteristics of the computer room temperature, and improving the robustness and generalization ability of the model. Using the FA algorithm to optimize the hyperparameters of the base learners can quickly find the globally optimal hyperparameter combination, avoiding the problem of being easily trapped in local optima in traditional hyperparameter tuning methods, and further improving the prediction accuracy of the base learners. The three base learners, LightGBM, LSTM, and NeuralProphet, have different characteristics and are respectively good at dealing with non-linear features, time series trends, and periodic features. By combining and optimizing these base learners, the model can better adapt to the complex patterns and diverse influencing factors in the computer room temperature data. Five-fold cross-validation uses the data multiple times for validation during the training process, effectively reducing the dependence of the model on a specific data distribution and reducing the risk of overfitting. When the FA algorithm optimizes the hyperparameters of the base learners, it can control the model complexity, enabling the model to ensure accuracy without overfitting the training data. The target base learners after five-fold cross-validation training and FA algorithm optimization have more reliable prediction performance. As the input of the first layer of the Stacking model, they can provide high-quality prediction results, laying a solid foundation for subsequent model integration.

[0059] S2: Based on the preset IEM algorithm, perform weighted processing on the prediction results of the target base learners;

[0060] It should be noted that the IEM algorithm is an improved weighted algorithm (Improved Ensemble Method) used to dynamically weight the prediction results of base learners. Compared with simple average weights or fixed weights, the IEM algorithm can dynamically adjust the weights according to the actual performance (such as error or correlation) of the base learners during the training or validation phase, enabling high-performance models to contribute more weight and low-performance models to contribute less weight, thereby improving the overall prediction accuracy. Weighted processing refers to weighted summation of the prediction results of multiple base learners according to the weights calculated by the IEM algorithm to integrate the prediction capabilities of each model and generate the final comprehensive prediction result.

[0061] Specifically, after the target base learners are trained, they are used to predict the test set or validation set respectively to generate corresponding prediction results. At the same time, the performance metrics (such as mean absolute error MAE or mean squared error MSE) of each base learner on the validation set are calculated as an important basis for weighted processing. The role of the performance metrics is to reflect the advantages and disadvantages of each base learner in the current task. The better the metrics (the lower the error), the greater the contribution of its prediction result in the final weighting.

[0062] Furthermore, based on the IEM algorithm, the weights are dynamically adjusted in combination with the performance metrics of the target base learners. The algorithm will preferentially assign higher weights to base learners with lower errors while suppressing the weight influence of models with higher errors, thereby effectively integrating the prediction capabilities of different models. Then, the prediction results of the target base learners are weighted and summed according to the calculated weights to generate the final comprehensive prediction result. The whole process can automatically balance the contributions of multiple base learners and avoid the negative impact of poor performance of a single model on the overall prediction result.

[0063] The IEM algorithm effectively reduces the prediction error and improves the prediction accuracy of the overall model by dynamically adjusting the weights, enabling base learners with superior performance to contribute more weight to the final prediction result. By weighted integration of the results of multiple base learners, the problem of large errors in a single model can be alleviated, the dependence of the model on a single learner can be reduced, and the adaptability of the comprehensive model to different data distributions can be improved. Dynamic weighted processing can automatically adjust the weight allocation of base learners in different tasks and data scenarios to ensure that the final prediction result is more stable and reliable. The comprehensive prediction result generated by weighted processing provides high-quality input features for the construction of subsequent meta-learners, thereby laying a foundation for the overall optimization of the Stacking model.

[0064] S3: Use the prediction result after weighted processing as input features, and based on the input features, construct a multiple linear regression to obtain a meta-learner;

[0065] It should be noted that the predicted results after weighted processing refer to the comprehensive predicted values generated by weighting the predicted results of multiple target base learners (such as LightGBM, LSTM, NeuralProphet) through the IEM algorithm. These weighted results retain the prediction information of each base learner and serve as the input features of the meta-learner. The input features refer to the weighted predicted results of multiple target base learners, which are input into the meta-learner in matrix form. Each column corresponds to the weighted predicted value of a base learner, and each row corresponds to a sample. Multiple linear regression is a simple and efficient regression model that predicts the target variable by linearly combining the input features. This model structure is simple and suitable for use as a meta-learner to further integrate the predicted results of the base learners into the final predicted results. As the learner in the second layer of the Stacking model, the meta-learner further learns the relationship between these predicted values and the actual target values by integrating the weighted predicted results of the base learners, and generates the final predicted results.

[0066] Specifically, the predicted results of the base learners after weighted processing are organized into an input feature matrix, where each column represents the predicted result of a base learner, and each row corresponds to the predicted value of a sample. Then, the actual observed values are extracted from the original data as the target variables, corresponding one by one with the input feature matrix. The construction of the feature matrix and the target variables ensures that the meta-learner can learn the mapping relationship between the predicted results of the base learners and the true target values.

[0067] Furthermore, based on the organized input features and target variables, a multiple linear regression model is constructed as the meta-learner. By fitting the input features and the target variables, the multiple linear regression model learns the linear combination weights of the predicted results of each base learner and generates the final predicted results. After training, the model performance is evaluated through the validation set to ensure that the meta-learner can effectively integrate the prediction capabilities of the base learners, and the trained model is saved for subsequent predictions.

[0068] Taking the prediction results after weighted processing as input features, multiple linear regression can further explore the potential information of the prediction results of each base learner, thereby improving the final prediction accuracy. As a meta-learner, multiple linear regression has a simple structure and high computational efficiency. It can quickly train and fit the relationship between input features and target variables, reduce the risk of overfitting, and avoid the computational burden caused by excessive model complexity. By learning the linear combination of the prediction results of the base learners, the meta-learner can reduce the impact of the prediction error of a single model on the final result, and at the same time enhance the adaptability to different data distributions, thereby improving the accuracy and robustness of the prediction. As the second layer of the Stacking model, the meta-learner integrates the prediction information of multiple base learners and generates the final prediction result, providing an efficient result integration mechanism for the overall model. By using the prediction results of the weighted base learners as input features and constructing a meta-learner using multiple linear regression, the prediction capabilities of multiple base learners are effectively integrated, which not only improves the overall prediction performance of the model, but also ensures computational efficiency and model robustness, providing strong support for the efficient application of the Stacking model.

[0069] S4: Based on the target base learner and the meta-learner, construct a Stacking computer room temperature prediction model.

[0070] It should be noted that the Stacking model is a two-layer ensemble learning model structure. The first layer consists of multiple base learners, which are used to capture diverse features of the data; the second layer is the meta-learner, which is used to integrate the prediction results of the base learners in the first layer, thereby improving the overall prediction performance.

[0071] Specifically, taking the target base learners (LightGBM, LSTM, and NeuralProphet) as the first layer of the Stacking model, training them respectively using the training data, and making predictions on the test set or validation set based on the trained base learners to generate the prediction results of multiple base learners. These results represent the understanding and prediction capabilities of different models for the computer room temperature characteristics, forming the prediction output of the first layer and providing input for the subsequent meta-learner.

[0072] Furthermore, based on the prediction results of the base learners in the first layer, taking these results as input features, construct the second-layer meta-learner (multiple linear regression). The meta-learner learns the mapping relationship between the weighted combination of the prediction results of the base learners and the actual target value by weighted combination, and generates the final comprehensive prediction value. After completing the model training, the Stacking model integrates the target base learner and the meta-learner to form a complete two-layer prediction framework.

[0073] The Stacking model significantly improves the overall prediction accuracy by combining the prediction capabilities of multiple target base learners and using a meta-learner to optimize and integrate their results. The combination of multiple base learners can capture different data features, reducing the dependence of a single model on the data distribution. The meta-learner further reduces the impact of the prediction errors of a single base learner, making the model more robust and adaptable. The two-layer model structure is clear, dividing the functions of different models: the first layer focuses on diversity prediction, and the second layer focuses on integration and optimization, thus enhancing the overall stability and consistency of the model. The Stacking model framework is highly flexible and can adjust the types or quantities of base learners according to different application scenarios. At the same time, it optimizes the integration results through the meta-learner to adapt to complex computer room temperature prediction tasks. By constructing a Stacking model based on target base learners and a meta-learner, multi-level prediction ability integration is achieved. The first-layer base learners provide diverse capture of data characteristics, and the second-layer meta-learner improves the prediction accuracy and robustness through integration and optimization, thus forming an efficient and stable computer room temperature prediction model.

[0074] In this embodiment, LightGBM, LSTM, and NeuralProphet are used as initial base learners. Based on the computer room temperature feature data, the initial base learners are trained through five-fold cross-validation, and the FA algorithm is used to optimize the hyperparameters of the initial base learners to obtain target base learners; based on the preset IEM algorithm, the prediction results of the target base learners are weighted; the weighted prediction results are used as input features, and based on the input features, a multiple linear regression is constructed to obtain a meta-learner; based on the target base learners and the meta-learner, a Stacking computer room temperature prediction model is constructed. In this embodiment, by using LightGBM, LSTM, and NeuralProphet as base learners, training the base learners through five-fold cross-validation, and using the FA algorithm to optimize their hyperparameters, the prediction accuracy and generalization ability of the base learners are improved; the prediction results of the target base learners are weighted based on the preset IEM algorithm to give full play to the advantages of each base learner and further reduce the prediction error; the weighted prediction results are used as input features, a multiple linear regression is constructed as the meta-learner, and finally, a Stacking model is constructed by combining the target base learners and the meta-learner, effectively integrating the prediction capabilities of multiple models and improving the accuracy of computer room temperature prediction.

[0075] Based on the above first embodiment, a second embodiment of the computer room temperature prediction method of this application is proposed. Please refer to Figure 2 , Figure 2 which is a schematic diagram of a sub-process in the second embodiment of the computer room temperature prediction method of this application.

[0076] As Figure 2As shown, in this embodiment, before step S1, it further includes:

[0077] S1a: Obtain the historical temperature data and environmental data of the computer room;

[0078] S1b: Preprocess the historical temperature data and the environmental data of the computer room to obtain preprocessed data, and the preprocessing includes one or more of data cleaning, timestamp processing, and data normalization;

[0079] S1c: Extract time series features from the preprocessed data to obtain the temperature feature data of the computer room.

[0080] It should be noted that the historical temperature data of the computer room refers to the temperature data collected by the temperature sensors or monitoring devices in the computer room, usually recorded in the form of a time series, including timestamps and corresponding temperature values. The environmental data refers to the external environmental information that affects the temperature change in the computer room, such as the outside temperature, humidity, air supply temperature of the air conditioner, equipment load, etc. These data are closely related to the temperature in the computer room. Data cleaning is a data preprocessing method used to remove missing values, outliers, or noise in the data to ensure data quality and consistency. Timestamp processing refers to formatting, filling, and aligning the time stamps of time series data to ensure the integrity and standardization of the time dimension. Data normalization refers to scaling the data to a fixed range (such as [0,1]) through linear transformation to eliminate the dimensional differences between different features, facilitating model training and feature extraction. Time series feature extraction refers to extracting feature information that is helpful for model training from the time dependence, periodicity, and trend of time series data, such as periodic encoding, sliding window statistical features, and autocorrelation.

[0081] Specifically, obtain the historical temperature data from the temperature monitoring system and sensors in the computer room, and combine the environmental related data (such as the outside air temperature, humidity, equipment load, etc.) to form an original data set containing timestamps, temperature values, and multi-dimensional environmental features. Detect missing values and outliers in the data. For missing values, they can be filled by interpolation method, and for outliers, they can be processed by sliding window smoothing or setting upper and lower limits. Unify the timestamp format to ensure the continuity of the time series; if there are data with irregular time intervals, interpolation filling or resampling is required to standardize the time interval. Scale the feature values so that they are distributed within a unified range (such as [0,1] or with a mean of 0 and a standard deviation of 1) to eliminate the scale differences between different data, facilitating model training.

[0082] Furthermore, based on the preprocessed dataset, explicit and implicit features of the time series are extracted to generate the data of the computer room temperature features. The periodic information (such as time periods, weekdays / weekends) in the timestamps is extracted and represented by sine or cosine encoding to preserve the time period characteristics. The local statistical features (such as mean, maximum, minimum, variance, etc.) of the time series are calculated using a sliding window to capture the local change patterns of temperature over time. The autocorrelation in the time series is analyzed to extract the delay patterns and internal rules between temperature values to help the model capture the time dependence. After the extraction is completed, a dataset of computer room temperature features for subsequent model training is generated.

[0083] Removing noise and outliers through data cleaning ensures the accuracy of the computer room temperature data; timestamp processing and normalization operations further standardize the data format and reduce the model performance fluctuations caused by data inconsistency. By introducing periodic, local statistical, and autocorrelation features in time series feature extraction, the model can better understand the dynamic patterns and long-term trends of computer room temperature changes, thereby improving the prediction ability. The preprocessing and feature extraction steps provide high-quality feature data for the subsequent prediction model, reduce the complexity of the original data, and improve the efficiency and accuracy of model training. By combining the time series feature extraction of environmental data, the model can adapt to diverse influencing factors of computer room temperature, improving the prediction adaptability and robustness for complex scenarios. This step generates a high-quality dataset of computer room temperature features through data cleaning, timestamp processing, normalization, and time series feature extraction, which not only ensures the accuracy and consistency of the data but also enhances the feature expression ability, providing a solid foundation for the construction of the subsequent prediction model.

[0084] Based on the above first embodiment, in this embodiment, step S1 includes:

[0085] S11: Using LightGBM, LSTM, and NeuralProphet as the initial base learners, and based on the five-fold cross-validation, dividing the computer room temperature feature data into a validation set and a training set;

[0086] S12: Training the initial base learners based on the validation set and the training set;

[0087] S13: Based on the results of the training, using the FA algorithm for position update and iteration;

[0088] S14: Determining the optimal hyperparameters according to the results of the position update and iteration;

[0089] S15: Retraining the initial base learners based on the optimal hyperparameters to obtain the target base learners.

[0090] It should be noted that five-fold cross-validation is a data partitioning and validation method that divides the data into 5 subsets. In turn, 1 subset is used as the validation set, and the remaining 4 subsets are used as the training set. Training and validation are performed iteratively to improve the generalization ability of the model. Position update and iteration are key steps in the FA algorithm. By calculating the attractiveness of fireflies based on light intensity (fitness) and updating their positions, the hyperparameter combination of the model is gradually optimized. The optimal hyperparameters are the globally optimal hyperparameter combination obtained through iterative search of the FA algorithm, which is used to improve the performance of the model.

[0091] Specifically, perform five-fold cross-validation on the computer room temperature feature data, divide the data into a training set and a validation set. In turn, use 4 subsets as the training set to train the base learner, and 1 subset as the validation set for validation to ensure that each data point is used as the validation set at least once. Use three initial base learners, LightGBM, LSTM, and NeuralProphet, to learn the data features on the training set respectively, and evaluate their prediction performance on the validation set. Save the prediction results and validation errors of each base learner as the input for subsequent hyperparameter optimization.

[0092] Furthermore, based on the training and validation results, use the FA algorithm for hyperparameter optimization. First, initialize the parameters of the FA algorithm (such as population size, maximum number of iterations) and the initial population position (i.e., hyperparameter combination). Calculate the position fitness (validation error) of each firefly, update the position according to the fitness, and move in a better direction. In each iteration, re-evaluate the fitness of the updated hyperparameter combination until the iteration number limit or convergence condition is reached. Finally, determine the globally optimal hyperparameter combination, and use this combination to retrain LightGBM, LSTM, and NeuralProphet to generate the target base learner, providing higher-quality prediction ability for the subsequent model.

[0093] This step trains the initial base learner through five-fold cross-validation, optimizes the hyperparameters using the FA algorithm, and finally generates the target base learner, which not only improves the prediction accuracy and generalization ability of the model, but also effectively enhances the adaptability and robustness of the model, providing high-quality basic input for the construction of the subsequent Stacking model.

[0094] In this embodiment, by using LightGBM, LSTM, and NeuralProphet as the initial base learners, based on the data of computer room temperature characteristics, the initial base learners are trained through five-fold cross-validation, and the FA algorithm is used to optimize the hyperparameters of the initial base learners to obtain the target base learners; based on the preset IEM algorithm, the prediction results of the target base learners are weighted; the weighted prediction results are used as input features, and based on the input features, a multiple linear regression is constructed to obtain the meta-learner; based on the target base learners and the meta-learner, a Stacking computer room temperature prediction model is constructed. In this embodiment, by using LightGBM, LSTM, and NeuralProphet as the base learners, training the base learners by five-fold cross-validation, and using the FA algorithm to optimize their hyperparameters, the prediction accuracy and generalization ability of the base learners are improved; based on the preset IEM algorithm, the prediction results of the target base learners are weighted to give full play to the advantages of each base learner and further reduce the prediction error; the weighted prediction results are used as input features, a multiple linear regression is constructed as the meta-learner, and finally, a Stacking model is constructed by combining the target base learners and the meta-learner, effectively integrating the prediction capabilities of multiple models and improving the accuracy of computer room temperature prediction.

[0095] Based on the above second embodiment, a third embodiment of the computer room temperature prediction method of the present application is proposed. Please refer to Figure 3 , Figure 3 which is a schematic diagram of a sub-process in the third embodiment of the computer room temperature prediction method of the present application.

[0096] In this embodiment, step S2 includes:

[0097] S21: Based on the preset weight allocation strategy, assign an initial weight to the target base learner;

[0098] S22: Obtain the prediction error of the target base learner on the validation set;

[0099] S23: Based on the prediction error, use the preset IEM algorithm to adjust the weight of the target base learner corresponding to the prediction error, and perform the weighted processing on the adjusted weight.

[0100] It should be noted that the initial weight refers to the initial weight assigned to the target base learner before adjustment, which is usually set according to the characteristics of the base learner, historical performance, or default value, and is used as the starting point for subsequent weight adjustment. The prediction error refers to the error between the prediction result of the target base learner on the validation set and the actual value, which is used to evaluate the performance of the base learner. Common metrics include mean squared error (MSE) or mean absolute error (MAE).

[0101] Specifically, based on a preset weight allocation strategy, initial weights are assigned to the target base learners. For example, equal initial weights (such as 1 / 3) are given to LightGBM, LSTM, and NeuralProphet, or weight ratios are preset according to experience. Then, the target base learners are used to predict the validation set, and their prediction errors are calculated. The smaller the error, the better the prediction performance of the base learner. The prediction error serves as a reference index for subsequent weight adjustment and provides input data for the IEM algorithm.

[0102] Furthermore, based on the prediction errors of the target base learners, the IEM algorithm is used to adjust the initial weights. The IEM algorithm dynamically adjusts the weights of the base learners with smaller errors to make their contributions to the final prediction greater, while reducing the weights of the base learners with larger errors, thereby balancing the influences of the base learners. The adjusted weights are used to perform weighted processing on the prediction results of the target base learners. The predicted values of each base learner are added according to the weight ratio to generate the final comprehensive prediction result. This process automatically optimizes the weight allocation, making the prediction result of the overall model more accurate and stable.

[0103] By dynamically adjusting the weights of the target base learners through the IEM algorithm, the base learners with smaller prediction errors contribute greater weights, thereby reducing the overall prediction error and improving the prediction accuracy of the model. Weakening the weights of the base learners with larger errors effectively alleviates the negative impact of the poor performance of a single model on the comprehensive prediction result and enhances the robustness of the model. The IEM algorithm can explore the characteristics of different base learners, give priority to the prediction ability of the models with better performance, and improve the effect of multi-model collaborative work. The dynamic weight allocation mechanism can adapt to the changes in data distribution in different scenarios, ensuring that the model can still maintain high prediction performance and stability in complex environments. The comprehensive prediction result after weighted processing is used as the input feature of the Stacking model, which can improve the learning effect of the meta-learner and further improve the prediction performance of the overall model. This step adjusts the weights of the target base learners through the IEM algorithm based on the prediction error and performs weighted processing, realizing dynamic optimization of weight allocation, improving the prediction accuracy and robustness of the model, and providing high-quality input features for the subsequent construction of the Stacking model, ensuring the performance of the comprehensive model.

[0104] Based on the above second embodiment, in this embodiment, step S3 includes:

[0105] S31: Using the prediction result after weighted processing as the input feature, and based on the sample order, organizing the input feature into a feature matrix, where each column of the feature matrix corresponds to the prediction output of a target base learner;

[0106] S32: Use the actual observation value of the initial base learner as the target variable, and pair the feature matrix with the target variable;

[0107] S33: Based on the pairing result, use a preset multiple linear regression method to fit the feature matrix with the target variable to obtain the meta-learner.

[0108] It should be noted that the sample order refers to the arrangement order of samples in time series data, ensuring the time consistency and logical correspondence between the input feature matrix and the target variable. The feature matrix is a two-dimensional matrix generated by organizing the prediction results of the weighted target base learners in the sample order, where each column corresponds to the prediction result of a base learner, and each row corresponds to the feature vector of a time series sample.

[0109] Specifically, organize the prediction results of the weighted target base learners in the sample order to form a feature matrix, where each column represents the prediction result of a base learner, and each row represents the feature vector of a sample. The arrangement of the feature matrix ensures the logical consistency of time series samples. Subsequently, extract the actual observation value of each time series sample from the original data as the target variable, and correspond the feature matrix with the target variable one by one to form paired data, providing input and supervision signals for the training of the meta-learner.

[0110] Furthermore, based on the paired data, use a preset multiple linear regression method to fit the feature matrix and the target variable, and learn the linear combination weights of the prediction results of different base learners in the feature matrix. Multiple linear regression generates a meta-learner that can integrate the prediction results of multiple base learners by minimizing the error between the predicted value and the target variable. After the fitting is completed, verify the performance of the meta-learner to ensure that it can effectively improve the accuracy and robustness of the final prediction result, and use the trained meta-learner for subsequent Stacking model prediction.

[0111] The feature matrix takes the weighted prediction results of the target base learners as input features. The meta-learner further optimizes the combination of these prediction results through multiple linear regression, effectively integrating the prediction capabilities of different base learners. The multiple linear regression model utilizes the linear relationship between the input features and the target variable to adjust the linear combination weights of the base learners' prediction results, further reducing the impact of the prediction errors of a single base learner on the final result. The multiple linear regression model, as the meta-learner, has a simple structure and high computational efficiency, can quickly fit the relationship between the feature matrix and the target variable, and at the same time avoid the overfitting problem that may be caused by complex models. By learning the prediction contributions of different base learners, the meta-learner can dynamically adapt to different data scenarios, improving the stability and adaptability of the integrated model. As the second-layer learner of the Stacking model, multiple linear regression further optimizes the prediction results of the target base learners, improving the prediction performance and robustness of the overall model. By organizing the weighted prediction results of the base learners into a feature matrix and using multiple linear regression to fit the meta-learner in combination with the target variable, this step realizes the optimized integration of the prediction capabilities of multiple models, not only improving the accuracy and efficiency of model prediction, but also enhancing the stability of the model, providing an efficient second-layer structure for the Stacking model.

[0112] Based on the above second embodiment, in this embodiment, step S4 includes:

[0113] S41: Training the meta-learner based on the prediction results of the target base learners and the actual observations;

[0114] S42: Learning the mapping relationship between the prediction results of the target base learners and the actual observations based on the training results;

[0115] S43: Combining the target base learners and the meta-learner into the Stacking computer room temperature prediction model based on the mapping relationship.

[0116] It should be noted that the actual observations refer to the real temperature values extracted from the computer room temperature feature data, which are used to compare with the prediction results of the target base learners to guide the training of the meta-learner. The mapping relationship refers to the linear or non-linear relationship between the prediction results of the target base learners and the actual observations learned by the meta-learner through training, which is used to generate the final prediction value.

[0117] Specifically, the prediction results of the target base learners are used as input features, and the actual observed values are used as target variables, which are input into the meta-learner for training. The meta-learner fits the relationship between the input features and the target variables, learns the contribution weights of the prediction results of different target base learners in the overall prediction, and generates an optimal linear combination or mapping function. During the training process, the validation set is used to evaluate the performance of the meta-learner to ensure that it can effectively reduce the prediction error and improve the reliability of the overall prediction result.

[0118] Furthermore, based on the mapping relationship learned by the meta-learner, the target base learner and the meta-learner are combined into a complete Stacking computer room temperature prediction model. The target base learner is responsible for providing diverse prediction results, and the meta-learner generates the final prediction value by integrating the relationship between the output of the target base learner and the actual observed values. The integrated Stacking model structure is clear: the first layer consists of multiple target base learners, and the second layer is the meta-learner, which fully utilizes the advantages of multi-model collaborative prediction to improve the accuracy and robustness of computer room temperature prediction.

[0119] By learning the mapping relationship between the prediction results of the target base learner and the actual observed values through the meta-learner, the characteristics of different base learners are fully exploited, and the overall prediction ability is optimized. The meta-learner reduces the impact of the prediction error of a single model by dynamically adjusting the contribution weights of the base learners, thereby improving the overall prediction accuracy of the Stacking model. By integrating the prediction results of multiple base learners, the Stacking model significantly reduces the impact of the performance fluctuation of a single base learner on the final prediction, enhancing the robustness and adaptability of the model. The first-layer target base learner focuses on the extraction of diverse features of the data, and the second-layer meta-learner is responsible for result integration and optimization. The hierarchical design improves the efficiency and stability of the model. Through the organic combination of the target base learner and the meta-learner, the Stacking model can flexibly adapt to diverse and complex computer room temperature prediction tasks and provide more accurate and reliable prediction results. This step integrates the target base learner and the meta-learner into a Stacking model through the training of the meta-learner and the learning of the mapping relationship, fully leveraging the prediction advantages of multiple base learners, optimizing the performance and adaptability of the overall model, and providing a strong technical guarantee for high-precision prediction of computer room temperature.

[0120] In this embodiment, by using LightGBM, LSTM, and NeuralProphet as the initial base learners, based on the computer room temperature feature data, the initial base learners are trained through five-fold cross-validation, and the FA algorithm is used to optimize the hyperparameters of the initial base learners to obtain the target base learners; based on the preset IEM algorithm, the prediction results of the target base learners are weighted; the weighted prediction results are used as input features, and based on the input features, multiple linear regression is constructed to obtain the meta-learner; based on the target base learners and the meta-learner, a Stacking computer room temperature prediction model is constructed. In this embodiment, by using LightGBM, LSTM, and NeuralProphet as the base learners, the base learners are trained using five-fold cross-validation, and the FA algorithm is used to optimize their hyperparameters, improving the prediction accuracy and generalization ability of the base learners; based on the preset IEM algorithm, the prediction results of the target base learners are weighted to give full play to the advantages of each base learner and further reduce the prediction error; the weighted prediction results are used as input features, multiple linear regression is constructed as the meta-learner, and finally, a Stacking model is constructed by combining the target base learners and the meta-learner, effectively integrating the prediction capabilities of multiple models and improving the accuracy of computer room temperature prediction.

[0121] An embodiment of the present application further provides a computer room temperature prediction device. Please refer to Figure 4 , Figure 4 which is a schematic diagram of the module structure of the computer room temperature prediction device in the embodiment of the present application. The computer room temperature prediction device includes:

[0122] A base learner module 401, configured to use LightGBM, LSTM, and NeuralProphet as the initial base learners, based on the computer room temperature feature data, train the initial base learners through five-fold cross-validation, and use the FA algorithm to optimize the hyperparameters of the initial base learners to obtain the target base learners;

[0123] A weighting module 402, configured to weight the prediction results of the target base learners based on the preset IEM algorithm;

[0124] A meta-learner module 403, configured to use the weighted prediction results as input features, and based on the input features, construct multiple linear regression to obtain the meta-learner;

[0125] A target module 404, configured to construct a Stacking computer room temperature prediction model based on the target base learners and the meta-learner.

[0126] The computer room temperature prediction device provided by the embodiment of the present application adopts the computer room temperature prediction method in the above embodiment, and can solve the technical problem of how to improve the accuracy of computer room temperature prediction. Compared with the prior art, the beneficial effects of the computer room temperature prediction device provided by the embodiment of the present application are the same as those of the computer room temperature prediction method provided by the above embodiment, and other technical features in the computer room temperature prediction device are the same as the features disclosed in the above embodiment method, which will not be elaborated here.

[0127] The present application provides a computer room temperature prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the computer room temperature prediction method in the above embodiment.

[0128] The following refers to Figure 5 , Figure 5 which is a schematic structural diagram of a device for the hardware operating environment involved in the computer room temperature prediction method in the embodiment of the present application, and shows a schematic structural diagram of a computer room temperature prediction device suitable for implementing the embodiment of the present application. Figure 5 The shown computer room temperature prediction device is only an example and should not bring any limitation to the functions and usage scope of the embodiment of the present application.

[0129] Such as Figure 5As shown, the computer room temperature prediction device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which may perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 into the random access memory (RAM: Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the computer room temperature prediction device are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems may be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the computer room temperature prediction device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a computer room temperature prediction device with various systems, it should be understood that it is not required to implement or have all the shown systems. Instead, more or fewer systems may be implemented or had.

[0130] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program codes for executing the methods shown in the flowcharts. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0131] The computer room temperature prediction device provided by the present application adopts the computer room temperature prediction method in the above embodiments, and can solve the technical problem of how to improve the accuracy of computer room temperature prediction. Compared with the prior art, the beneficial effects of the computer room temperature prediction device provided by the present application are the same as those of the computer room temperature prediction method provided by the above embodiments, and other technical features in the computer room temperature prediction device are the same as those disclosed in the method of the previous embodiment, and will not be elaborated here.

[0132] It should be understood that the various parts disclosed in the present application may be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in a suitable manner in any one or more embodiments or examples.

[0133] As described above, this is only the specific implementation manner of the present application. However, the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all of them should be covered by the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

[0134] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the computer room temperature prediction method in the above embodiments.

[0135] The above computer-readable storage medium carries one or more programs. When the above one or more programs are executed by a computer room temperature prediction device, the computer room temperature prediction device is caused to: use LightGBM, LSTM, and NeuralProphet as initial base learners, based on the computer room temperature feature data, train the initial base learners through five-fold cross-validation, and optimize the hyperparameters of the initial base learners by using the FA algorithm to obtain target base learners; based on a preset IEM algorithm, perform weighted processing on the prediction results of the target base learners; use the weighted prediction results as input features, and based on the input features, construct a multiple linear regression to obtain a meta-learner; based on the target base learners and the meta-learner, construct a Stacking computer room temperature prediction model. Computer program code for performing the operations of the present application can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages such as Java, Smalltalk, and C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0136] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks may occur in a different order than that noted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0137] The modules described in the embodiments of the present application can be implemented in software or in hardware. In some cases, the name of the module does not constitute a limitation on the unit itself.

[0138] The readable storage medium provided by the present application is a computer-readable storage medium, and the computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned computer room temperature prediction method, which can solve the technical problem of how to improve the accuracy of computer room temperature prediction. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the computer room temperature prediction method provided by the above embodiments, and will not be elaborated here.

[0139] An embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the computer room temperature prediction method as described above are implemented.

[0140] The computer program product provided by the present application can solve the technical problem of how to improve the accuracy of computer room temperature prediction. Compared with the prior art, the beneficial effects of the computer program product provided by the embodiments of the present application are the same as those of the computer room temperature prediction method provided by the above embodiments, and will not be elaborated here.

[0141] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be included in the patent scope of the present application by the same token.

Claims

1. A method for predicting temperature in a computer room, characterized in that: The method comprises: LightGBM, LSTM, and NeuralProphet are used as initial base learners. Based on the temperature feature data of the computer room, the initial base learners are trained through five-fold cross validation, and the FA algorithm is used to optimize the hyperparameters of the initial base learners to obtain the target base learner. Based on a preset IEM algorithm, weighted processing is performed on the prediction results of the target base learner; The weighted prediction results are used as input features, and based on the input features, a multivariate linear regression is constructed to obtain a meta-learner; Based on the target base learner and the meta learner, a temperature prediction model for a stacking computer room is constructed.

2. The method according to claim 1, characterized in that Before the step of using LightGBM, LSTM, and NeuralProphet as initial base learners, training the initial base learners through five-fold cross validation based on the temperature feature data of the computer room, and optimizing the hyperparameters of the initial base learners using the FA algorithm to obtain the target base learner, the step further includes: Obtain historical temperature data and environmental data of the computer room; Preprocessing the historical temperature data of the computer room and the environmental data to obtain preprocessed data, wherein the preprocessing includes one or more of data cleaning, timestamp processing, and data normalization; Time series feature extraction is performed on the preprocessed data to obtain the temperature feature data of the computer room.

3. The method according to claim 1, characterized in that The step of using LightGBM, LSTM, and NeuralProphet as initial base learners, training the initial base learners through five-fold cross validation based on the temperature feature data of the computer room, and optimizing the hyperparameters of the initial base learners using the FA algorithm to obtain the target base learner includes: LightGBM, LSTM, and NeuralProphet are used as initial base learners, and based on the five-fold cross validation, the temperature feature data of the computer room is divided into a validation set and a training set; Based on the validation set and the training set, training the initial base learner; Based on the training result, the FA algorithm is used to perform position update and iteration; Determining optimal hyperparameters according to the results of the position update and iteration; Based on the optimal hyperparameters, the initial base learner is retrained to obtain the target base learner.

4. The method according to claim 3, characterized in that The step of weighting the prediction result of the target base learner based on the preset IEM algorithm includes: Based on a preset weight allocation strategy, assigning initial weights to the target base learner; Obtaining a prediction error of the target base learner on the validation set; Based on the prediction error, the preset IEM algorithm is used to adjust the weight of the target basis learner corresponding to the prediction error, and the weighted processing is performed on the adjusted weight.

5. The method according to claim 4, characterized in that The step of using the weighted prediction results as input features and constructing a multivariate linear regression based on the input features to obtain a meta-learner includes: The weighted prediction results are used as input features, and based on the sample order, the input features are sorted into a feature matrix, where each column of the feature matrix corresponds to a prediction output of a target base learner; Using the actual observation value of the initial base learner as the target variable, and pairing the feature matrix with the target variable; Based on the pairing results, a preset multiple linear regression method is used to fit the feature matrix and the target variable to obtain the meta-learner.

6. The method according to claim 1, characterized in that The step of constructing a temperature prediction model for a stacking computer room based on the target base learner and the meta learner includes: Training the meta-learner based on the prediction results and actual observations of the target base learner; Based on the training results, learning the mapping relationship between the prediction results of the target base learner and the actual observation values; Based on the mapping relationship, the target base learner and the meta learner are combined into the stacking room temperature prediction model.

7. A device for predicting temperature in a computer room, characterized in that: The device comprises: The base learner module is used to use LightGBM, LSTM, and NeuralProphet as initial base learners, train the initial base learners through five-fold cross validation based on the temperature feature data of the computer room, and use the FA algorithm to optimize the hyperparameters of the initial base learners to obtain the target base learner; A weighting module, used for performing weighted processing on the prediction result of the target base learner based on a preset IEM algorithm; A meta-learner module, used to use the weighted prediction results as input features, and construct a multivariate linear regression based on the input features to obtain a meta-learner; The target module is used to build a temperature prediction model for a Stacking computer room based on the target base learner and the meta learner.

8. A computer device, characterized in that: The device comprises: a memory, a processor, and a computer room temperature prediction program stored in the memory and executable on the processor, wherein the computer room temperature prediction program is configured to implement the steps of the computer room temperature prediction method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores a computer room temperature prediction program, which, when executed by a processor, implements the steps of the computer room temperature prediction method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the method for predicting the temperature of a computer room according to any one of claims 1 to 6 are implemented.

Images