Machine room space energy consumption prediction method based on long and short term memory network

Abstract

The invention provides a machine room space energy consumption prediction method and system based on a long and short term memory network, and the method comprises the following steps: (a) constructing a three-dimensional digital twin model of a target machine room, and generating time sequence simulation data based on computational fluid dynamics (CFD) simulation; (b) preprocessing the time sequence simulation data; (c) constructing and training a long-short term memory (LSTM) network model; (d) operating parameters based on the current machine room; and (e) displaying the energy-saving operation suggestion and the operation state of the machine room in real time through a visual interface to assist an administrator to make a decision. According to the method, the computer room three-dimensional model is constructed by utilizing the digital twinning technology, time sequence data covering normal and abnormal working conditions (such as air conditioner faults and extreme temperatures) is generated in combination with CFD simulation, the problem of model overfitting caused by insufficient real data in a traditional method is solved, the training sample range is expanded through simulation data, and the training efficiency is improved. The adaptive capacity of the LSTM model to a complex scene is enhanced, and the generalization of a prediction result and the robustness of a system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer room space energy consumption prediction, and in particular to a computer room space energy consumption prediction method based on a long short-term memory network. Background Art

[0002] Traditional data center energy management relies primarily on sensor-based linear control algorithms and static prediction models. These methods have significant limitations when dealing with complex and changing computer room environments (such as dynamic load changes, non-uniform temperature distribution, and sudden equipment failures). Specifically,

[0003] 1. Data sample limitations:

[0004] Real computer room operating data usually only covers routine operating conditions within a safe range, and lacks data on abnormal scenarios (such as air conditioning failures and extreme high temperatures). This results in a single model training sample and makes it difficult to predict energy consumption and risks in extreme situations.

[0005] Neural networks trained on real data are prone to overfitting, lack generalization capabilities, and are unable to adapt to complex nonlinear dynamic systems in reality.

[0006] 2. Static nature and experience dependence of control strategy:

[0007] Traditional methods are mostly based on fixed rules (such as setting a constant temperature threshold) or manual experience to adjust air conditioning parameters. They lack dynamic optimization capabilities and find it difficult to balance energy efficiency (PUE optimization) and safety (overheating risk avoidance) in real time.

[0008] Existing algorithms have a weak ability to capture time series features and cannot effectively model the long-term dependence between energy consumption and temperature changes in computer rooms.

[0009] 3. Lack of visualization and management tools:

[0010] Key parameters such as the computer room's airflow organization and temperature distribution lack intuitive visualization, making it difficult for administrators to quickly locate hot spots or evaluate the effectiveness of control strategies.

[0011] The decision-making process relies on offline analysis and manual intervention, which has a slow response speed and cannot meet the energy consumption optimization needs with high real-time requirements.

[0012] Therefore, a method for predicting computer room space energy consumption based on long short-term memory network is proposed. Summary of the Invention

[0013] The present invention aims to solve the problems raised in the background technology and provides a method for predicting energy consumption of computer room space based on long short-term memory network.

[0014] The specific technical solutions are as follows:

[0015] A method for predicting energy consumption of a computer room space based on a long short-term memory network includes the following steps:

[0016] (a) Build a three-dimensional digital twin model of the target computer room and generate time-series simulation data based on computational fluid dynamics (CFD) simulation. The time-series simulation data includes temperature distribution, power usage effectiveness (PUE), and air conditioning operating parameters.

[0017] (b) preprocessing the time series simulation data, including partitioning the data set through a sliding window, performing wavelet transform denoising and feature extraction, to generate training data;

[0018] (c) constructing and training a long short-term memory (LSTM) network model, inputting the training data, and outputting a PUE prediction value and overheating probability;

[0019] (d) Based on the current operating parameters of the computer room, the LSTM model is used to predict the PUE and overheating probability after shutting down an air conditioner or increasing the air conditioner temperature set point, and generate energy-saving operation recommendations that meet safety criteria;

[0020] (e) The energy-saving operation suggestions and the operating status of the computer room are displayed in real time through a visual interface to assist the administrator in making decisions.

[0021] The above-mentioned computer room space energy consumption prediction method based on long short-term memory network, wherein the LSTM network model in step (c) includes 1 input layer, 2 LSTM layers, 1 Dropout layer, 1 fully connected layer and 1 output layer, wherein the number of time step neurons in the LSTM layer is 64, the dropout rate of the Dropout layer is 20%, the activation function adopts a combination of tanh and sigmoid, and the output layer adopts the ReLU function.

[0022] In the above-mentioned method for predicting computer room space energy consumption based on long short-term memory network, the length of the sliding window in step (b) is 64, the sliding step is 5, and the prediction step is 10; the wavelet transform is used to decompose the original data into coefficients of different frequency bands, and select low-frequency coefficients as effective features to be input into the LSTM model.

[0023] In the above-mentioned method for predicting computer room space energy consumption based on long short-term memory networks, the logic for generating energy-saving operation suggestions in step (d) includes:

[0024] (d1) If the PUE after shutting down one air conditioner is lower than the current PUE without causing cabinet overheating, and there are more than three air conditioners currently in operation, it is recommended to shut down one air conditioner;

[0025] (d2) If the PUE after increasing the air conditioner temperature set point is lower than the current PUE and does not cause overheating, it is recommended to increase the temperature set point;

[0026] (d3) If none of the above conditions are met, it is recommended to maintain the current operating status.

[0027] The above-mentioned method for predicting computer room space energy consumption based on long short-term memory network, wherein the visualization interface is constructed by Unity3D, includes:

[0028] (e1) Cabinet temperature and air conditioning sensible cooling power are dynamically displayed in the form of color mapping;

[0029] (e2) Operation suggestions are pushed in real time through the interactive panel;

[0030] (e3) Air conditioning control parameters are adjusted through sliders, and the airflow organization of the computer room can be observed from multiple perspectives.

[0031] In the above-mentioned method for predicting computer room space energy consumption based on long short-term memory networks, the CFD simulation data covers the operating status of the computer room under abnormal working conditions, including air conditioning failure or extreme ambient temperature, to enhance the generalization ability of the LSTM model.

[0032] The present invention further provides a computer room space energy consumption prediction system based on a long short-term memory network, which is used to implement the above-mentioned computer room space energy consumption prediction method based on a long short-term memory network, comprising:

[0033] The digital twin module is used to build a three-dimensional model of the computer room and generate time-series simulation data through CFD simulation;

[0034] Data processing module, used for sliding window division, wavelet denoising and feature extraction of simulation data;

[0035] LSTM model training module, used to build and train a neural network to predict PUE and overheating probability;

[0036] Run the suggestion generation module to generate energy-saving control strategies based on the model prediction results;

[0037] Visual interaction module, used to display the operating status and suggestions of the computer room in real time.

[0038] In the above-mentioned computer room space energy consumption prediction system based on long short-term memory network, the LSTM model training module adopts the Adam optimization algorithm, the learning rate is 0.001, the maximum number of iterations is 3000, and the batch size is 64.

[0039] In the above-mentioned computer room space energy consumption prediction system based on long short-term memory network, the input of the operation suggestion generation module includes the current PUE, the number of air conditioners in operation and the temperature set point, and the output is one of "turning off one air conditioner", "raising the temperature set point" or "maintaining the current state".

[0040] The present invention also provides a computer-readable storage medium storing a computer program, which implements the steps of the above method when executed.

[0041] The present invention has the following beneficial effects:

[0042] 1. Improved data generation and generalization capabilities:

[0043] Digital twin technology is used to build a three-dimensional model of the computer room, and combined with CFD simulation to generate time series data covering normal and abnormal operating conditions (such as air conditioning failure and extreme temperatures), which solves the problem of model overfitting caused by insufficient real data in traditional methods.

[0044] By expanding the range of training samples through simulation data, the adaptability of the LSTM model to complex scenarios is enhanced, and the generalization of prediction results and system robustness are improved.

[0045] 2. Time series data processing and model optimization:

[0046] The sliding window data division, wavelet transform denoising and feature extraction technology are used to effectively extract key information from time series data, reduce noise interference, improve the quality of input data, and thus improve the model prediction accuracy.

[0047] Through the multi-layer structure of the LSTM network (input layer, LSTM layer, Dropout layer, etc.) and its parameter configuration (such as the number of neurons, activation function combination), the model's ability to capture long-term dependencies is enhanced. At the same time, the Dropout mechanism is used to suppress overfitting and ensure the stability and reliability of the prediction results.

[0048] 3. Intelligent balance between safety and energy efficiency:

[0049] Based on the decision logic of safety criteria (such as the lower limit of the number of air conditioners and the overheating probability threshold), energy-saving operation suggestions are dynamically generated to reduce PUE while avoiding the risk of equipment overheating, achieving coordinated control of energy efficiency optimization and safe operation.

[0050] By predicting the energy consumption and risks of different control strategies (turning off the air conditioner or adjusting the temperature set point), the optimal solution is provided to avoid inefficiencies or misoperations caused by relying on experience.

[0051] 4. Improved interactivity and management efficiency:

[0052] The visualization interface uses Unity3D to achieve dynamic mapping of the computer room's operating parameters (such as temperature color distribution and multi-perspective display of airflow organization), converting complex data into intuitive visual information to help administrators quickly locate problems and respond.

[0053] Interactive functions (such as sliders to adjust parameters and real-time suggestion push) simplify operational processes, improve decision-making efficiency, and reduce manual management costs.

[0054] 5. Systematization and standardization support:

[0055] The modular design (digital twin, data processing, model training, decision generation, and visualization) forms a complete closed loop, ensuring the scalability and operability of the technical solution and facilitating adaptation to the customized needs of computer rooms of different sizes.

[0056] By solidifying the method process through computer-readable storage media, the rapid deployment and standardized application of technical solutions can be achieved, enhancing practicality and promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flowchart of a method for predicting energy consumption in a computer room based on a long short-term memory network is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The technical solution of the present invention will be further described below with reference to the accompanying drawings and through specific implementation methods.

[0059] Among them, the drawings are only used for illustrative purposes and represent only schematic diagrams rather than actual pictures, and should not be understood as limiting this patent; in order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0060] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "inside", "outside" and the like indicate an orientation or position relationship based on the orientation or position relationship shown in the drawings, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting this patent. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0061] In the description of the present invention, unless otherwise expressly specified or limited, when the term "connection" or the like appears to indicate a connection relationship between components, such term should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be internal communication between two components or an interaction between two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood in specific circumstances.

[0062] Example

[0063] The method for predicting energy consumption of computer room space based on long short-term memory network provided in this embodiment is as follows: Figure 1 As shown, the following steps are included:

[0064] (a) Build a 3D digital twin model of the target computer room and generate time-series simulation data based on computational fluid dynamics (CFD) simulation. The time-series simulation data includes temperature distribution, power usage effectiveness (PUE), and air conditioning operating parameters.

[0065] (b) Preprocessing the time series simulation data, including partitioning the data set through sliding windows, wavelet transform denoising and feature extraction to generate training data;

[0066] (c) Build and train a long short-term memory (LSTM) network model, input training data, and output PUE prediction value and overheating probability;

[0067] (d) Based on the current computer room operating parameters, the LSTM model is used to predict the PUE and overheating probability after shutting down an air conditioner or raising the air conditioner temperature set point, generating energy-saving operation recommendations that meet safety criteria;

[0068] (e) Energy-saving operation suggestions and computer room operation status are displayed in real time through a visual interface to assist administrators in making decisions.

[0069] By adopting the above technical solution, multi-dimensional time series data is generated through digital twin models and CFD simulation. Energy consumption and overheating risks are predicted by LSTM networks, and energy-saving suggestions are dynamically generated and visualized. This achieves accurate prediction and safe optimization control of computer room energy consumption, thereby improving energy management efficiency.

[0070] The LSTM network model in step (c) includes 1 input layer, 2 LSTM layers, 1 Dropout layer, 1 fully connected layer and 1 output layer, wherein the number of time step neurons in the LSTM layer is 64, the dropout rate of the Dropout layer is 20%, the activation function adopts a combination of tanh and sigmoid, and the output layer adopts the ReLU function.

[0071] By adopting the above technical solution, through specific network structure and parameter configuration (such as LSTM layer, Dropout layer and activation function), the model's temporal learning ability is enhanced, overfitting is prevented, and the stability and reliability of the prediction results are improved.

[0072] In step (b), the length of the sliding window is 64, the sliding step is 5, and the prediction step is 10. Wavelet transform is used to decompose the original data into coefficients of different frequency bands, and the low-frequency coefficients are selected as effective features to be input into the LSTM model.

[0073] By adopting the above technical solution, through sliding window division and wavelet transform processing, the feature extraction quality of time series data is optimized, noise interference is reduced, the effectiveness of model input is improved, and thus the prediction accuracy is improved.

[0074] The logic for generating energy-saving operation suggestions in step (d) includes:

[0075] (d1) If the PUE after shutting down one air conditioner is lower than the current PUE without causing cabinet overheating, and there are more than three air conditioners currently in operation, it is recommended to shut down one air conditioner;

[0076] (d2) If the PUE after increasing the air conditioner temperature set point is lower than the current PUE and does not cause overheating, it is recommended to increase the temperature set point;

[0077] (d3) If none of the above conditions are met, it is recommended to maintain the current operating status.

[0078] By adopting the above technical solution and decision-making logic based on safety criteria (such as limiting the number of air conditioners and overheating judgment), energy-saving suggestions can be ensured to reduce PUE while avoiding the risk of equipment overheating, achieving dual protection of safety and energy efficiency.

[0079] The visual interface is built using Unity3D and includes:

[0080] (e1) Cabinet temperature and air conditioning sensible cooling power are dynamically displayed in the form of color mapping;

[0081] (e2) Operation suggestions are pushed in real time through the interactive panel;

[0082] (e3) Air conditioning control parameters are adjusted through sliders, and the airflow organization of the computer room can be observed from multiple perspectives.

[0083] By adopting the above technical solution, the operating status of the computer room (such as temperature distribution and control parameters) is dynamically mapped through a visual interface, providing intuitive multi-perspective interactive functions to help administrators quickly understand complex data and make efficient decisions.

[0084] Among them, CFD simulation data covers the operating status of the computer room under abnormal working conditions, including air conditioning failure or extreme ambient temperature, to enhance the generalization ability of the LSTM model.

[0085] By adopting the above technical solution and training the model with simulation data covering abnormal working conditions, the system's predictive adaptability to extreme scenarios (such as air conditioning failure and high temperature environment) is enhanced, and the model generalization and system robustness are improved.

[0086] This embodiment further provides a computer room space energy consumption prediction system based on a long short-term memory network, which is used to implement the above-mentioned computer room space energy consumption prediction method based on a long short-term memory network, including:

[0087] The digital twin module is used to build a three-dimensional model of the computer room and generate time-series simulation data through CFD simulation;

[0088] Data processing module, used for sliding window division, wavelet denoising and feature extraction of simulation data;

[0089] LSTM model training module, used to build and train a neural network to predict PUE and overheating probability;

[0090] Run the suggestion generation module to generate energy-saving control strategies based on the model prediction results;

[0091] Visual interaction module, used to display the operating status and suggestions of the computer room in real time.

[0092] By adopting the above technical solution and integrating multi-link technical means through modular system design (digital twin, data processing, model training, etc.), a complete closed loop from data generation to decision support is achieved, thereby improving the operability and scalability of the method.

[0093] The LSTM model training module uses the Adam optimization algorithm, with a learning rate of 0.001, a maximum number of iterations of 3000, and a batch size of 64.

[0094] By adopting the above technical solution, through optimizing algorithms (such as Adam) and training parameter settings, the model convergence speed is accelerated, the training efficiency is improved, and the global optimality of the model parameters is ensured.

[0095] The inputs of the operation suggestion generation module include the current PUE, the number of air conditioners in operation, and the temperature set point, and the output is one of "turn off one air conditioner", "increase the temperature set point", or "maintain the current state".

[0096] By adopting the above technical solution, by clarifying the input and output rules (such as the number of air conditioners and temperature set points), the decision-making logic is simplified, the accuracy and feasibility of operation suggestions are improved, and the risk of misoperation is reduced.

[0097] This embodiment also provides a computer-readable storage medium storing a computer program. When the program is executed, the steps of the above-mentioned method for predicting the energy consumption of a computer room space based on a long short-term memory network are implemented. The method flow is solidified by the computer-readable storage medium to ensure the standardized implementation and rapid deployment of the technical solution, thereby enhancing the practicality and promotion of the method.

[0098] In summary, the method and system for predicting computer room space energy consumption based on the long short-term memory network proposed in this embodiment achieve multi-dimensional optimization effects through the following technical solutions:

[0099] 1. Improved data generation and generalization capabilities:

[0100] Digital twin technology is used to build a three-dimensional model of the computer room, and combined with CFD simulation to generate time series data covering normal and abnormal operating conditions (such as air conditioning failure and extreme temperatures), which solves the problem of model overfitting caused by insufficient real data in traditional methods.

[0101] By expanding the range of training samples through simulation data, the adaptability of the LSTM model to complex scenarios is enhanced, and the generalization of prediction results and system robustness are improved.

[0102] 2. Time series data processing and model optimization:

[0103] The sliding window data division, wavelet transform denoising and feature extraction technology are used to effectively extract key information from time series data, reduce noise interference, improve the quality of input data, and thus improve the model prediction accuracy.

[0104] Through the multi-layer structure of the LSTM network (input layer, LSTM layer, Dropout layer, etc.) and its parameter configuration (such as the number of neurons, activation function combination), the model's ability to capture long-term dependencies is enhanced. At the same time, the Dropout mechanism is used to suppress overfitting and ensure the stability and reliability of the prediction results.

[0105] 3. Intelligent balance between safety and energy efficiency:

[0106] Based on the decision logic of safety criteria (such as the lower limit of the number of air conditioners and the overheating probability threshold), energy-saving operation suggestions are dynamically generated to reduce PUE while avoiding the risk of equipment overheating, achieving coordinated control of energy efficiency optimization and safe operation.

[0107] By predicting the energy consumption and risks of different control strategies (turning off the air conditioner or adjusting the temperature set point), the optimal solution is provided to avoid inefficiencies or misoperations caused by relying on experience.

[0108] 4. Improved interactivity and management efficiency:

[0109] The visualization interface uses Unity3D to achieve dynamic mapping of the computer room's operating parameters (such as temperature color distribution and multi-perspective display of airflow organization), converting complex data into intuitive visual information to help administrators quickly locate problems and respond.

[0110] Interactive functions (such as sliders to adjust parameters and real-time suggestion push) simplify operational processes, improve decision-making efficiency, and reduce manual management costs.

[0111] 5. Systematization and standardization support:

[0112] The modular design (digital twin, data processing, model training, decision generation, and visualization) forms a complete closed loop, ensuring the scalability and operability of the technical solution and facilitating adaptation to the customized needs of computer rooms of different sizes.

[0113] By solidifying the method process through computer-readable storage media, the rapid deployment and standardized application of technical solutions can be achieved, enhancing practicality and promotion.

[0114] Summary of working principle process:

[0115] 1. Data generation and model building stage:

[0116] Step 1: Build a 3D digital twin model of the target computer room. Use CFD simulation to simulate the temperature distribution, PUE, and air conditioning operating parameters of the computer room under different operating conditions (normal and abnormal), and generate simulation data with time series characteristics.

[0117] Step 2: Preprocess the simulation data, including:

[0118] Step 2.1 Sliding window division: Divide the continuous time series data into segments of fixed length (such as 64 time steps) to form model input samples.

[0119] Step 2.2 Wavelet transform denoising: decompose the data into different frequency bands, filter out high-frequency noise, and extract low-frequency effective features as model input.

[0120] Step 2.3 Feature extraction: Select key parameters (such as temperature gradient and air conditioning load) as training data to improve the input information density.

[0121] 2. Model training and prediction stage:

[0122] Step 3: Build an LSTM network model. Its structure includes an input layer (receiving preprocessed data), a dual LSTM layer (capturing temporal dependencies), a dropout layer (randomly discarding some neurons to prevent overfitting), a fully connected layer (integrating features), and an output layer (predicting PUE and overheating probability).

[0123] Step 4: The Adam optimization algorithm (learning rate 0.001) is used to train the model, and the parameters are updated through back propagation. The maximum number of iterations (3000) ensures model convergence, and the batch size (64) balances training efficiency and memory usage.

[0124] 3. Run the suggestion generation phase:

[0125] Step 5: Based on the current computer room operating parameters (such as the number of air conditioners, temperature set point, and real-time PUE), the trained LSTM model is used to predict the PUE change and overheating risk of two energy-saving strategies (turning off one air conditioner and raising the temperature set point).

[0126] Step 6: Generate recommendations based on security criteria:

[0127] If shutting down an air conditioner can reduce PUE without triggering overheating, and the current number of air conditioners is greater than 3, it is recommended to shut down one air conditioner.

[0128] If increasing the temperature set point can reduce PUE without risking overheating, then adjusting the temperature is recommended;

[0129] If none of the above conditions are met, maintain the current state to ensure safety first.

[0130] 4. Visualization and decision support stage:

[0131] Step 7: Use the Unity3D visualization interface to display the computer room's operating status in real time:

[0132] Cabinet temperature and air conditioning power are dynamically mapped in the form of heat maps, intuitively reflecting local hot spots and energy consumption distribution;

[0133] Operation suggestions are pushed through the interactive panel, and administrators can adopt them with one click or manually adjust parameters (such as adjusting air conditioning temperature with a slider);

[0134] The multi-view observation function allows administrators to view the airflow organization in the computer room, helping to optimize equipment layout or adjust cooling strategies.

[0135] 5. Closed-loop optimization and expanded applications:

[0136] The system continuously collects real-time data from the computer room, updates the digital twin model, and iteratively trains the LSTM network to form a dynamic optimization closed loop;

[0137] The modular design supports functional expansion (such as new sensor data access and multi-computer room collaborative management) to adapt to future technology upgrade needs.

[0138] Core logic closed loop:

[0139] Data generation → feature optimization → model prediction → security decision-making → visual feedback → dynamic iteration form a complete chain from simulation to actual application, realizing intelligent and refined energy consumption management of the computer room.

[0140] The above are only preferred embodiments of the present invention and do not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting energy consumption of computer room space based on long short-term memory network, characterized in that: The following steps are involved: (a) Build a three-dimensional digital twin model of the target computer room and generate time series simulation data based on computational fluid dynamics (CFD) simulation. The time series simulation data includes temperature distribution, power usage efficiency (PUE), and air conditioning operating parameters. (b) preprocessing the time series simulation data, including partitioning the data set through a sliding window, performing wavelet transform denoising and feature extraction, to generate training data; (c) constructing and training a long short-term memory (LSTM) network model, inputting the training data, and outputting a PUE prediction value and overheating probability; (d) Based on the current operating parameters of the computer room, the LSTM model is used to predict the PUE and overheating probability after shutting down an air conditioner or increasing the air conditioner temperature set point, and generate energy-saving operation recommendations that meet safety criteria; (e) The energy-saving operation suggestions and the operating status of the computer room are displayed in real time through a visual interface to assist the administrator in making decisions.

2. The method for predicting energy consumption of computer room space based on long short-term memory network according to claim 1 is characterized in that: The LSTM network model in step (c) includes 1 input layer, 2 LSTM layers, 1 Dropout layer, 1 fully connected layer and 1 output layer, wherein the number of time step neurons in the LSTM layer is 64, the dropout rate of the Dropout layer is 20%, the activation function adopts a combination of tanh and sigmoid, and the output layer adopts the ReLU function.

3. The method for predicting energy consumption of a computer room space based on a long short-term memory network according to claim 1 is characterized in that: The length of the sliding window in step (b) is 64, the sliding step is 5, and the prediction step is 10; the wavelet transform is used to decompose the original data into coefficients of different frequency bands, and select low-frequency coefficients as effective features to input into the LSTM model.

4. The method for predicting energy consumption of a computer room space based on a long short-term memory network according to claim 1 is characterized in that: The logic for generating energy-saving operation suggestions in step (d) includes: (d1) If the PUE after shutting down one air conditioner is lower than the current PUE without causing cabinet overheating, and there are more than three air conditioners currently in operation, it is recommended to shut down one air conditioner; (d2) If the PUE after increasing the air conditioner temperature set point is lower than the current PUE and does not cause overheating, it is recommended to increase the temperature set point; (d3) If none of the above conditions are met, it is recommended to maintain the current operating status.

5. The method for predicting computer room space energy consumption based on long short-term memory network according to claim 1 is characterized in that: The visualization interface is built using Unity3D and includes: (e1) Cabinet temperature and air conditioning sensible cooling power are dynamically displayed in the form of color mapping; (e2) Operation suggestions are pushed in real time through the interactive panel; (e3) Air conditioning control parameters are adjusted through sliders, and the airflow organization of the computer room can be observed from multiple perspectives.

6. The method for predicting energy consumption of a computer room space based on a long short-term memory network according to claim 1 is characterized in that: The CFD simulation data covers the operating status of the computer room under abnormal working conditions, including air conditioning failure or extreme ambient temperature, to enhance the generalization ability of the LSTM model.

7. A computer room space energy consumption prediction system based on long short-term memory network, characterized in that: The method for predicting energy consumption of a computer room space based on a long short-term memory network according to any one of claims 1 to 6 comprises: The digital twin module is used to build a three-dimensional model of the computer room and generate time-series simulation data through CFD simulation; Data processing module, used for sliding window division, wavelet denoising and feature extraction of simulation data; LSTM model training module, used to build and train a neural network to predict PUE and overheating probability; Run the suggestion generation module to generate energy-saving control strategies based on the model prediction results; Visual interaction module, used to display the operating status and suggestions of the computer room in real time.

8. The computer room space energy consumption prediction system based on long short-term memory network according to claim 7 is characterized in that: The LSTM model training module adopts the Adam optimization algorithm, with a learning rate of 0.001, a maximum number of iterations of 3000, and a batch size of 64.

9. The computer room space energy consumption prediction system based on long short-term memory network according to claim 7 is characterized in that: The input of the operation suggestion generation module includes the current PUE, the number of air conditioners in operation, and the temperature set point, and the output is one of "turn off one air conditioner", "increase the temperature set point" or "maintain the current state".

10. A computer-readable storage medium storing a computer program, characterized in that: When the program is executed, the steps of the method according to any one of claims 1 to 5 are implemented.

Images