Liquid cooling equipment control method and system, electronic equipment and storage medium

Abstract

The invention provides a liquid cooling equipment control method and system, electronic equipment and a storage medium, and relates to the technical field of cooling control, and the method comprises the steps: collecting the state data of a temperature and humidity sensor; predicting the temperature and humidity data according to the state data of the temperature and humidity sensor; based on the prediction result of the temperature and humidity data, predicting the state of the liquid cooling equipment and a plurality of corresponding control actions to obtain a predicted state of the liquid cooling equipment and a plurality of corresponding predicted control actions; based on the prediction state of the liquid cooling equipment, determining a target control action which most accords with a preset energy consumption condition and a safety condition from the plurality of prediction control actions; and executing the target control action on the liquid cooling equipment. The intelligent level of environment monitoring is improved, the energy efficiency, the stability, the safety and the like are remarkably improved, and the system has wide application prospects and market value.

Description

Technical Field

[0001] The present disclosure relates to the field of cooling control technology, and in particular to a liquid cooling device control method, system, electronic device and storage medium. Background Art

[0002] With the rapid development of information technology and the widespread use of data centers, the dense deployment of servers and other electronic equipment has led to a large amount of heat generated in the computer room. Traditional air cooling systems often find it difficult to meet the growing heat dissipation needs when dealing with high-density servers, which in turn affects the performance and stability of the equipment. Liquid cooling technology has gradually become an effective solution for data center cooling due to its superior thermal conductivity and heat capacity.

[0003] However, the control methods of liquid cooling equipment in related technologies mostly rely on manual intervention and lack intelligent control methods, resulting in low cooling efficiency and energy waste. In addition, existing control strategies often fail to make full use of real-time monitoring data, resulting in slow response speed and inaccurate adjustment of the cooling system. Therefore, there is an urgent need for an intelligent control method based on temperature and humidity sensor data that can predict environmental changes in real time and automatically adjust the operating status of liquid cooling equipment. Summary of the invention

[0004] The embodiments of the present disclosure provide a liquid cooling device control method, system, electronic device and storage medium, aiming to solve the problems existing in the above-mentioned background technology.

[0005] In order to solve the above technical problems, the present disclosure is implemented as follows: In a first aspect, an embodiment of the present disclosure provides a liquid cooling device control method, the method comprising: Collect status data of temperature and humidity sensors; Predicting temperature and humidity data according to the status data of the temperature and humidity sensor; Based on the prediction result of the temperature and humidity data, predict the state of the liquid cooling device and the corresponding multiple control actions to obtain the predicted state of the liquid cooling device and the corresponding multiple predicted control actions; Based on the predicted state of the liquid cooling device, determining a target control action that best meets the preset energy consumption conditions and safety conditions from the multiple predicted control actions; The target control action is performed on the liquid cooling device.

[0006] Optionally, predicting the temperature and humidity data according to the state data of the temperature and humidity sensor includes: Monitoring the collected status data of the temperature and humidity sensor; When it is detected that the state data of the temperature and humidity sensor exceeds a preset temperature threshold or humidity threshold, feature extraction is performed on the state data of the temperature and humidity sensor, and a time feature is extracted from the state data of the temperature and humidity sensor; Inputting the state data of the temperature and humidity sensor after feature extraction into an LSTM time series model, wherein the LSTM time series model is pre-trained using the historical state data of the temperature and humidity sensor collected within a preset period; The temperature and humidity data are predicted by the LSTM time series model.

[0007] Optionally, train the LSTM time series model according to the following steps: Based on the time feature, the historical state data of the temperature and humidity sensor is divided into multiple time series data; Sorting the multiple time series data in chronological order; Dividing the sorting results of the multiple time series data into a training set and a test set according to a preset ratio; Constructing the LSTM time series model, and training the LSTM time series model using the training set; After each training round, the LSTM time series model is tested using the test set to determine the accuracy of the LSTM time series model; When the accuracy of the LSTM time series model does not reach a preset threshold, the hyperparameters of the LSTM time series model are adjusted, and the steps are performed: the LSTM time series model is trained using the training set until the accuracy of the LSTM time series model reaches a preset threshold.

[0008] Optionally, the state of the liquid cooling device and the corresponding multiple control actions are predicted based on the prediction result of the temperature and humidity data to obtain the predicted state of the liquid cooling device and the corresponding multiple predicted control actions, including: Inputting the prediction result of the temperature and humidity data into a recurrent neural network model, wherein the recurrent neural network model is pre-trained using the historical status data of the temperature and humidity sensor and the liquid cooling device collected within a preset period; Predicting the state of the liquid cooling device and corresponding multiple control actions through the recurrent neural network model; A feature vector output by the recurrent neural network is obtained, wherein the feature vector includes a predicted state of the liquid cooling device and a corresponding plurality of predicted control actions.

[0009] Optionally, the determining, based on the predicted state of the liquid cooling device, a target control action that best meets preset energy consumption conditions and safety conditions from the multiple predicted control actions includes: Applying a Q-learning algorithm to a feature vector output by the recurrent neural network, and determining a predicted state of the liquid cooling device and a corresponding plurality of predicted control actions based on the feature vector; Find the corresponding state-action pair from the Q table defined by the Q-learning algorithm, and determine the action of the state-action pair with the highest Q value as the target control action; The Q-learning algorithm is defined as follows: Creating a Q-table for the Q-learning algorithm, the Q-table comprising a plurality of state-action pairs and their corresponding Q values; A reward function is set for the Q-learning algorithm, wherein the reward function is set to give a negative reward to the Q value when it is detected that the state data does not meet the preset energy consumption conditions and safety conditions; and give a positive reward to the Q value when it is detected that the state data meets the preset energy consumption conditions and safety conditions; An ε-greedy strategy is set for the Q-learning algorithm, and an initial value and a decay rate of ε are set. The ε-greedy strategy is set to randomly select a state-action pair in the Q table with probability ε, and select a state-action pair with a highest Q value in the Q table with probability 1-ε.

[0010] Optionally, the method further comprises: Based on the ε-greedy strategy, a state-action pair in the Q table is randomly selected with probability ε, and a target state-action pair with the highest Q value in the Q table is selected with probability 1-ε; According to the target state-action pair selected based on the ε-greedy strategy, a corresponding control action is generated, and the value of ε is adjusted according to the decay rate; Execute the control action corresponding to the target state-action pair on the liquid cooling device, and obtain the current state data of the electric meter and the temperature and humidity sensor while waiting for a preset holding time; Based on a preset reward function, the Q value corresponding to the target state-action pair is updated, and the number of learning times of the Q-learning is updated. When it is checked that the current state data does not meet the energy consumption condition and the safety condition, a negative reward is given to the Q value corresponding to the target state-action pair; when it is checked that the current state data meets the energy consumption condition and the safety condition, a positive reward is given to the Q value corresponding to the target state-action pair; When the number of learning times is lower than the preset number of iterations, return to step: generate a corresponding control action according to the target state-action pair selected based on the ε-greedy strategy, adjust the value of ε according to the decay rate, and select the state-action pair with the highest Q value in the Q table with probability 1-ε; until the number of learning times is not lower than the preset number of iterations.

[0011] Optionally, the recurrent neural network model is trained according to the following steps: Performing feature extraction on the historical status data of the temperature and humidity sensor and the liquid cooling device collected within a preset period, and extracting time features from the historical status data of the temperature and humidity sensor and the liquid cooling device; Based on the time characteristics, the historical status data of the temperature and humidity sensor and the liquid cooling device are divided into multiple time series data; Sorting the multiple time series data in chronological order; Dividing the sorting results of the multiple time series data into a training set and a test set according to a preset ratio; An input layer, a hidden layer, and an output layer of a recurrent neural network model are defined, wherein the output layer is defined to output a feature vector including a predicted state of the liquid cooling device and a corresponding plurality of predicted control actions; Training the recurrent neural network model using the training set; After each training round, the recurrent neural network model is tested using the test set to determine the accuracy of the recurrent neural network model; When the accuracy of the recurrent neural network model does not reach a preset threshold, the hyperparameters of the recurrent neural network model are adjusted, and the steps are performed: the recurrent neural network model is trained using the training set until the accuracy of the recurrent neural network model reaches a preset threshold.

[0012] In a second aspect, an embodiment of the present disclosure provides a liquid cooling device control system, which is applied to execute the steps of a liquid cooling device control method, wherein the system includes: a power environment monitoring host, a management platform server, a temperature and humidity sensor, and a liquid cooling device; The power environment monitoring host is used to collect status data of the temperature and humidity sensors; The management platform server is used to predict the temperature and humidity data according to the status data of the temperature and humidity sensor; based on the prediction results of the temperature and humidity data, predict the status of the liquid cooling device and the corresponding multiple control actions to obtain the predicted status of the liquid cooling device and the corresponding multiple predicted control actions; based on the predicted status of the liquid cooling device, determine the target control action that best meets the preset energy consumption conditions and safety conditions from the multiple predicted control actions; and execute the target control action on the liquid cooling device.

[0013] In a third aspect, an embodiment of the present disclosure provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program implements the steps of a liquid cooling device control method when executed by the processor.

[0014] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a liquid cooling device control method are implemented.

[0015] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects: The present disclosure can monitor the environmental changes in the computer room in real time by collecting data from temperature and humidity sensors, and can obtain the current temperature and humidity status in time, avoiding the inefficiency of relying on manual monitoring and response in traditional methods, thereby improving the response speed of environmental control. By predicting the collected temperature and humidity data, potential environmental change trends can be identified in advance. The liquid cooling equipment can actively adjust the control strategy before the environmental parameters become abnormal, avoiding the risk of overheating or insufficient cooling of the equipment due to lack of prediction ability. The present disclosure determines the target control action based on the prediction results, making the operation of the liquid cooling equipment more accurate and efficient, and can dynamically adjust the operating status of the liquid cooling equipment according to real-time data and prediction information, reducing unnecessary energy consumption and improving cooling efficiency. Therefore, the present disclosure not only improves the intelligent level of environmental monitoring, but also brings significant improvements in energy efficiency, stability and safety, and has broad application prospects and market value. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a schematic diagram of the steps of a liquid cooling device control method provided by an embodiment of the present disclosure; Figure 2 It is a schematic diagram of the tuning strategy flow of the recurrent neural network model in the present disclosure; Figure 3 It is a schematic diagram of the overall logic flow of a liquid cooling device control method provided by an embodiment of the present disclosure; Figure 4 It is a structural block diagram of a liquid cooling equipment control system provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0019] In today's era of rapid development of informatization and digitalization, as the core facility for information storage and processing, the operation efficiency and environmental control of data centers are particularly important. With the dense deployment of servers and network equipment, the problem of heat accumulation in the computer room is becoming increasingly serious, and traditional cooling methods can no longer meet the dual needs of high efficiency and energy saving and emission reduction. Therefore, liquid cooling technology has gradually become an ideal choice for data center cooling due to its superior thermal management capabilities. However, the liquid cooling equipment control method of the related technology often relies on manual intervention and lacks intelligent decision support, resulting in low cooling efficiency and energy waste. The present disclosure proposes a liquid cooling equipment control method, which aims to achieve efficient control of liquid cooling equipment through intelligent technical means. Its core concept is to use temperature and humidity sensors to collect environmental data in real time, and combine advanced machine learning algorithms, especially LSTM time series models and recurrent neural networks (RNNs) to predict and analyze temperature and humidity data. Based on the prediction results, control instructions for target control actions of liquid cooling equipment can be generated autonomously, thereby achieving precise environmental control. The starting point of the present disclosure is to improve the environmental monitoring capabilities of data centers, ensure that equipment operates in the best working state, and reduce energy consumption and operating costs. Through intelligent control methods, not only can the cooling efficiency be improved, but also the stability and security of the system can be enhanced, providing strong support for the sustainable development of data centers.

[0020] Figure 1 1 is a schematic diagram of the steps of a liquid cooling device control method provided by an embodiment of the present disclosure, the method comprising: Step S101, collecting status data of the temperature and humidity sensor; Step S102, predicting the temperature and humidity data according to the status data of the temperature and humidity sensor; Step S103, predicting the state of the liquid cooling device and the corresponding multiple control actions based on the prediction result of the temperature and humidity data, to obtain the predicted state of the liquid cooling device and the corresponding multiple predicted control actions; Step S104, based on the predicted state of the liquid cooling device, determining a target control action that best meets the preset energy consumption conditions and safety conditions from the multiple predicted control actions; Step S105: executing the target control action on the liquid cooling device.

[0021] Involving step S101, in step S101, the status data in the computer room is collected in real time through a high-precision temperature and humidity sensor to obtain key indicators related to the computer room environment. In the embodiment of the present disclosure, the status data of the temperature and humidity sensor includes the following aspects: one is temperature data. The sensor monitors the ambient temperature in the computer room in real time, in degrees Celsius (°C), reflecting the heat dissipation of the equipment in the computer room and the cooling requirements of the environment. The second is humidity data. The sensor also monitors the relative humidity in the computer room, expressed in percentage (%). The sensor monitors the temperature and humidity changes in the computer room in real time through the built-in detection element to ensure the timeliness of the data. The data can be collected at a preset time interval, such as every one minute. The status data should contain the timestamp of the corresponding collection time to facilitate subsequent feature processing. The collected status data is transmitted to the power environment monitoring host through a wired or wireless communication protocol (such as Modbus, MQTT, etc.) to form a data stream. After receiving the data, the power environment monitoring host performs preliminary storage and processing on it to prepare for subsequent predictive analysis.

[0022] Involving step S102, the purpose of this step is to take corresponding control measures in advance by predicting future environmental changes to ensure the safety and efficiency of the equipment in the computer room. The temperature and humidity in the computer room are affected by many factors, including equipment load, external climate, operating status of the cooling system, etc. The dynamic changes of these factors make it difficult to predict environmental parameters. Through the analysis and modeling of historical data, these change patterns can be better understood. If control is based solely on real-time monitoring data, it may cause a lag in response and fail to respond to sudden environmental changes in a timely manner. By predicting future temperature and humidity trends, measures can be taken in advance to prevent environmental parameters from exceeding the safety range. The results of the prediction of temperature and humidity data mainly include future temperature prediction values ​​and future humidity prediction values. Specifically, the future temperature prediction value is based on the current temperature state data and historical data, predicting the temperature change trend in the future period of time, and presenting it in the form of a time series, such as the expected temperature value in the next 1 hour, 2 hours, or even longer, and the future temperature prediction value is similar. In addition to specific values, the results of the prediction of temperature and humidity data can also include the change trend of temperature and humidity (such as rising, falling or remaining stable), which helps to determine when control measures need to be taken. In actual operation, the prediction process usually uses machine learning algorithms, such as LSTM time series models or other time series analysis methods. These models can capture the time series characteristics of temperature and humidity changes by training historical state data, thereby improving the accuracy of prediction. Through the prediction process of this step, it can provide a scientific basis for the subsequent determination of target control actions, ensure that the liquid cooling equipment operates in the best state, and thus improve the overall efficiency and reliability of the data center.

[0023] Involving step S103, using the predicted temperature and humidity data, combined with the historical operating data of the liquid cooling equipment (such as flow, temperature, energy consumption, etc.), a machine learning model (such as a recurrent neural network or other deep learning model) is used to predict the future state of the liquid cooling equipment. The predicted state includes the operating temperature of the liquid cooling equipment, the flow of the coolant, the energy consumption of the equipment, etc. Based on the predicted state of the liquid cooling equipment, a series of possible control actions are generated. The control actions include adjusting the flow and temperature of the coolant, the state of the switch cooling valve, starting or stopping the operation of the liquid cooling equipment, etc., to ensure that the liquid cooling equipment can operate effectively under the predicted environmental conditions and meet the temperature and humidity requirements of the computer room. Finally, the predicted state of the liquid cooling equipment and the corresponding multiple predicted control actions are output, wherein the predicted state of the liquid cooling equipment is used to describe the working state of the liquid cooling equipment at a specific point in the future. Multiple predicted control actions list a series of possible control actions for subsequent steps to select the optimal target control action.

[0024] Involving step S104, for each predictive control action, its impact on the state of the liquid cooling equipment is evaluated, especially the impact on energy consumption and safety. The energy consumption condition can be to check the energy consumption level that each control action may cause after execution to ensure that it is within the preset energy consumption range. For example, the control action may affect the flow and temperature of the coolant, thereby affecting the overall energy consumption. The safety condition can be to ensure that all control actions will not cause equipment overheating, coolant leakage or other safety hazards after execution. The definition of safety conditions and energy consumption conditions can be in the form of safety thresholds and energy consumption thresholds. Further, the threshold can be selected to set a specific value or a range interval. After evaluating all predictive control actions, the target control action that best meets these conditions will be selected based on the preset energy consumption conditions and safety conditions. Specifically, all control actions can be sorted according to the energy consumption and safety evaluation results, and those actions with low energy consumption and safety are given priority. If there are multiple objectives (such as minimum energy consumption and maximum safety), a multi-objective optimization algorithm can be used to balance these objectives and select the optimal target control action.

[0025] Involving step S105, according to the target control action determined in step S104, the target control action is actually executed to adjust the operating state of the liquid cooling device. This step is the link of converting predictions and decisions into actual operations, which can be to generate instructions corresponding to the target control action, including adjusting the coolant flow, switching the cooling valve, and starting and stopping the liquid cooling device. The liquid cooling device performs the actual corresponding action according to the received control instruction. While executing the target control action, it will continue to monitor the temperature and humidity status in the computer room in real time to evaluate the effect of the control action.

[0026] The present disclosure can monitor the environmental changes in the computer room in real time by collecting data from temperature and humidity sensors, and can obtain the current temperature and humidity status in time, avoiding the inefficiency of relying on manual monitoring and response in traditional methods, thereby improving the response speed of environmental control. By predicting the collected temperature and humidity data, potential environmental change trends can be identified in advance. The liquid cooling equipment can actively adjust the control strategy before the environmental parameters become abnormal, avoiding the risk of overheating or insufficient cooling of the equipment due to lack of prediction ability. The present disclosure determines the target control action based on the prediction results, making the operation of the liquid cooling equipment more accurate and efficient, and can dynamically adjust the operating status of the liquid cooling equipment according to real-time data and prediction information, reducing unnecessary energy consumption and improving cooling efficiency. Therefore, the present disclosure not only improves the intelligent level of environmental monitoring, but also brings significant improvements in energy efficiency, stability and safety, and has broad application prospects and market value.

[0027] Exemplarily, predicting the temperature and humidity data based on the status data of the temperature and humidity sensor includes: monitoring the collected status data of the temperature and humidity sensor; when it is detected that the status data of the temperature and humidity sensor exceeds a preset temperature threshold or humidity threshold, performing feature extraction on the status data of the temperature and humidity sensor, and extracting time features from the status data of the temperature and humidity sensor; inputting the status data of the temperature and humidity sensor after feature extraction into an LSTM time series model, wherein the LSTM time series model is pre-trained using historical status data of the temperature and humidity sensor collected within a preset period; and predicting the temperature and humidity data using the LSTM time series model.

[0028] When the status data from the high-precision temperature and humidity sensor is collected, the status data is monitored. The status data is compared with the preset temperature threshold and humidity threshold. For example, the temperature threshold may be set to 27°C and the humidity threshold may be set to 70%. In the case where the status data is detected to exceed the preset threshold, feature extraction will be performed. The purpose of feature extraction is to extract information useful for prediction from the original temperature and humidity data, especially time features. Time features can include the timestamp, hour, date, etc. of the collected data. Assume that the timestamp collected at 28°C and 75% is "2023-10-01 14:30". Time features will be extracted, such as "14 hours" means that it is currently 2:30 pm. After the features are extracted, the processed temperature and humidity status data is input into the LSTM (Long Short-Term Memory) time series model. The LSTM model is trained by collecting historical status data within a preset period (such as the past 30 days), and can capture the time series characteristics of temperature and humidity changes. It should be noted that the collection period of historical data and the prediction period of future data involved in the embodiment of the present disclosure can be adjusted in hours, which can be 24 hours, 30 hours, or a week (7 days), a month (30 days), and the embodiment of the present disclosure does not limit this. After the training is completed, the current status data (such as 28°C and 75%) and the extracted time features will be used as input for the LSTM model to predict. Finally, the LSTM model will predict the temperature and humidity data based on the input data. The LSTM model will output the predicted values ​​of temperature and humidity in the future, which can be presented in the form of a time series, such as the expected values ​​of temperature and humidity in the next 1 hour and 2 hours. It should be noted that when the status data is collected, before feature extraction, a series of preprocessing of the status data is required, including data cleaning, data format conversion, and data standardization or normalization, to ensure that the collected status data is of high quality and uniform format before feature extraction, and can effectively reflect the actual situation of the computer room environment. The preprocessing step lays a good foundation for subsequent feature extraction and model training.

[0029] Through the above steps, effective predictions can be made based on real-time monitored temperature and humidity data combined with historical data, which can not only perceive future environmental changes in advance, but also provide support for control decisions of liquid cooling equipment, ensuring the safety of the computer room environment and efficient operation of the equipment.

[0030] Exemplarily, the LSTM time series model is trained according to the following steps: based on time features, the historical status data of the temperature and humidity sensor is divided into multiple time series data; the multiple time series data are sorted in chronological order; the sorted results of the multiple time series data are divided into a training set and a test set according to a preset ratio; the LSTM time series model is constructed, and the LSTM time series model is trained using the training set; after each training round, the LSTM time series model is tested using the test set to determine the accuracy of the LSTM time series model; when the accuracy of the LSTM time series model does not reach a preset threshold, the hyperparameters of the LSTM time series model are adjusted, and the step of training the LSTM time series model using the training set is executed until the accuracy of the LSTM time series model reaches a preset threshold.

[0031] The historical status data of the temperature and humidity sensor is divided into multiple time series data according to the time characteristics. This can be achieved by selecting a fixed time window, such as using the data of the past 24 hours as a time series sample. Suppose there is a week of temperature and humidity data, which is recorded every hour. Every 24 hours of data can be divided into a time series sample. In this way, if the data is from "2023-10-01 00:00" to "2023-10-07 23:00", 7 samples will be obtained, each containing 24 hours of temperature and humidity data. After segmentation, make sure that these time series data are arranged in chronological order, because the LSTM model relies on the sequential information of the time series for learning. For the above 7 samples, their time order should be from "2023-10-01 00:00" to "2023-10-07 23:00", and each sample is arranged in sequence. Divide these time series data into training sets and test sets according to the preset ratio (for example, 80% training set and 20% test set). The training set is used to train the model, while the test set is used to evaluate the performance of the model. Assuming there are 7 samples, at a ratio of 80%, the first 5 samples ("2023-10-01" to "2023-10-05") will be used for the training set, and the last 2 samples ("2023-10-06" and "2023-10-07") will be used for the test set. At this stage, build the LSTM model and define its structure, including the input layer, hidden layer, and output layer. The input of the LSTM model can be a three-dimensional array in the format of (number of samples, time step, number of features). If each sample contains 24 hours of temperature and humidity data, and each sample has 2 features (temperature and humidity), the shape of the input data is (5, 24, 2) for the training set. The LSTM model is trained using the training set data, and the back propagation algorithm can be used to update the parameters. During the training process, the LSTM model will continuously adjust its weights to minimize the error between the predicted value and the true value. During the training process, the LSTM model learns the relationship between temperature and humidity, such as the change trend of humidity when the temperature rises. After each training round, the model is evaluated using the test set to calculate the accuracy or other performance indicators of the LSTM model (such as mean square error, mean absolute error, etc.). Assume that the mean square error between the temperature and humidity values ​​predicted by the model on the test set and the actual values ​​is 0.5, and the model accuracy is 85%. If the accuracy of the LSTM model does not reach the preset threshold (for example, 90%), the hyperparameters of the LSTM model need to be adjusted, and the LSTM model can also be adjusted by means of K-fold cross validation. Hyperparameters include learning rate, batch size, number of LSTM layers, number of hidden units, etc. If the current learning rate is 0.001, you may try to adjust it to 0.0005, or increase the number of LSTM layers from 1 to 2 layers.After adjusting the hyperparameters, the training set is used to train the LSTM model again and tested again until the accuracy of the LSTM model reaches the preset threshold.

[0032] Through the above steps, the LSTM time series model can effectively learn the changing patterns of temperature and humidity from historical data and make accurate predictions based on this. This ensures that the LSTM model can function in a real environment, thereby achieving intelligent control of the computer room environment.

[0033] Exemplarily, the state of the liquid cooling device and the corresponding multiple control actions are predicted based on the prediction results of the temperature and humidity data to obtain the predicted state of the liquid cooling device and the corresponding multiple predicted control actions, including: inputting the prediction results of the temperature and humidity data into a recurrent neural network model, the recurrent neural network model is pre-trained using the historical state data of the temperature and humidity sensor and the liquid cooling device collected within a preset period; predicting the state of the liquid cooling device and the corresponding multiple control actions through the recurrent neural network model; obtaining a feature vector output by the recurrent neural network, the feature vector including the predicted state of the liquid cooling device and the corresponding multiple predicted control actions.

[0034] The predicted results of temperature and humidity data are input into a recurrent neural network (RNN) model pre-trained with the historical status data of the temperature and humidity sensor and the liquid cooling device. The historical status data of the liquid cooling device refers to the numerical data and status data of the liquid cooling device, which may include information such as the set value of the CDU or the switch status of the cooling valve, and is collected in real time based on the status data feedback from the liquid cooling device (such as the liquid cooling CDU). The recurrent neural network model predicts the current status and possible control actions of the liquid cooling device based on the input temperature and humidity prediction results and the previous historical status data. For example, the recurrent neural network model may predict that the status of the liquid cooling device is "normal operation" based on the input temperature and humidity prediction results (28°C, 45%RH), and generate multiple corresponding predictive control actions, action A, set the temperature of the liquid cooling device to 26°C; action B, set the temperature of the liquid cooling device to 27°C; action C, maintain the current temperature setting. The output of the recurrent neural network model is a feature vector containing the current temperature and humidity, the predicted status of the liquid cooling device, and the corresponding multiple predictive control actions, which are used in the subsequent decision-making process. In the above example, the feature vector output by the recurrent neural network model can be [28, 45, 1, 26, 27, 28], where 28 is the current temperature (temperature prediction result), 45 is the current humidity (humidity prediction result), 1 is the predicted state of the liquid cooling equipment (normal operation), 26 is the temperature setting for action A, 27 is the temperature setting for action B, and 28 is the temperature setting for action C (keep the current setting).

[0035] The above steps can effectively predict the current state of the liquid cooling equipment and its possible control actions by inputting the predicted results of temperature and humidity data and the historical state data of the liquid cooling equipment into the pre-trained recurrent neural network (RNN) model. The contextual information of the historical data is utilized to enable the recurrent neural network model to understand and respond to environmental changes more accurately. The output feature vector integrates the current environmental parameters, equipment status and control options, providing a comprehensive information basis for the subsequent decision-making process. Through this intelligent prediction and control, the efficient operation of the liquid cooling equipment can be achieved, the stability and safety of the computer room environment can be improved, thereby optimizing energy use and reducing operating costs.

[0036] Exemplarily, based on the predicted state of the liquid cooling device, determining the target control action that best meets the preset energy consumption conditions and safety conditions from the multiple predicted control actions includes: applying the Q-learning algorithm to the feature vector output by the recurrent neural network, and determining the predicted state of the liquid cooling device and the corresponding multiple predicted control actions based on the feature vector; searching for the corresponding state-action pair from the Q table defined by the Q-learning algorithm, and determining the action of the state-action pair with the highest Q value as the target control action.

[0037] The feature vector generated by the recurrent neural network (RNN) based on the input data (such as temperature and humidity, CDU status, cold valve status, etc.) contains the current status information of the liquid cooling equipment, that is, the predicted status of the liquid cooling equipment. The feature vector is a highly abstract representation of the environmental status and can effectively represent the operation status of the liquid cooling system. Figure 2 is a schematic diagram of the tuning strategy flow of the recurrent neural network model in the present disclosure, such as Figure 2 As shown, based on the feature vector, the Q-learning algorithm can infer the future state of the liquid cooling equipment. For example, suppose the feature vector indicates that the current temperature is 30°C, the humidity is 40%RH, the CDU state is normal, and the cold valve is closed. Based on the information of the feature vector, the Q-learning algorithm can predict the possible future state of the liquid cooling equipment under the current state (such as the temperature may rise to 32°C, and the humidity may rise to 45%RH). Based on the current state, the Q-learning algorithm generates multiple possible predictive control actions. For example, the following four predictive control actions, action A, lower the CDU temperature setting; action B, open the cold valve; action C, increase the coolant flow; action D, keep the current state unchanged. As mentioned earlier, the Q table stores the Q value of each state-action pair, which represents the expected benefit of taking an action in a specific state. Assume that the state-action pair corresponding to the current state in the Q table is as follows: (state 1, action A) ->Q value = 5; (state 1, action B) ->Q value = 10; (state 1, action C) ->Q value = 7; (State 1, Action D) -> Q value = 3; Search all actions in the current state according to the Q table and select the action with the highest Q value as the target control action. In this example, the Q value of action B (opening the cold valve) is 10, which is the highest among all actions, so it is determined as the target control action.

[0038] Exemplarily, the Q-learning algorithm is defined according to the following steps: creating a Q-table for the Q-learning algorithm, the Q-table including multiple state-action pairs and their corresponding Q values; setting a reward function for the Q-learning algorithm, the reward function being set to give a negative reward to the Q value when it is detected that the state data does not meet the preset energy consumption conditions and safety conditions; giving a positive reward to the Q value when it is detected that the state data meets the preset energy consumption conditions and safety conditions; setting an ε-greedy strategy for the Q-learning algorithm, and setting an initial value and decay rate of ε, the ε-greedy strategy being set to randomly select a state-action pair in the Q table with a probability of ε, and select the state-action pair with the highest Q value in the Q table with a probability of 1-ε.

[0039] The key components of the Q-learning algorithm mainly include the creation of the Q-table, the setting of the reward function, and the implementation of the ε-greedy strategy. The following will explain these concepts and their role in Q-learning in detail one by one. The Q-table is the core data structure of the Q-learning algorithm, which is used to store the Q value of each state-action pair. The Q value represents the expected benefit of taking an action in a specific state. The rows of the Q-table represent different states, and the columns represent the actions that can be taken. The Q value of each state-action pair reflects the cumulative reward that may be obtained in the future after taking the action in that state. By continuously updating the Q value, the algorithm can learn the optimal action selection in different states. The reward function is a feedback mechanism used in the Q-learning algorithm to evaluate each state-action pair. It gives corresponding rewards or penalties based on the performance of the current state. Specifically, when the state data does not meet the preset energy consumption conditions or safety conditions, a negative reward is given. It means that the action taken has led to adverse consequences, and the algorithm will tend to reduce the selection of such actions. On the contrary, when the state data meets the preset energy consumption conditions and safety conditions, a positive reward is given. This encourages the algorithm to choose such actions in the future to achieve better performance and safety. The ε-greedy strategy is a commonly used exploration and exploitation balance strategy in Q-learning, which encourages the exploration of new state-action pairs by introducing a randomness while also exploiting the known best choice. By combining the feature vector output by the recurrent neural network model and the Q-learning algorithm, the best control action can be intelligently selected to optimize the operating state of the liquid cooling equipment. The present disclosure provides a decision-making method for realizing automatic control using machine learning technology to improve the energy efficiency and safety of the data center.

[0040] Exemplarily, the method also includes: based on the ε greedy strategy, randomly selecting the state-action pair in the Q table with probability ε, and selecting the target state-action pair with the highest Q value in the Q table with probability 1-ε; generating a corresponding control action according to the target state-action pair selected based on the ε greedy strategy, and adjusting the value of ε according to the decay rate; executing the control action corresponding to the target state-action pair on the liquid cooling device, and obtaining the current state data of the electric meter and the temperature and humidity sensor while waiting for a preset holding time; updating the Q value corresponding to the target state-action pair based on a preset reward function, and updating the learning function of the Q-learning. times, and when it is checked that the current state data does not meet the energy consumption condition and the safety condition, a negative reward is given to the Q value corresponding to the target state-action pair; when it is checked that the current state data meets the energy consumption condition and the safety condition, a positive reward is given to the Q value corresponding to the target state-action pair; when the number of learning times is lower than the preset number of iterations, return to step: according to the target state-action pair selected based on the ε greedy strategy, a corresponding control action is generated, and the value of ε is adjusted according to the decay rate, and the state-action pair with the highest Q value in the Q table is selected with probability 1-ε; until the number of learning times is not lower than the preset number of iterations.

[0041] When selecting an action, a state-action pair is randomly selected with probability ε (exploration), and the state-action pair with the highest Q value is selected with probability 1-ε (exploitation). Assume ε = 0.1, which means there is a 10% probability of random selection and a 90% probability of selecting the action with the highest Q value in the Q table. If the Q table of the current state is as follows: (state 1, action A) ->Q value = 5; (state 1, action B) ->Q value = 10; (State 1, Action C) ->Q value = 7; then randomly select action A, B or C with a probability of 10%, and select action B with a probability of 90% (Q value = 10). Generate the corresponding control action based on the selected target state-action pair. For example, assuming that action B is selected (set the temperature of the liquid cooling device to 27°C), the generated control action is to set the temperature of the liquid cooling device to 27°C. Adjust the value of ε according to the decay rate to reduce the frequency of exploration. Assuming the decay rate is 0.01, the initial ε is 0.1, and the adjusted new ε is 0.1 - 0.01 = 0.09. Please refer to Figure 2 After executing the generated control action, wait for the preset holding time (for example, 1 minute), and then obtain the current state data from the electric meter and temperature and humidity sensor. The current state data of the electric meter and temperature and humidity sensor can also be collected in real time and stored in advance, and then obtained and further judged while waiting for the preset holding time. For example, the current temperature is 26°C, the current humidity is 50%RH, and the current energy consumption is 200W. Based on the preset reward function, the Q value of the target state-action pair is updated, and positive or negative rewards are given according to whether the current state data meets the energy consumption conditions and safety conditions ( Figure 2 Take whether the temperature and humidity exceed the threshold as an example to define the safety condition). As mentioned above, the energy consumption condition and the safety condition can be a value or a range. Assume that the energy consumption condition is expected to be ≤250W, and the safety condition is expected to be between 20°C and 30°C. In this example, the current state meets the energy consumption condition (200W ≤ 250W) and the safety condition (26°C is between 20°C and 30°C), so a positive reward (for example, +5) is given. Assuming that the current Q value is 10, the updated Q value can be calculated by the following formula: New Q value = old Q value + learning rate * (reward + future Q value - old Q value); assuming the learning rate is 0.1 and the future Q value is 10 (assuming no change), then the new Q value = 10 + 0.1 * (5 + 10 - 10) = 10 + 0.5 = 10.5. Check whether the current number of learning times is lower than the preset number of iterations (for example, 100 times). If so, return to step: Generate the corresponding control action based on the target state-action pair selected based on the ε-greedy strategy, and adjust the value of ε according to the decay rate, select the state-action pair with the highest Q value in the Q table with probability 1-ε, continue to select the target state-action pair, generate control actions and adjust the value of ε. The above process will continue until the number of learning times reaches the preset number of iterations.

[0042] Through the above steps, the control strategy of the liquid cooling equipment can be continuously optimized through the ε greedy strategy and Q-learning algorithm. After each control action is executed, the Q value is updated according to the current state data, thereby improving the accuracy and efficiency of future decisions, and ultimately achieving the goals of energy saving and safety.

[0043] Exemplarily, the recurrent neural network model is trained according to the following steps: feature extraction is performed on the historical status data of the temperature and humidity sensor and the liquid cooling device collected within a preset period, and time features are extracted from the historical status data of the temperature and humidity sensor and the liquid cooling device; based on the time features, the historical status data of the temperature and humidity sensor and the liquid cooling device are divided into multiple time series data; the multiple time series data are sorted in chronological order; the sorted results of the multiple time series data are divided into a training set and a test set according to a preset ratio; the input layer, hidden layer and output layer of the recurrent neural network model are defined, and the output layer is defined as outputting a feature vector including the predicted state of the liquid cooling device and the corresponding multiple predicted control actions; the recurrent neural network model is trained using the training set; after each training round, the recurrent neural network model is tested using the test set to determine the accuracy of the recurrent neural network model; when the accuracy of the recurrent neural network model does not reach the preset threshold, the hyperparameters of the recurrent neural network model are adjusted, and the steps are performed: the recurrent neural network model is trained using the training set until the accuracy of the recurrent neural network model reaches the preset threshold.

[0044] Similar to the LSTM time series model training process, features are extracted from the historical status data of the temperature and humidity sensor and the liquid cooling device. Based on the extracted time features, the historical status data is divided into multiple time series data. The multiple time series data are sorted in chronological order. The sorted time series data is divided into a training set and a test set according to a preset ratio. Define the input layer, hidden layer, and output layer of the recurrent neural network model. For example, the input layer is defined as inputting the temperature and humidity data of the past 24 hours (24 temperature and humidity values). The hidden layer is defined as 2 LSTM layers, each with 64 units. The output layer is defined as outputting a feature vector containing the predicted state of the liquid cooling device and the corresponding multiple predicted control actions. Use the training set to train the recurrent neural network model. Use the fit method of the Keras library for training. The batch size can be set to 32 and the training rounds to 50. After each training round, use the test set to test the model and determine the accuracy. Use the evaluate method to calculate the accuracy of the test set. If the accuracy of the recurrent neural network model does not reach the preset threshold, adjust the hyperparameters. Assuming the preset threshold is 90%, if the current accuracy is 85%, you can consider adjusting the learning rate (such as from 0.001 to 0.0005), increasing the number of hidden layer units (such as from 64 to 128), and increasing the number of training rounds (such as from 50 to 100). After adjusting the hyperparameters, re-use the training set to train the recurrent neural network model. Continue to test the accuracy of the recurrent neural network model and adjust the hyperparameters until the accuracy of the recurrent neural network model reaches the preset threshold.

[0045] Through the above steps, the recurrent neural network model can be effectively trained to accurately predict the state and control actions of the liquid cooling equipment. The process shows how to optimize model performance and ultimately achieve intelligent control through steps such as feature extraction, time series processing, model definition, training, and evaluation.

[0046] Figure 3 is a schematic diagram of the overall logic flow of a liquid cooling device control method provided by an embodiment of the present disclosure, such as Figure 3As shown in the figure, the temperature and humidity sensor monitors the temperature and humidity in the computer room in real time, and sends the collected status data to the power environment monitoring host. After collection, the status data of the temperature and humidity sensor is first monitored. If the temperature or humidity is detected to exceed the preset threshold, feature extraction is performed to extract the time feature. The extracted status data is input into the pre-trained LSTM time series model to predict the temperature and humidity data. According to the prediction results, the prediction results of the temperature and humidity data are input into the recurrent neural network model to predict the status and control actions of the liquid cooling equipment. The feature vector output by the recurrent neural network is obtained. The target control action is determined from multiple predictive control actions using the Q-learning algorithm. The predicted data is analyzed by the intelligent algorithm. According to the target control action determined in the decision control stage, a control instruction is sent to the liquid cooling equipment. According to the received instruction, the liquid cooling equipment adjusts the flow rate, temperature and other parameters of the coolant to keep the temperature and humidity in the computer room within the set range. After the equipment is controlled, wait for a preset holding time to obtain the current status data of the electric meter and the temperature and humidity sensor. Based on the preset reward function, the Q value of the target state-action pair is updated. If the current state data meets or does not meet the energy consumption and safety conditions, a positive or negative reward is given respectively. The probability of selecting an action is adjusted according to the ε-greedy strategy, and the decision control and device control steps are repeated until the preset number of learning times is reached. By continuously updating the strategy, the control effect is gradually optimized, and the accuracy and efficiency of intelligent control are improved.

[0047] Exemplarily, the method also includes: based on the historical status data of the liquid cooling device, optimizing the control strategy of the liquid cooling device using a generative adversarial network, the generative adversarial network including a generator and a discriminator, wherein the generator is used to generate potential control strategies for the liquid cooling device, and the discriminator is used to discriminate the difference between the generated control strategy and the actual control strategy; optimizing the control strategy generated by the generator through the discriminator to improve the energy efficiency and stability of the control strategy; and controlling the liquid cooling device based on the control strategy generated by the generator.

[0048] Generative Adversarial Network (GAN) is a deep learning framework that consists of two adversarial neural networks, a generator and a discriminator. They compete with each other to gradually optimize the output quality of the generator. The historical state data of the liquid cooling equipment includes the operation data of the CDU and the cooling valve in different time periods. The task of the generator is to generate potential liquid cooling equipment control strategies from the input data (historical state data). Including operating parameters, such as adjusting the coolant flow rate, fan speed, etc. The generator may generate control strategies that are far different from the actual application in the initial stage, but through adversarial training with the discriminator, the generator will gradually improve its strategy and make it closer and closer to the optimal state. The task of the discriminator is to evaluate the control strategy generated by the generator and compare it with the actual strategy used. The discriminator tries to distinguish the quality of the generated strategy from the actual strategy. The discriminator is used to feedback to the generator which generated strategies are undesirable and which strategies are closer to the actual optimal strategy. In adversarial training, the generator continuously generates new control strategies, while the discriminator tries to distinguish the difference between these strategies and the actual optimal control strategy. The generator continuously optimizes its output through feedback from the discriminator. Whenever the discriminator can easily distinguish between the generated strategy and the real strategy, the generator will adjust its parameters to generate a better control strategy. The discriminator also becomes more and more accurate during the training process. It will become more and more adept at identifying the defects of the generated control strategy, thereby providing more valuable feedback to the generator. This training process is an iterative process of gradual improvement. As the training progresses, the control strategy generated by the generator becomes closer and closer to the actual optimal control strategy, and finally achieves the effect of being close to or better than the control strategy designed by humans.

[0049] After multiple optimizations of the discriminator feedback, the generator is finally able to generate a control scheme that is very close to the real control strategy. The generated control strategy can not only operate under various temperature and humidity conditions, but also consider energy efficiency optimization issues, that is, how to minimize energy consumption while ensuring device performance. When the generative adversarial network has been trained sufficiently and the generator is able to generate high-quality control strategies, these strategies can be used to actually control liquid cooling equipment. The strategies generated by the generator will dynamically adapt to the operating environment of the equipment and adjust operations based on the real-time status data of the equipment. For example, during the operation of the liquid cooling equipment, if the temperature is too high, the generator may generate a strategy to increase the coolant flow rate or increase the fan speed; if the temperature is too low, it will generate a strategy to reduce these parameters to maintain the balance of the system.

[0050] By introducing a generative adversarial network (GAN), the control strategy of liquid cooling equipment can be automatically generated and optimized to achieve the goal of improving energy efficiency and stability. Through continuous adversarial training between the generator and the discriminator, the generated control strategy gradually approaches the optimal one, and in some cases can even surpass the traditional manually designed strategy. Ultimately, these generated strategies can be applied to the actual operation of liquid cooling equipment, enabling liquid cooling equipment to respond to complex environmental changes more intelligently and stably, reduce energy consumption and maintain efficient operation.

[0051] Figure 4 is a structural block diagram of a liquid cooling device control system provided by an embodiment of the present disclosure, such as Figure 4 As shown, the steps of executing a liquid cooling equipment control method are applied, and the system includes: a power environment monitoring host, a management platform server, a temperature and humidity sensor, and a liquid cooling equipment; the power environment monitoring host is used to collect status data of the temperature and humidity sensor; the management platform server is used to predict the temperature and humidity data according to the status data of the temperature and humidity sensor; based on the prediction results of the temperature and humidity data, the state of the liquid cooling equipment and the corresponding multiple control actions are predicted to obtain the predicted state of the liquid cooling equipment and the corresponding multiple predicted control actions; based on the predicted state of the liquid cooling equipment, the target control action that best meets the preset energy consumption conditions and safety conditions is determined from the multiple predicted control actions; and the target control action is executed on the liquid cooling equipment.

[0052] The power environment monitoring host is responsible for collecting status data from temperature and humidity sensors in real time, including environmental parameters such as temperature and humidity in the computer room. The power environment monitoring host performs preliminary processing and storage on these data, providing a basis for subsequent data analysis and prediction. The power environment monitoring host is usually equipped with a high-performance processor and storage unit to support rapid data collection and processing. Optionally, a data communication module can be integrated to send the collected data to the management platform server.

[0053] The management platform server receives data from the power environment monitoring host and analyzes and predicts the temperature and humidity data. Based on the prediction results, the state of the liquid cooling device and the corresponding multiple control actions are further predicted to obtain the predicted state of the liquid cooling device and the corresponding multiple predicted control actions, and finally determine the target control action for the liquid cooling device (for example, adjusting the temperature setting and flow rate of the liquid cooling device). The management platform server uses advanced data analysis algorithms, such as machine learning and deep learning algorithms, to improve the accuracy of the prediction. Optionally, a user interface is provided to facilitate operators to monitor the system status and manually adjust the control strategy.

[0054] The temperature and humidity sensor monitors the temperature and humidity in the room in real time and transmits the data to the power environment monitoring host. It has the characteristics of high precision and high sensitivity and can work stably under different environmental conditions. It has a variety of communication interfaces (such as RS485, Wi-Fi, etc.) to facilitate the connection with the monitoring host.

[0055] Liquid cooling equipment performs corresponding cooling operations according to the target control actions issued by the management platform server, such as adjusting the flow and temperature of the coolant to keep the temperature and humidity in the computer room within the set range. It includes components such as cooling units (CDUs) and cooling valves, which can efficiently absorb and transfer heat. It has automatic control capabilities and can automatically adjust the operating status according to the control instructions generated by the target control actions.

[0056] In the disclosed embodiment, the temperature and humidity sensors collect data regularly and send it to the power environment monitoring host. The power environment monitoring host processes the data and sends it to the management platform server. The management platform server analyzes the data and predicts future temperature and humidity changes, and generates target control actions. Finally, the target control action is sent to the liquid cooling device, and the liquid cooling device performs the corresponding cooling operation according to the instruction. Through real-time monitoring and data prediction, the liquid cooling device control system can quickly respond to changes in the computer room environment and ensure that the liquid cooling equipment is always in the best working condition. Through intelligent control algorithms (such as Q-learning, etc.), the liquid cooling equipment control system can continuously optimize the control strategy and improve energy efficiency and stability.

[0057] The liquid cooling equipment control system provided by the present disclosure realizes real-time monitoring and intelligent control of the computer room environment through the collaborative work of multiple modules. Each module coordinates and works together to ensure the efficient operation of the liquid cooling equipment, reduce energy consumption, and improve the overall performance and reliability of the data center.

[0058] An embodiment of the present disclosure also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, the various processes in the above-mentioned embodiment of a liquid cooling device control method are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0059] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process in the above-mentioned embodiment of a liquid cooling device control method is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the embodiments can be referred to each other.

[0060] It should be understood by those skilled in the art that the embodiments of the present disclosure may be provided as methods, devices, and electronic devices. Therefore, the embodiments of the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present disclosure may take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] The embodiments of the present disclosure are described with reference to the flowcharts and / or block diagrams of the methods, apparatuses, and electronic devices according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing terminal device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded into a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable terminal device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0062] Although the preferred embodiments of the present disclosure have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present disclosure.

[0063] Finally, it should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "include..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements. The above is a detailed introduction to a liquid cooling device control method, system, electronic device and storage medium provided by the present disclosure. The principles and implementation methods of the present disclosure are explained in this article using specific examples. The description of the above embodiments is only used to help understand the method and its core idea of ​​the present disclosure; at the same time, for those of ordinary skill in the art, according to the idea of ​​the present disclosure, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present disclosure.

Claims

1. A liquid cooling device control method, characterized in that: The method comprises: Collect status data of temperature and humidity sensors; Predicting temperature and humidity data according to the status data of the temperature and humidity sensor; Based on the prediction result of the temperature and humidity data, predict the state of the liquid cooling device and the corresponding multiple control actions to obtain the predicted state of the liquid cooling device and the corresponding multiple predicted control actions; Based on the predicted state of the liquid cooling device, determining a target control action that best meets the preset energy consumption conditions and safety conditions from the multiple predicted control actions; The target control action is performed on the liquid cooling device.

2. The method according to claim 1, characterized in that The predicting of temperature and humidity data according to the state data of the temperature and humidity sensor includes: Monitoring the collected status data of the temperature and humidity sensor; When it is detected that the state data of the temperature and humidity sensor exceeds a preset temperature threshold or humidity threshold, feature extraction is performed on the state data of the temperature and humidity sensor, and a time feature is extracted from the state data of the temperature and humidity sensor; Inputting the state data of the temperature and humidity sensor after feature extraction into an LSTM time series model, wherein the LSTM time series model is pre-trained using the historical state data of the temperature and humidity sensor collected within a preset period; The temperature and humidity data are predicted by the LSTM time series model.

3. The method according to claim 2, characterized in that Follow the steps below to train the LSTM time series model: Based on the time feature, the historical state data of the temperature and humidity sensor is divided into multiple time series data; Sorting the multiple time series data in chronological order; Dividing the sorting results of the multiple time series data into a training set and a test set according to a preset ratio; Constructing the LSTM time series model, and training the LSTM time series model using the training set; After each training round, the LSTM time series model is tested using the test set to determine the accuracy of the LSTM time series model; When the accuracy of the LSTM time series model does not reach a preset threshold, the hyperparameters of the LSTM time series model are adjusted, and the steps are performed: the LSTM time series model is trained using the training set until the accuracy of the LSTM time series model reaches a preset threshold.

4. The method according to claim 1, characterized in that The state of the liquid cooling device and the corresponding multiple control actions are predicted based on the prediction result of the temperature and humidity data to obtain the predicted state of the liquid cooling device and the corresponding multiple predicted control actions, including: Inputting the prediction result of the temperature and humidity data into a recurrent neural network model, wherein the recurrent neural network model is pre-trained using the historical status data of the temperature and humidity sensor and the liquid cooling device collected within a preset period; Predicting the state of the liquid cooling device and corresponding multiple control actions through the recurrent neural network model; A feature vector output by the recurrent neural network is obtained, wherein the feature vector includes a predicted state of the liquid cooling device and a corresponding plurality of predicted control actions.

5. The method according to claim 4, characterized in that The step of determining the target control action that best meets the preset energy consumption condition and safety condition from the multiple predicted control actions based on the predicted state of the liquid cooling device includes: Applying a Q-learning algorithm to a feature vector output by the recurrent neural network, and determining a predicted state of the liquid cooling device and a corresponding plurality of predicted control actions based on the feature vector; Find the corresponding state-action pair from the Q table defined by the Q-learning algorithm, and determine the action of the state-action pair with the highest Q value as the target control action; The Q-learning algorithm is defined as follows: Creating a Q-table for the Q-learning algorithm, the Q-table comprising a plurality of state-action pairs and their corresponding Q values; A reward function is set for the Q-learning algorithm, wherein the reward function is set to give a negative reward to the Q value when it is detected that the state data does not meet the preset energy consumption conditions and safety conditions; and give a positive reward to the Q value when it is detected that the state data meets the preset energy consumption conditions and safety conditions; An ε-greedy strategy is set for the Q-learning algorithm, and an initial value and a decay rate of ε are set. The ε-greedy strategy is set to randomly select a state-action pair in the Q table with probability ε, and select a state-action pair with a highest Q value in the Q table with probability 1-ε.

6. The method according to claim 5, characterized in that The method further comprises: Based on the ε-greedy strategy, a state-action pair in the Q table is randomly selected with probability ε, and a target state-action pair with the highest Q value in the Q table is selected with probability 1-ε; According to the target state-action pair selected based on the ε-greedy strategy, a corresponding control action is generated, and the value of ε is adjusted according to the decay rate; Execute the control action corresponding to the target state-action pair on the liquid cooling device, and obtain the current state data of the electric meter and the temperature and humidity sensor while waiting for a preset holding time; Based on a preset reward function, the Q value corresponding to the target state-action pair is updated, and the number of learning times of the Q-learning is updated. When it is checked that the current state data does not meet the energy consumption condition and the safety condition, a negative reward is given to the Q value corresponding to the target state-action pair; when it is checked that the current state data meets the energy consumption condition and the safety condition, a positive reward is given to the Q value corresponding to the target state-action pair; When the number of learning times is lower than the preset number of iterations, return to step: generate a corresponding control action according to the target state-action pair selected based on the ε-greedy strategy, adjust the value of ε according to the decay rate, and select the state-action pair with the highest Q value in the Q table with probability 1-ε; until the number of learning times is not lower than the preset number of iterations.

7. The method according to claim 4, characterized in that Follow the steps below to train the recurrent neural network model: Performing feature extraction on the historical status data of the temperature and humidity sensor and the liquid cooling device collected within a preset period, and extracting time features from the historical status data of the temperature and humidity sensor and the liquid cooling device; Based on the time characteristics, the historical status data of the temperature and humidity sensor and the liquid cooling device are divided into multiple time series data; Sorting the multiple time series data in chronological order; Dividing the sorting results of the multiple time series data into a training set and a test set according to a preset ratio; An input layer, a hidden layer, and an output layer of a recurrent neural network model are defined, wherein the output layer is defined to output a feature vector including a predicted state of the liquid cooling device and a corresponding plurality of predicted control actions; Training the recurrent neural network model using the training set; After each training round, the recurrent neural network model is tested using the test set to determine the accuracy of the recurrent neural network model; When the accuracy of the recurrent neural network model does not reach a preset threshold, the hyperparameters of the recurrent neural network model are adjusted, and the steps are performed: the recurrent neural network model is trained using the training set until the accuracy of the recurrent neural network model reaches a preset threshold.

8. A liquid cooling equipment control system, characterized in that: Applied to executing the steps of the method according to any one of claims 1 to 7, the system comprises: a power environment monitoring host, a management platform server, a temperature and humidity sensor, and a liquid cooling device; The power environment monitoring host is used to collect status data of the temperature and humidity sensors; The management platform server is used to predict the temperature and humidity data according to the status data of the temperature and humidity sensor; Based on the prediction results of the temperature and humidity data, the state of the liquid cooling device and the corresponding multiple control actions are predicted to obtain the predicted state of the liquid cooling device and the corresponding multiple predicted control actions; based on the predicted state of the liquid cooling device, the target control action that best meets the preset energy consumption conditions and safety conditions is determined from the multiple predicted control actions; and the target control action is executed on the liquid cooling device.

9. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein when the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images