Intelligent chilled water storage system optimization control method based on multivariable prediction and reinforcement learning

Abstract

The invention discloses an intelligent chilled water storage system optimization control method based on multivariable prediction and reinforcement learning, and the method comprises the steps: enabling the system to collect the environment temperature, humidity, load demands and state data of a cold storage pool in real time through a sensor, generating a multivariable prediction model through a machine learning algorithm, and predicting the load demands in real time; based on the prediction result and the state of the cold storage pool, the system dynamically adjusts a cold capacity distribution strategy through an optimization algorithm, and a model prediction control technology is used for calculating a cold capacity release scheme in advance; the reinforcement learning algorithm adaptively optimizes a control strategy according to historical data and real-time feedback; meanwhile, a built-in intelligent fault diagnosis mechanism monitors the equipment state in real time, a standby control strategy is automatically triggered when abnormity is found, stable switching is ensured when a fault occurs, and the stability and robustness of the system are improved; according to the intelligent chilled water storage system optimization control method based on multivariable prediction and reinforcement learning, through an intelligent algorithm and accurate model prediction, cooling capacity distribution is optimized, dynamic adjustment is conducted, the energy utilization efficiency is improved, waste is reduced, and the cooling effect is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of adaptive control of water storage systems, particularly to an intelligent water storage system optimization control method based on multivariate prediction and reinforcement learning, suitable for the field of energy optimization management in building environments through water storage technology. BACKGROUND

[0002] The intelligent water storage system optimization control method based on multivariate prediction and reinforcement learning refers to dynamically adjusting the operating parameters of the water storage system through intelligent control algorithms to optimize system performance and maximize energy efficiency. Water storage systems are commonly used in building air conditioning systems or industrial refrigeration, storing cold energy during off-peak hours such as night, and then using the stored water for cooling during peak hours, thereby balancing energy consumption. Real-time monitoring of key parameters such as water storage temperature, cooling load, external environment temperature, and equipment operating status is achieved through sensors. Based on these data, the system can understand the current operating conditions in real time. The system establishes a mathematical model, combines historical data and real-time monitoring data, and predicts future load demand and environmental conditions. For example, predicting future air conditioning load, weather changes, and building temperature changes to help the system make optimal decisions. Based on the model prediction results, the adaptive control algorithm dynamically adjusts the control strategy of the water storage system, such as starting the storage process in advance when the predicted load demand is high, or increasing the storage of cold water during low demand periods. At the same time, the system can also adjust the cooling efficiency according to the actual temperature of the stored water to reduce energy consumption. Intelligent optimization algorithms are used to optimize control parameters. These algorithms adjust control parameters to achieve the best state of system operation, ensuring stable indoor temperature and reducing energy consumption. The adaptive control method also includes fault detection and fault tolerance mechanisms. When the system encounters an anomaly or failure, the control method can automatically adjust the system operation to ensure stable cooling effect while avoiding energy waste.

[0003] However, the existing intelligent water storage system optimization control method based on multivariate prediction and reinforcement learning, while having significant advantages in improving system efficiency and reducing energy consumption, still has some defects and challenges. Existing water storage systems typically rely on pre-set control strategies, failing to fully consider complex load fluctuations and environmental changes, resulting in insufficient system energy efficiency, serious energy waste, and problems such as response lag and insufficient control precision. SUMMARY

[0004] In view of the deficiencies in the prior art, the purpose of the present application is to provide an intelligent water cold storage system optimization control method based on multivariate prediction and reinforcement learning, which optimizes the distribution of cold energy through intelligent algorithms and accurate model prediction, and dynamically adjusts during system operation to improve the energy utilization efficiency of the system, reduce energy waste and improve cooling effect.

[0005] The technical solution adopted by the present application to solve its technical problems is:

[0006] The intelligent water cold storage system optimization control method based on multivariate prediction and reinforcement learning comprises the following steps:

[0007] Real-time collection of environmental temperature, humidity, load demand and cold storage pool state data through sensors;

[0008] Using a machine learning algorithm, combining the collected environmental data, generating a multivariate prediction model for real-time prediction of load demand, the prediction model is trained based on environmental temperature, humidity, time period, historical load factors;

[0009] According to the predicted load demand and the current state of the water cold storage pool, the distribution strategy of cold energy is dynamically adjusted through an optimization algorithm;

[0010] In the process of cold energy distribution, model predictive control technology is used to calculate the cold energy release scheme in the future period in advance according to real-time data and load prediction results;

[0011] During system operation, the control strategy is adaptively adjusted through a reinforcement learning algorithm, which continuously optimizes the control strategy through historical data and real-time feedback;

[0012] The system has an intelligent fault diagnosis mechanism that monitors the working state of the equipment in real time, automatically triggers the backup control strategy for adjustment when an abnormality is identified, and ensures that the system switches when a device fails through data integration and analysis of multiple sensors.

[0013] As a preferred method of real-time collection of environmental temperature, humidity, load demand and cold storage pool state data by sensors:

[0014] Temperature sensors and humidity sensors are deployed in the target environment to collect air temperature and humidity data in real time;

[0015] The cooling demand of the current system is monitored through load sensors, and the cooling load demand is indirectly calculated through temperature difference, pressure change and flow rate;

[0016] The temperature sensor in the cold storage pool is used to monitor the cold energy storage state of the cold storage pool, and a pressure sensor is deployed to detect the pressure change in the cold storage pool to calculate the current cooling capacity of the cold storage pool;

[0017] All sensor data is transmitted to the central control system through the real-time data acquisition unit, and the data acquisition unit adopts data preprocessing functions;

[0018] The central control system performs dynamic analysis based on the real-time collected data, and real-time monitors the environmental temperature, humidity, load demand and cold storage pool state.

[0019] As preferred, a machine learning algorithm is used to generate a multivariate prediction model based on the collected environmental data for real-time prediction of load demand. The prediction model is trained based on environmental temperature, humidity, time period, and historical load factors using the following method:

[0020] The real-time collected temperature, humidity, time period, and historical load environmental data are processed for missing value processing, outlier detection and removal. If data is missing, it is completed by interpolation and mean filling methods.

[0021] A multivariate time series prediction machine learning algorithm is used, which includes linear regression, decision tree regression, random forest regression, support vector regression, and neural network.

[0022] The feature selection method is used to select features that have a greater impact on load prediction, and to remove redundant and irrelevant features. The method used is:

[0023] The correlation between temperature, humidity, time period, and historical load features and load demand is calculated, and features with high correlation to load demand are used.

[0024] If there is multicollinearity between features, PCA is used to reduce the feature dimension and extract principal components.

[0025] The data set is divided into training set, validation set and test set. The selected machine learning model is trained on the training set, and cross-validation is used to evaluate the model performance and optimize the hyperparameters.

[0026] The grid search or random search method is used to optimize the hyperparameters of the model, and different evaluation indicators are used to evaluate the performance of the model on the validation set and test set.

[0027] The real-time collected environmental data is input into the trained prediction model, and the model outputs the real-time load demand prediction value based on the input environmental features.

[0028] As preferred, based on the predicted load demand and the current state of the water cold storage pool, an optimization algorithm is used to dynamically adjust the cold distribution strategy. The method is:

[0029] Based on environmental data and historical load prediction results, real-time load demand is obtained, and the current state of the water cold storage pool includes current cold storage, maximum storage capacity of the cold storage pool, and cold storage and discharge rate of the cold storage pool.

[0030] According to the real-time load demand and the current state of the water storage pool, the distribution of cold energy is dynamically adjusted, the storage or release of cold energy is performed in advance through prediction of the change of load demand, and the load prediction result is balanced with the charging and discharging capacity of the water storage pool;

[0031] Based on the above, the use of the water storage pool is optimized through cold energy distribution optimization;

[0032] According to the above target, linear programming, mixed integer linear programming, dynamic programming, genetic algorithm or particle swarm optimization algorithm are used to dynamically adjust the cold energy distribution strategy, so as to maximize the use of the cold energy reserve of the water storage pool.

[0033] As a preferred, in the process of cold energy distribution, model predictive control technology is used to calculate the cold energy release scheme in the future period of time in advance according to real-time data and load prediction results, and the method is as follows:

[0034] The input real-time data includes the current system temperature, cooling load and environmental conditions, the cooling load demand in the future period of time is predicted according to historical data and external conditions, and the prediction method is based on machine learning and time series analysis technology;

[0035] The future cold energy distribution is optimized by minimizing energy consumption, load balance and response time, the model predictive control technology predicts the system state in the future period of time, calculates a series of control operations, and through rolling optimization, the control algorithm adjusts the cold energy distribution strategy in real time according to new measurement data and prediction results;

[0036] Through the model predictive control technology, the system calculates the cold energy release scheme in the future period of time in each control cycle, and based on the current load, predicted demand, environmental conditions and other data, the optimal cold energy distribution at each time point is calculated in advance for several hours or days.

[0037] As a preferred, in the process of system operation, the control strategy is adaptively adjusted through reinforcement learning algorithm, and the method for reinforcement learning to continuously optimize the control strategy through historical data and real-time feedback is as follows:

[0038] Through continuous interaction with the environment, reinforcement learning continuously adjusts the strategy according to real-time state, load demand and energy efficiency feedback data, through Q-learning or deep Q network reinforcement learning algorithm, the system adjusts the strategy according to actual feedback after each control cycle;

[0039] An initial control strategy or model is set, and the initial state information of the system is obtained, the system selects an action according to the current state, achieves the target through cold energy distribution, and obtains rewards or penalties according to feedback;

[0040] Through multiple interactions and feedback, the reinforcement learning algorithm continuously updates the strategy and adjusts the control action. Reinforcement learning optimizes historical data to respond to temperature changes, load fluctuations, and environmental changes, and makes real-time adjustments.

[0041] As a preferred, the system has an intelligent fault diagnosis mechanism built-in, which monitors the working state of the equipment in real time. When an abnormality is identified, it automatically triggers the backup control strategy for adjustment. The mechanism integrates and analyzes data from multiple sensors to ensure smooth switching when equipment fails.

[0042] Temperature, pressure, current, and vibration sensors are distributed throughout the equipment to monitor its health in real time. Data is transmitted to the central control system or edge computing platform in real time through PLC and embedded device data acquisition systems.

[0043] Data fusion processes data from multiple sensors using Kalman filtering and complementary filtering techniques to extract vibration patterns and current waveform state characteristics. Mathematical methods or machine learning models are used to identify whether the equipment is abnormal.

[0044] A physical model of the equipment is used to predict its state, and sensor data and model calculations are combined to detect faults. Machine learning models are trained on normal and fault state equipment data.

[0045] The system compares data with diagnostic models and identifies whether the equipment is in a fault state based on a set fault threshold. When an abnormality is found, the system marks it as a fault signal.

[0046] Redundant control strategies are designed for different types of equipment failures. When the main equipment cannot continue to work, it automatically switches to backup equipment or backup systems.

[0047] Hot backup control: The backup equipment remains in standby mode while the system is normal, ensuring that it can take over when the main equipment fails.

[0048] Cold backup control: The backup equipment is normally in an off state and starts when the main equipment fails.

[0049] Degraded control: If the backup equipment is not sufficient to run at full load, the system switches to degraded mode.

[0050] When a fault is detected, the system immediately switches to backup equipment or backup control systems. The system maintains process smoothness during the switching process.

[0051] The entire fault diagnosis process is performed in real time. When a fault occurs or is confirmed, the system provides real-time feedback on the fault and initiates switching operations. It also monitors the status of the backup equipment and continues to monitor it through sensor feedback.

[0052] Another technical problem to be solved by the present application is to provide an electronic device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the intelligent water storage cooling system optimization control method based on multivariate prediction and reinforcement learning as described above.

[0053] Another technical problem to be solved by the present application is to provide a computer readable storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the intelligent water storage cooling system optimization control method based on multivariate prediction and reinforcement learning.

[0054] The present application has the following advantages:

[0055] Through real-time load prediction and dynamic scheduling, energy waste and excessive distribution of cold energy are avoided, and the overall energy efficiency of the system is improved; the model predictive control technology and reinforcement learning algorithm are used to accurately adjust the system according to different environments and load demands, reduce the response lag of the system, and improve the control precision; the built-in intelligent fault detection and fault tolerance mechanism ensures that the system can operate stably even under equipment failure or external disturbance, and reduces the downtime caused by failure; the system can adaptively adjust according to different cooling demands and environmental changes, has strong flexibility, and is suitable for different types of buildings and load fluctuations. DETAILED DESCRIPTION

[0056] The principles and features of the present application are described below, and the examples are only used to explain the present application and are not intended to limit the scope of the present application. In the following paragraphs, the present application is described in more detail by way of example. The advantages and features of the present application will be more apparent from the following description and claims.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0058] Embodiments

[0059] The technical solution adopted by the present application to solve its technical problems is:

[0060] The intelligent water storage cooling system optimization control method based on multivariate prediction and reinforcement learning comprises the following steps:

[0061] Real-time collection of environmental temperature, humidity, load demand and storage pool state data by sensors;

[0062] Adopting machine learning algorithm, combining with the collected environmental data, a multivariate prediction model is generated for real-time prediction of load demand, which is trained based on environmental temperature, humidity, time period, and historical load factors;

[0063] According to the predicted load demand and the current state of the water storage pool, the distribution strategy of cold energy is dynamically adjusted through optimization algorithm;

[0064] In the process of cold energy distribution, model predictive control technology is adopted to calculate the cold energy release scheme in the future period in advance according to real-time data and load prediction results;

[0065] In the process of system operation, the control strategy is adaptively adjusted through reinforcement learning algorithm, which continuously optimizes the control strategy through historical data and real-time feedback;

[0066] The system has an intelligent fault diagnosis mechanism, which monitors the working state of the equipment in real time, and automatically triggers the standby control strategy for adjustment when an abnormality is identified. The mechanism ensures the system switches when a device failure occurs through data integration and analysis of multiple sensors.

[0067] Through real-time load prediction and dynamic cold energy distribution, the accuracy and efficiency of cold energy supply are ensured, and energy waste is reduced. Model predictive control and reinforcement learning enable the system to have high foresight and adaptive ability, and can flexibly adjust the strategy according to changes in environment and load demand. The built-in intelligent fault diagnosis and standby control strategy ensure that the system can respond quickly when a device failure occurs, ensuring smooth operation of the system. Based on the integration of sensor data, machine learning and reinforcement learning, the system can continuously optimize the control strategy according to real-time feedback and historical data, improving the intelligent level of the system.

[0068] The method for real-time collection of environmental temperature, humidity, load demand and storage pool state data by sensors is as follows:

[0069] Temperature sensors and humidity sensors are deployed in the target environment to collect air temperature and humidity data in real time;

[0070] Load sensors are used to monitor the cooling demand of the current system, and the cooling load demand is indirectly calculated through temperature difference, pressure change, and flow rate;

[0071] Temperature sensors in the storage pool are used to monitor the cold energy storage state of the storage pool, and pressure sensors are deployed to detect the pressure change in the storage pool to calculate the current cooling capacity of the storage pool;

[0072] All sensor data are transmitted to the central control system through real-time data acquisition units, and the data acquisition units have data preprocessing functions;

[0073] The central control system performs dynamic analysis based on real-time collected data, and monitors the environmental temperature, humidity, load demand and cold storage pool state in real time.

[0074] Through real-time data collection of temperature sensors, humidity sensors, load sensors and cold storage pool state monitoring devices, the system can always obtain the latest environmental information; through load sensors and indirect measurement methods, the system can accurately monitor cooling demand. These sensors help reflect the subtle changes in load fluctuations, avoiding excessive or insufficient cooling supply; deploying temperature sensors and pressure sensors in the cold storage pool not only enables real-time monitoring of the cold storage state of the cold storage pool, but also monitors the pressure changes in the pool, thereby calculating the current cooling capacity of the cold storage pool; the data of all sensors is transmitted to the central control system through real-time data acquisition units, and the data preprocessing function can reduce data noise and eliminate unnecessary information, improving the efficiency and accuracy of data transmission; through real-time data collection and dynamic analysis, the central control system can timely adjust the operation strategy and perform adaptive control according to environmental changes and load demand; the system can monitor various indicators such as cold storage pool, load, environment, etc. through real-time data, and the central control system can quickly identify and take necessary measures in case of abnormal conditions.

[0075] A multivariate prediction model is generated by using machine learning algorithms combined with collected environmental data for real-time prediction of load demand. The prediction model is trained based on environmental temperature, humidity, time period, and historical load factors using the following method:

[0076] The real-time collected temperature, humidity, time period, and historical load environmental data are processed for missing value processing, outlier detection and removal. If data is missing, it is completed by interpolation and mean filling methods.

[0077] A multivariate time series prediction machine learning algorithm is used, which includes linear regression, decision tree regression, random forest regression, support vector regression, and neural network.

[0078] Feature selection methods are used to select features that have a greater impact on load prediction, and to remove redundant and irrelevant features. The method used is:

[0079] The correlation between temperature, humidity, time period, and historical load features and load demand is calculated, and features with high correlation to load demand are used.

[0080] If there is multicollinearity between features, PCA is used to reduce the feature dimension and extract principal components.

[0081] The data set is divided into training set, validation set and test set, the selected machine learning model is trained on the training set, and cross-validation is used to evaluate the performance of the model and optimize the hyperparameters.

[0082] The grid search or random search method is used for model hyperparameter tuning, and different evaluation indicators are used to evaluate the performance of the model on the validation set and test set.

[0083] The real-time collected environmental data is input into the trained prediction model, and the model outputs the real-time load demand prediction value according to the input environmental characteristics.

[0084] By training a multivariate prediction model based on environmental data, the load demand can be accurately predicted under the influence of various factors, especially when the external conditions such as temperature and humidity change, accurate load prediction can be provided in time; real-time prediction of load demand can help the system dynamically adjust the cold distribution and optimize energy use, avoid energy waste, and improve the accuracy of cold supply; through data cleaning methods such as missing value processing, outlier detection and removal, it can ensure that the data used for training is of high quality and consistency; feature selection and PCA dimensionality reduction help to extract features that have a greater impact on load demand prediction, reducing redundant and irrelevant features; cross-validation is used to evaluate the performance of the model, and grid search or random search methods are used to optimize the hyperparameters of the model, which can further improve the accuracy and generalization ability of the prediction model; this scheme can output real-time load demand prediction values based on real-time collected environmental data and trained models, supporting real-time decision-making.

[0085] According to the predicted load demand and the current state of the water storage pool, the method for dynamically adjusting the cold distribution strategy is:

[0086] Based on environmental data and historical load prediction results, real-time load demand is obtained, and the current state of the water storage pool includes current cold storage, maximum storage capacity of the storage pool, and charging and discharging rate of the storage pool;

[0087] According to the real-time load demand and the current state of the water storage pool, the cold distribution is dynamically adjusted, the change of load demand is predicted, and the storage or release of cold is performed in advance, and the load prediction result is balanced with the charging and discharging capacity of the water storage pool;

[0088] Based on the above, the cold distribution is optimized to use the water storage pool;

[0089] According to the above target, linear programming, mixed integer linear programming, dynamic programming, genetic algorithm or particle swarm optimization algorithm are used to dynamically adjust the cold distribution strategy to maximize the use of cold storage of the water storage pool.

[0090] By dynamically adjusting the allocation of cold energy, the load demand and the storage capacity of the water storage pool can be effectively balanced, over-storage or unnecessary cold energy release can be avoided, and the maximum use of cold energy can be ensured. By reasonably adjusting the allocation of cold energy, the provision of cold energy through traditional cooling methods during high load can be avoided, and the energy procurement cost during peak hours can be reduced. Predicting load demand and storing or releasing cold energy in advance can avoid the risk of insufficient or excessive cold energy caused by load fluctuations, thereby improving the stability of the system. Using a prediction model combined with an optimization algorithm strategy, the cold energy allocation is not only based on current demand but also considers future demand changes, so that the system can adapt to different load change scenarios. Using multiple optimization algorithms, the most suitable algorithm can be flexibly adjusted and selected according to different application scenarios and actual situations, improving the optimization effect.

[0091] During the cold energy allocation process, the model predictive control technology is used to calculate the cold energy release scheme for the future period in advance based on real-time data and load prediction results:

[0092] The input real-time data includes the current system temperature, cooling load, and environmental conditions. The future cooling load demand for a period of time is predicted based on historical data and external conditions. The prediction method is based on machine learning and time series analysis technology.

[0093] By minimizing energy consumption, load balancing, and response time, the future allocation of cold energy is optimized. The model predictive control technology predicts the system state for a period of time in the future, calculates a series of control operations, and through rolling optimization, the control algorithm adjusts the cold energy allocation strategy in real time based on new measurement data and prediction results.

[0094] Through the model predictive control technology, the system calculates the cold energy release scheme for the future period in each control cycle. Based on current load, predicted demand, environmental conditions, and other data, the optimal cold energy allocation at each time point is calculated several hours or days in advance.

[0095] By predicting future cooling load and releasing or storing cold energy in advance, energy consumption increases caused by emergency response can be avoided. Minimizing energy consumption as an optimization goal, combined with factors such as time-of-use electricity prices, the system will preferentially store cold energy during low electricity price periods and release it during high electricity price periods, optimizing operating costs. The rolling optimization mechanism allows the system to re-optimize the cold energy allocation scheme in each control cycle, even if the prediction is biased, it can be corrected in the next cycle, enhancing the robustness of the system. MPC allows system constraints such as cold energy reserve limits, chiller start-stop constraints, cold storage pool charging and discharging rates, comfort temperature constraints, and others to be modeled uniformly. Machine learning and time series prediction models can be seamlessly integrated as load prediction modules to improve prediction accuracy.

[0096] During system operation, the control strategy is adaptively adjusted through reinforcement learning algorithm. Reinforcement learning continuously optimizes the control strategy through historical data and real-time feedback:

[0097] Reinforcement learning continuously adjusts the strategy according to real-time state, load demand, and energy efficiency feedback data through continuous interaction with the environment. Through Q-learning or deep Q network reinforcement learning algorithm, the system adjusts the strategy according to actual feedback after each control period.

[0098] Set the initial control strategy or model and obtain the initial state information of the system. The system selects actions based on the current state, achieves the target through cold distribution, and obtains rewards or penalties based on feedback.

[0099] Through multiple interactions and feedback, the reinforcement learning algorithm continuously updates the strategy and adjusts the control actions. Reinforcement learning optimizes based on historical data to respond to temperature changes, load fluctuations, and environmental changes for real-time adjustment.

[0100] Reinforcement learning can adaptively adjust the strategy in the face of different environments, load fluctuations, and device conditions, and is more flexible and robust than traditional rule-based or model predictive control methods. Reinforcement learning does not require accurate modeling of the complex physical or mathematical relationships of the cold distribution system in advance. Through interaction with the environment, the system can learn the optimal control strategy through experience. Reinforcement learning algorithm continuously optimizes the strategy during long-term operation. Even if the initial control strategy of the system is simple or not ideal, the strategy will gradually optimize and improve system performance as the number of interactions increases. Reinforcement learning not only optimizes energy consumption, but also considers multiple goals such as load balancing, comfort, response time, and device service life. The system can find a balance between multiple constraints and goals. Reinforcement learning is suitable for nonlinear and complex systems, such as the complex nonlinear relationship between temperature, load, and environmental factors in the cold distribution process.

[0101] The system has an intelligent fault diagnosis mechanism that monitors the working state of the device in real time. When an abnormality is identified, it automatically triggers a backup control strategy for adjustment. The mechanism integrates and analyzes data from multiple sensors to ensure smooth switching in the event of device failure.

[0102] Temperature, pressure, current, and vibration sensors are distributed throughout the device to monitor its health in real time. Data is transmitted to the central control system or edge computing platform in real time through PLC and embedded device data acquisition systems.

[0103] Data fusion processes data from multiple sensors through Kalman filtering and complementary filtering techniques, extracts vibration patterns and current waveform state characteristics, and uses mathematical methods or machine learning models to identify whether the equipment is abnormal.

[0104] A physical model of the equipment is used to predict its state, and sensor data and model calculations are combined to detect faults.

[0105] The system compares data with diagnostic models and identifies whether the equipment is in a fault state through a set fault threshold. When an anomaly is found, the system marks it as a fault signal.

[0106] According to different fault types of the equipment, a redundant control strategy is designed. When the main equipment cannot continue to work, it automatically switches to the standby equipment or backup system:

[0107] Hot backup control: the standby equipment works normally but is in standby state to ensure that it takes over when the main equipment fails.

[0108] Cold backup control: the standby equipment is normally in an off state and starts when the main equipment fails.

[0109] Degraded control: if the standby equipment is not sufficient to run at full load, the system switches to degraded mode.

[0110] When a fault is detected, the system immediately switches to the standby equipment or backup control system, and the system will maintain the smoothness of the process during the switching process.

[0111] The entire fault diagnosis process is real-time. Whenever a fault occurs or is confirmed, the system provides real-time feedback on the fault information and starts switching operations, and monitors the status of the standby equipment. Through sensor feedback data, the system will continue to monitor.

[0112] Through multi-layer redundancy strategy, the system will not be paralyzed when the equipment fails, ensuring continuous operation. Based on PLC and edge computing platform, data acquisition, processing and response have millisecond or second level timeliness, meeting the needs of high real-time scenarios. With the help of multi-sensor fusion and AI recognition technology, the accuracy and advance of fault detection are improved, reducing false positives and false negatives. The entire system can automatically adjust the strategy according to different equipment, working conditions and operating conditions, improving flexibility. Cold backup does not consume energy when the equipment is not working, and hot backup runs with low power consumption, effectively balancing reliability and energy efficiency. In the future, more types of sensors can be connected, more standby strategies can be expanded, or more complex intelligent algorithms can be introduced. Fault detection, decision-making, switching and feedback are automatically executed, greatly reducing the risk of manual intervention and misoperation.

[0113] The embodiment also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the intelligent water storage cooling system optimization control method based on multivariate prediction and reinforcement learning when executing the program.

[0114] The embodiment also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the intelligent water storage cooling system optimization control method based on multivariate prediction and reinforcement learning.

[0115] Real-time environmental data collection is performed through a sensor network, and the types and configurations of the sensors are adjusted according to actual applications to collect environmental data in multiple dimensions including temperature, humidity, and system load state.

[0116] Load prediction adopts a multivariate dynamic modeling method, and the cold demand of the system in a future period of time is predicted through analysis of the environmental data, and the process combines multiple statistical learning or machine learning algorithms to dynamically adapt to load changes.

[0117] In the application, the adaptive adjustment and learning mechanism of the system performs real-time strategy optimization through an intelligent optimization algorithm, the optimization mechanism combines reinforcement learning or similar machine learning technology, adjusts according to real-time feedback of the system, and continuously learns and improves the control strategy, the learning mechanism automatically adjusts according to the input of the multidimensional data, adapts to environmental changes and load fluctuations, and thus more efficient cold management and system operation are realized.

[0118] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of each method can be included. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0119] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the system is divided into different functional units or modules to complete all or part of the above-mentioned functions.

[0120] The above-mentioned embodiments of the present application are not a limitation on the protection scope of the present application, and the embodiments of the present application are not limited thereto. According to the above-mentioned content of the present application, other various forms of modification, replacement or change of the above-mentioned structure of the present application can be made according to ordinary technical knowledge and conventional means in the art without departing from the above-mentioned basic technical idea of the present application, which shall fall within the protection scope of the present application.

Claims

1. An intelligent chilled water storage system optimization control method based on multivariate prediction and reinforcement learning, characterized in that, The method comprises the following steps: Real-time collection of environmental temperature, humidity, load demand and cold storage pool state data through sensors; Using machine learning algorithms, combined with collected environmental data, a multivariate prediction model is generated to predict real-time load demand. This prediction model is trained based on environmental temperature, humidity, time period, and historical load factors; According to the predicted load demand and the current state of the water cold storage pool, the distribution strategy of cold energy is dynamically adjusted through an optimization algorithm; During the distribution of cold energy, model predictive control technology is used to calculate the cold energy release plan for the future period in advance based on real-time data and load prediction results; During system operation, the control strategy is adaptively adjusted through reinforcement learning, which continuously optimizes the control strategy through historical data and real-time feedback; The system has an intelligent fault diagnosis mechanism that monitors the working state of the equipment in real time. When an anomaly is identified, the backup control strategy is automatically triggered for adjustment. The mechanism ensures system switching in the event of equipment failure through data integration and analysis of multiple sensors. 2.The intelligent chilled water storage system optimization control method based on multivariate prediction and reinforcement learning of claim 1, wherein, The method for real-time collection of environmental temperature, humidity, load demand and cold storage pool state data is as follows: Temperature sensors and humidity sensors are deployed in the target environment to collect real-time air temperature and humidity data; Load sensors are used to monitor the cooling demand of the current system, and the cooling load demand is calculated indirectly through temperature difference, pressure change, and flow rate; Temperature sensors in the cold storage pool are used to monitor the cold storage state of the cold storage pool, and pressure sensors are used to detect the pressure change in the cold storage pool to calculate the current cooling capacity of the cold storage pool; All sensor data is transmitted to the central control system through real-time data acquisition units, which have data preprocessing functions; The central control system dynamically analyzes the real-time collected data to monitor the environmental temperature, humidity, load demand and cold storage pool state in real time.

3. The intelligent chilled water storage system optimization control method based on multivariate prediction and reinforcement learning according to claim 2, characterized in that, The method for generating a multivariate prediction model using machine learning algorithms combined with collected environmental data to predict real-time load demand is as follows: The real-time collected temperature, humidity, time period, and historical load environmental data are processed for missing values, abnormal value detection and removal. If data is missing, it is completed by interpolation and mean filling methods; A multivariate time series prediction machine learning algorithm is used, which includes linear regression, decision tree regression, random forest regression, support vector regression, and neural network; Feature selection methods are used to select features that have a greater impact on load prediction, remove redundant and irrelevant features, and the methods used are: Calculate the correlation between temperature, humidity, time period, historical load features and load demand, and use features that are highly correlated with load demand; If there is multicollinearity between features, use PCA to reduce feature dimension and extract principal components; Divide the data set into training set, validation set and test set, train the selected machine learning model on the training set, use cross-validation to evaluate the model performance and optimize the hyperparameters; The grid search or random search method is used for the hyperparameter tuning of the model, and different evaluation indexes are used to evaluate the performance of the model on the validation set and the test set. The real-time collected environmental data is input into the trained prediction model, and the model outputs the real-time load demand prediction value according to the input environmental characteristics.

4. The intelligent chilled water storage system optimization control method based on multivariate prediction and reinforcement learning according to claim 3, characterized in that, According to the predicted load demand and the current state of the water storage pool, the method for dynamically adjusting the cold distribution strategy is: Based on the environmental data and historical load prediction results, the real-time load demand is obtained, and the current state of the water storage pool includes the current cold storage capacity, the maximum storage capacity of the water storage pool, and the charging and discharging rate of the water storage pool. According to the real-time load demand and the current state of the water storage pool, the cold distribution is dynamically adjusted, the change of the load demand is predicted, and the storage or release of the cold is performed in advance, and the load prediction result is balanced with the charging and discharging capacity of the water storage pool. Based on the above, the cold distribution optimization of the water storage pool is performed. According to the above target, linear programming, mixed integer linear programming, dynamic programming, genetic algorithm or particle swarm optimization algorithm is used to dynamically adjust the cold distribution strategy, and the cold storage capacity of the water storage pool is maximized.

5. The intelligent chilled water storage system optimization control method based on multivariate prediction and reinforcement learning according to claim 4, characterized in that, In the cold distribution process, the model predictive control technology is used to calculate the cold release scheme in the future period of time in advance according to the real-time data and the load prediction result. The real-time data is input, including the current system temperature, cold load, and environmental conditions, the future cold load demand in a period of time is predicted according to the historical data and external conditions, and the prediction method is based on machine learning and time series analysis technology. The future cold distribution is optimized by minimizing energy consumption, load balance, and response time, the model predictive control technology predicts the system state in the future period of time, calculates a series of control operations, and adjusts the cold distribution strategy in real time according to the new measurement data and prediction result through rolling optimization. Through the model predictive control technology, the system calculates the cold release scheme in the future period of time in each control cycle, and calculates the optimal cold distribution at each time point based on the current load, predicted demand, environmental conditions and other data several hours or days in advance.

6. The intelligent chilled water storage system optimization control method based on multivariate prediction and reinforcement learning according to claim 5, characterized in that, In the system running process, the control strategy is adaptively adjusted through the reinforcement learning algorithm, and the reinforcement learning algorithm continuously optimizes the control strategy through historical data and real-time feedback. The reinforcement learning algorithm continuously adjusts the strategy according to the real-time state, load demand, and energy efficiency feedback data through continuous interaction with the environment, and the system adjusts the strategy according to the actual feedback after each control cycle through Q-learning or deep Q network reinforcement learning algorithm. The initial control strategy or model is set, and the initial state information of the system is obtained, the system selects actions according to the current state, achieves the target through cold distribution, and obtains rewards or penalties according to feedback. Through multiple interactions and feedback, the reinforcement learning algorithm continuously updates the strategy and adjusts the control action, and the reinforcement learning algorithm optimizes the strategy according to historical data to respond to temperature changes, load fluctuations and environmental changes, and adjusts in real time.

7. The intelligent chilled water storage system optimization control method based on multivariate prediction and reinforcement learning according to claim 6, characterized in that, The system has an intelligent fault diagnosis mechanism that monitors the working state of the equipment in real time. When an abnormality is identified, the system automatically triggers a backup control strategy for adjustment. The mechanism integrates and analyzes data from multiple sensors to ensure that the system can switch when a device fails. Temperature, pressure, current, and vibration sensors are distributed throughout the equipment to monitor its health in real time. Data is transmitted to the central control system or edge computing platform in real time through PLC and embedded device data acquisition systems. Data fusion processes data from multiple sensors using Kalman filtering and complementary filtering techniques to extract vibration patterns and current waveform state characteristics. Mathematical methods or machine learning models are used to identify whether the equipment is abnormal. A physical model of the equipment is used to predict its state. Sensor data and model calculations are combined to detect faults, or machine learning models are used to train equipment data under normal and fault conditions. The system compares data with diagnostic models and identifies whether the equipment is in a fault state based on a set fault threshold. When an abnormality is found, the system marks it as a fault signal. Redundant control strategies are designed based on different fault types of the equipment. When the main equipment cannot continue to work, the system automatically switches to a backup device or backup system. Hot backup control: The backup device remains in standby mode while the system is normal to ensure that it can take over when the main device fails. Cold backup control: The backup device is normally in an off state and starts when the main device fails. Degraded control: If the backup device is not sufficient to run at full load, the system switches to degraded mode. When a fault is detected, the system immediately switches to a backup device or backup control system. The system maintains process smoothness during the switching process. The entire fault diagnosis process is performed in real time. When a fault occurs or is confirmed, the system provides real-time feedback on the fault information and starts switching operations. The system monitors the status of the backup device and continues to monitor it through sensor feedback.

8. An electronic device, comprising: The computer program product includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the intelligent water storage and cooling system optimization control method based on multivariate prediction and reinforcement learning as claimed in any one of claims 1-7.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the intelligent water storage and cooling system optimization control method based on multivariate prediction and reinforcement learning as claimed in any one of claims 1-7.