Multi-parameter collaborative intelligent control system for cold and hot machine rooms based on edge computing

Abstract

The present application relates to the technical field of cold and heat machine room control, in particular to a multi-parameter collaborative intelligent control system for cold and heat machine rooms based on edge computing, which comprises a perception layer, an edge computing layer and an execution layer, uses an LSTM neural network to predict cold and heat loads, dynamically corrects PID proportionality coefficients and introduces a feedforward signal, constructs a DQN multi-target weighted reward function based on traditional PID energy consumption, and realizes collaborative regulation and control of energy consumption, parameter compliance and equipment stability. A sensor precision weighted fusion algorithm is used to build a triple adaptive mechanism, a double abnormality detection mode is designed, and the edge computing layer and the cloud breakpoint continue transmission. The perception layer sensor is wirelessly hot-pluggable, and the execution layer forms a closed loop with a feedback module. The present application improves the accuracy and timeliness of regulation and control, reduces the dependence on the cloud, realizes energy consumption optimization and rapid abnormality processing, and adapts to dynamic changes in machine room load.

Description

Technical Field

[0001] This invention relates to the field of cold and hot data center control technology, specifically to a multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing. Background Technology

[0002] Cold and heating equipment rooms are core energy control units in various buildings and industrial production processes. Their operating status directly affects environmental stability, equipment lifespan, and overall energy consumption. Traditional cold and heating equipment rooms often employ a single PID control mode, which can only passively adjust based on the current load, resulting in problems such as adjustment lag and over-adjustment. At the same time, traditional control methods do not consider multiple objectives such as energy consumption, parameter compliance, and equipment stability in a coordinated manner, which can easily lead to the sacrifice of other indicators in pursuit of a single indicator. For example, excessive pursuit of parameter compliance may cause a surge in energy consumption, or the attempt to reduce energy consumption may cause equipment overload operation.

[0003] In addition, traditional data center control relies heavily on centralized cloud computing, which makes it easy to lose control capabilities when the network is interrupted. Furthermore, the multi-dimensional operating parameters collected by sensors are not accurately fused, resulting in large parameter errors. The parameters and sensor fusion weights of existing control algorithms are mostly fixed values, which cannot be dynamically adjusted according to the actual operating status of the data center. This leads to a decrease in control accuracy over time and makes it difficult to adapt to the dynamic changes in the data center's hot and cold loads. Summary of the Invention

[0004] To address the shortcomings of existing technologies, a multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing is provided. This system solves problems such as lag in traditional PID control, imbalance in multi-objective regulation, low accuracy of sensor parameter fusion, inability to adaptively adjust algorithm parameters, and high dependence on the cloud. It achieves multi-objective collaborative intelligent control of cold and hot data centers with the lowest energy consumption, compliant environmental parameters, and stable equipment operation.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by this invention is a multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing. This system achieves intelligent multi-parameter control of the data center through the collaborative use of a PID correction coefficient calculation formula and a DQN multi-objective weighted reward function based on traditional PID energy consumption. The system includes: A formula for calculating the PID correction coefficient is constructed, and the PID proportional coefficient is dynamically adjusted based on the predicted values ​​of the cooling and heating loads of the cooling and heating room. The formula is as follows: = 1 + (Predicted load - Current load) / Current load ;in, For PID correction coefficients, This is the initial proportional coefficient of the PID execution unit. This is the corrected PID proportional coefficient; when When >1, it increases proportionally. To speed up the adjustment process, when When it is less than 1, it decreases proportionally. To avoid over-adjustment, the predicted cold and hot load values ​​are used as feedforward signals for the PID execution unit to adjust the PID control target threshold in advance. Construct a DQN multi-objective weighted reward function with traditional single PID control as the energy consumption benchmark, the expression of which is: ,and ;in, For energy consumption weight, For parameter compliance weights, To ensure stable weighting of the equipment, This is an energy consumption reward item, calculated as follows: =1-(Current data center energy consumption / Baseline energy consumption), where the baseline energy consumption is the average energy consumption of the cold and hot data center under the traditional single PID control mode; Rewards for meeting parameter targets. This is a reward for equipment stability. The reward function guides the DQN algorithm to discover the correlation patterns of data center parameters, outputs the optimization direction of equipment operating parameters, and combines it with the PID control after being corrected by the above formula to output precise operating control commands for each execution device in the cold and hot data centers, so as to achieve multi-objective coordinated regulation of data center energy consumption, environmental parameter compliance, and stable equipment operation.

[0006] The aforementioned edge computing-based multi-parameter collaborative intelligent control system for cold and hot data centers, The value is 0.4. The value is 0.4. The value is 0.2; The value selection rule is: when the computer room environmental parameters and pipeline operation parameters are all within the preset threshold range. =1, otherwise =0.1; the The value selection rule is: when the core equipment in the computer room is not overloaded or faulty. =1, when the core equipment is slightly overloaded =0.5, when the core equipment fails. =0.

[0007] The aforementioned edge computing-based multi-parameter collaborative intelligent control system for cold and hot data centers uses an LSTM neural network processing unit as the input to output the predicted cold and hot load values ​​for the next 1-2 hours. The LSTM neural network processing unit takes the multi-dimensional operating parameter sequence of the cold and hot data center after fusion processing within the past hour as input and outputs the predicted cold and hot load values ​​for the data center for the next 1-2 hours. The multi-dimensional operating parameters include pipeline medium pressure P, pipeline medium flow rate Q, core equipment operating power W, data center ambient temperature T, and data center ambient humidity H. Each parameter is standardized and mapped to the [0,1] interval, and the multi-dimensional operating parameters serve as the state space State of the DQN algorithm, i.e., State=[P,Q,W,T,H].

[0008] The aforementioned multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing further includes a sensing layer and an edge computing layer. The sensing layer collects the original multi-dimensional operating parameters of the data center and transmits them to the edge computing layer. The edge computing layer uses a sensor accuracy weighted fusion algorithm to fuse the original multi-dimensional operating parameters. The weighting calculation process of this sensor accuracy weighted fusion algorithm is as follows: based on the nominal accuracy error value of each type of sensor, the accuracy weight coefficient of each type of sensor is calculated. ,in, For the first The nominal accuracy error value of the sensor class; the accuracy weight coefficients of all single-class sensors are normalized to obtain the normalized weights. , The number of similar sensors participating in the fusion; based on The raw parameters collected by multiple sensors of the same type are weighted and summed to obtain the fused value of a single type of parameter.

[0009] The aforementioned edge computing-based multi-parameter collaborative intelligent control system for cold and hot data centers also includes an execution layer. This execution layer is communicatively connected to PID execution units, and each execution unit is equipped with a feedback module. The feedback module collects the operating status and changes in data center parameters after the equipment performs control operations and generates feedback data, which is then transmitted to the collaborative control layer. The system constructs a triple adaptive mechanism. The collaborative control layer calculates the normalized weight update amount of the sensor accuracy-weighted fusion algorithm based on the feedback data, realizing adaptive adjustment of the sensor fusion weights. The specific update algorithm is as follows: ,in, For the updated number Sensor-like normalized weights, For the previous version Sensor-like normalized weights, For the first The difference between the absolute value of the error between the parameters acquired by the sensor class and the fused parameters and the average error of all sensors. For the first The absolute value of the error between the parameters acquired by the sensor and the fused parameters. This represents the average error of sensors of the same type.

[0010] The aforementioned edge computing-based multi-parameter collaborative intelligent control system for cold and hot data centers further includes a triple adaptive mechanism: the collaborative control layer iteratively optimizes the training parameters of the LSTM neural network operation unit and the reward function weights of the DQN algorithm based on the feedback data, achieving dual adaptive adjustment of algorithm parameters and sensor fusion weights; simultaneously, the edge computing layer optimizes the local collaborative decision-making logic in conjunction with the feedback data, forming a triple adaptive control system of "adaptive algorithm parameters, adaptive sensor fusion weights, and adaptive local decision-making logic".

[0011] The aforementioned multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing involves the edge computing layer performing dual anomaly detection—threshold and trend analysis—on the preprocessed parameters. The trend analysis employs a mathematical model combining first-order linear fitting and a sliding window, specifically implemented by setting the sliding window length to... , For 5-10 consecutive acquisition cycles, the parameter sequence within the window... Perform a first-order linear fit, the fitting formula is as follows: ,in, The slope is the fitted slope. The intercept is used to calculate the slope. The absolute value and the preset trend threshold The ratio when When parameters show abnormal change trends, a local early warning is immediately triggered and an emergency control command is generated when the parameter data exceeds the preset value threshold or is determined to be an abnormal change trend by the above trend analysis model. At the same time, the abnormal data and the key operating data of the computer room are synchronized to the cloud management platform.

[0012] The aforementioned edge computing-based multi-parameter collaborative intelligent control system for cold and hot data centers includes a front-end computing layer that performs data denoising, outlier removal, and data standardization before fusing the original multi-dimensional operating parameters. The sensing layer includes temperature sensors, humidity sensors, pressure sensors, flow sensors, and equipment status sensors. All sensors are connected to the edge nodes of the edge computing layer via wireless communication and support hot-swapping. Temperature and humidity sensors are evenly deployed in various areas of the data center, pressure and flow sensors are deployed at key nodes of the cold and hot piping, and equipment status sensors are deployed on core equipment such as chillers, heaters, fans, and water pumps to collect the operating power, operating time, and fault signals of the core equipment.

[0013] The aforementioned multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing adopts a breakpoint resume mechanism between the edge computing layer and the cloud management platform. When the network is interrupted, the edge computing layer independently completes the local control of the data center. When the network is restored, the data center operation data, system control records, and anomaly detection records are automatically synchronized to the cloud management platform, enabling remote monitoring of the data center, historical data tracing, and anomaly tracing by the cloud.

[0014] The aforementioned edge computing-based multi-parameter collaborative intelligent control system for cold and hot data centers includes an execution layer comprising a refrigeration execution unit, a heating execution unit, a ventilation execution unit, and a valve execution unit. The refrigeration execution unit is connected to the chiller and chilled water pump, the heating execution unit is connected to the heater and hot water pump, the ventilation execution unit is connected to the fan, and the valve execution unit is connected to various valves in the cold and hot pipelines. The refrigeration and heating execution units achieve regulation by adjusting the start / stop status and operating power of the corresponding equipment, the ventilation execution unit adjusts the ventilation volume by adjusting the fan speed, and the valve execution unit controls the flow rate of the pipeline medium by adjusting the valve opening. Each execution unit completes collaborative regulation operations based on the precise operating control commands output by the PID execution unit.

[0015] The beneficial effects of the multi-parameter collaborative intelligent control system for cold and hot computer rooms based on edge computing in this invention are that it achieves PID feedforward regulation through load prediction, and combined with dynamically corrected PID proportional coefficient, it eliminates the regulation lag problem of traditional PID control, while avoiding over-regulation and improving the timeliness and accuracy of regulation. A DQN multi-objective weighted reward function is constructed based on the energy consumption of traditional PID control. It takes into account the three core indicators of energy consumption, parameter compliance, and equipment stability, and solves the imbalance problem of traditional single-objective control. It achieves the minimum energy consumption while ensuring the normal operation of the computer room. A sensor accuracy-weighted fusion algorithm is used to fuse multi-dimensional operating parameters and adaptively adjust the fusion weights, eliminating the influence of sensor errors and dimensional differences, and providing high-quality data for control decisions. A triple adaptive mechanism is constructed to enable the sensor fusion weights, algorithm training parameters, and edge decision logic to be dynamically optimized based on the actual operation feedback of the data center, ensuring the long-term stability of system control accuracy and adapting to the dynamic changes in data center load. The system utilizes an edge computing layer to perform parameter processing, algorithm computation, anomaly detection, and local control. Combined with a breakpoint resume mechanism, the system can continue to work independently even when the network is interrupted, reducing dependence on the cloud and improving system reliability. Employing a dual anomaly detection mechanism combining threshold and trend analysis, it can promptly detect abnormal parameter values ​​and trends, triggering local early warnings and emergency controls, while simultaneously synchronizing with the cloud to achieve rapid anomaly processing and tracing. The sensing layer sensors use wireless communication and support hot-swapping, which facilitates sensor deployment, expansion, and maintenance, reducing system operation and maintenance costs. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the operation flow of the control system of the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, the edge computing-based multi-parameter collaborative intelligent control system for cold and hot data centers achieves intelligent multi-parameter control of the data center through the collaborative calculation formula of PID correction coefficient and the DQN multi-objective weighted reward function with traditional PID as the energy consumption benchmark. This includes: A formula for calculating the PID correction coefficient is constructed, and the PID proportional coefficient is dynamically adjusted based on the predicted values ​​of the cooling and heating loads of the cooling and heating room. The formula is as follows: = 1 + (Predicted load - Current load) / Current load ;in, For PID correction coefficients, This is the initial proportional coefficient of the PID execution unit. This is the corrected PID proportional coefficient; when When >1, it increases proportionally. To speed up the adjustment process, when When it is less than 1, it decreases proportionally. To avoid over-adjustment, the predicted values ​​of hot and cold loads are used as feedforward signals for the PID execution unit to adjust the target threshold of PID control in advance; this enables advance control of load changes and eliminates the adjustment lag problem of traditional PID control.

[0019] Construct a DQN multi-objective weighted reward function with traditional single PID control as the energy consumption benchmark, the expression of which is: ,and ;in, For energy consumption weight, For parameter compliance weights, To ensure stable weighting of the equipment, This is an energy consumption reward item, calculated as follows: =1-(Current data center energy consumption / Baseline energy consumption), where the baseline energy consumption is the average energy consumption of the cold and hot data center under the traditional single PID control mode; The lower the current energy consumption, the larger the value, guiding the algorithm to optimize towards lower energy consumption. Rewards for meeting parameter targets. This is a reward for equipment stability. The reward function guides the DQN algorithm to discover the correlation patterns of data center parameters, outputs the optimization direction of equipment operating parameters, and combines it with the PID control after being corrected by the above formula to output precise operating control commands for each execution device in the cold and hot data centers, so as to achieve multi-objective coordinated regulation of data center energy consumption, environmental parameter compliance, and stable equipment operation.

[0020] The value is 0.4. The value is 0.4. The value is set to 0.2; this weighting ratio balances energy consumption optimization with the basic stability of the data center operation. The value selection rule is: when the computer room environmental parameters and pipeline operation parameters are all within the preset threshold range. =1, otherwise =0.1; the guiding algorithm ensures that the core operating parameters of the computer room meet the standards; the The value selection rule is: when the core equipment in the computer room is not overloaded or faulty. =1, when the core equipment is slightly overloaded =0.5, when the core equipment fails. =0; the guiding algorithm avoids abnormal device operation.

[0021] The reward function guides the DQN algorithm to discover the correlation patterns of equipment room parameters, outputs the optimization direction of equipment operating parameters, and combines it with the PID control modified by the above formula to output precise operating control commands for each execution device in the cold and hot equipment room, thereby achieving multi-objective coordinated regulation.

[0022] The predicted heating and cooling load values ​​are output by an LSTM neural network processing unit. This unit takes the multi-dimensional operating parameter sequence of the heating and cooling room over the past hour (after fusion processing) as input and outputs the predicted heating and cooling load values ​​for the next 1-2 hours, providing a data foundation for PID feedforward regulation and coefficient correction. The multi-dimensional operating parameters include pipeline medium pressure P, pipeline medium flow rate Q, core equipment operating power W, room ambient temperature T, and room ambient humidity H. Each parameter is standardized and mapped to the [0,1] interval, and these multi-dimensional operating parameters serve as the state space State of the DQN algorithm, i.e., State=[P,Q,W,T,H]. Standardization eliminates the dimensional differences between parameters, improving the prediction accuracy of the LSTM neural network.

[0023] It also includes a perception layer and an edge computing layer. The perception layer collects raw multi-dimensional operating parameters of the computer room and transmits them to the edge computing layer. The edge computing layer uses a sensor accuracy weighted fusion algorithm to fuse the raw multi-dimensional operating parameters. The weighting calculation process of this sensor accuracy weighted fusion algorithm is as follows: based on the nominal accuracy error value of each type of sensor, the accuracy weight coefficient of each type of sensor is calculated. ,in, For the first The nominal accuracy error value of a sensor class is used; the higher the sensor accuracy (the smaller the error), the larger the accuracy weighting coefficient. The accuracy weighting coefficients of all single-class sensors are normalized to obtain the normalized weights. , To ensure the number of similar sensors participating in the fusion is equal, normalization is applied to guarantee that the sum of the weights is 1, thus avoiding weight imbalance; based on By weighted summation of raw parameters collected by multiple sensors of the same type, a fusion value of a single type of parameter is obtained, thus achieving accurate fusion of parameters from multiple sensors of the same type.

[0024] It also includes an execution layer, which is communicatively connected to the PID execution units, and each execution unit is equipped with a feedback module. The feedback module collects the operating status and changes in equipment room parameters after the device performs control operations and generates feedback data, which is then transmitted to the collaborative control layer. The system constructs a triple adaptive mechanism. The collaborative control layer calculates the normalized weight update amount of the sensor accuracy weighted fusion algorithm based on the feedback data, realizing the adaptive adjustment of the sensor fusion weights. The specific update algorithm is as follows: ,in, For the updated number Sensor-like normalized weights, For the previous version Sensor-like normalized weights, For the first The difference between the absolute value of the error between the parameters acquired by the sensor class and the fused parameters and the average error of all sensors. For the first The absolute value of the error between the parameters acquired by the sensor and the fused parameters. This represents the average error of sensors of the same type. The update algorithm assigns higher weights to sensors with smaller acquisition errors, further improving parameter fusion accuracy.

[0025] The triple adaptive mechanism also includes: the collaborative control layer iteratively optimizes the training parameters of the LSTM neural network operation unit and the reward function weights of the DQN algorithm based on the feedback data, so as to achieve dual adaptive adjustment of algorithm parameters and sensor fusion weights; at the same time, the edge computing layer optimizes the local collaborative decision-making logic in combination with the feedback data, forming a triple adaptive control system of "adaptive algorithm parameters, adaptive sensor fusion weights, and adaptive local decision-making logic".

[0026] Based on feedback data, the collaborative control layer iteratively optimizes the training parameters of the LSTM neural network operation unit and the reward function weights of the DQN algorithm, so that the load prediction model and the multi-objective optimization algorithm are adapted to the actual operating status of the data center, avoiding the decrease in model accuracy caused by changes in the data center operating environment.

[0027] The edge computing layer optimizes local collaborative decision-making logic by combining feedback data and adjusts the control strategies at the edge, ensuring that the independent control capabilities of the edge match the real-time operational needs of the data center, thereby improving the accuracy of edge control. The edge computing layer performs dual anomaly detection—threshold and trend analysis—on the preprocessed parameters to promptly identify parameter anomalies and trigger emergency handling. The trend analysis employs a mathematical model combining first-order linear fitting and a sliding window, specifically implemented by setting the sliding window length to... , For 5-10 consecutive acquisition cycles, the parameter sequence within the window... Perform a first-order linear fit, the fitting formula is as follows: ,in, The slope is the fitted slope. The intercept is used to calculate the slope. The absolute value and the preset trend threshold The ratio when The system detects abnormal trends in parameters; when parameter data exceeds preset threshold values ​​or is determined to be abnormal by the aforementioned trend analysis model, the system immediately triggers a local alert and generates emergency control commands, while simultaneously synchronizing the abnormal data and critical operational data of the data center to the cloud management platform. This enables rapid response and handling of anomalies.

[0028] Before fusing the original multidimensional operating parameters, the edge computing layer sequentially performs preprocessing operations such as data denoising, outlier removal, and data standardization. Data denoising eliminates random noise collected by the sensors, outlier removal eliminates invalid data caused by sensor malfunctions or external interference, and data standardization maps each parameter to the [0, 1] interval, providing a high-quality data foundation for subsequent fusion, prediction, and algorithm operation.

[0029] The sensing layer includes temperature sensors, humidity sensors, pressure sensors, flow sensors, and equipment status sensors. All sensors connect to the edge nodes of the edge computing layer via wireless communication and support hot-swapping for easy maintenance and expansion. Temperature and humidity sensors are evenly deployed throughout the data center, pressure and flow sensors are deployed at key nodes in the hot and cold piping, and equipment status sensors are deployed on core equipment such as chillers, heaters, fans, and pumps to collect data on operating power, runtime, and fault signals. This enables real-time monitoring of equipment status.

[0030] The edge computing layer and cloud management platform employ a breakpoint resume mechanism. When the network is interrupted, the edge computing layer independently performs local control of the data center. Once the network is restored, it automatically synchronizes the data center's operating data, system control records, and anomaly detection records to the cloud management platform, enabling remote monitoring, historical data tracing, and anomaly source identification of the data center from the cloud. This provides data support for the long-term management and optimization of the data center.

[0031] The execution layer includes a refrigeration execution unit, a heating execution unit, a ventilation execution unit, and a valve execution unit. The refrigeration execution unit is connected to the chiller and chilled water pump, while the heating execution unit is connected to the heater and hot water pump. It controls the heating and cooling loads by adjusting the start / stop status and operating power of the corresponding equipment. The ventilation execution unit is connected to the fan, adjusting the fan speed to regulate the ventilation volume and adapt to the temperature and humidity control requirements of the computer room environment. The valve execution unit is connected to various valves in the hot and cold pipelines, controlling the flow rate of the pipeline medium by adjusting the valve opening, thus achieving precise control of pipeline parameters. The refrigeration and heating execution units achieve regulation by adjusting the start / stop status and operating power of the corresponding equipment; the ventilation execution unit regulates the ventilation volume by adjusting the fan speed; and the valve execution unit controls the flow rate of the pipeline medium by adjusting the valve opening. Each execution unit completes coordinated control operations based on the precise operating control commands output by the PID execution unit.

[0032] Each execution unit is equipped with a feedback module. The feedback module collects the operating status and changes in computer room parameters after the equipment performs control operations and generates feedback data, which is transmitted to the collaborative control layer to provide data support for the system's adaptive adjustment.

[0033] The present invention will be further described in detail below with reference to specific embodiments. This embodiment takes the intelligent control of a commercial cold and hot room as an example to illustrate the implementation process and effect of the system. Example 1

[0034] System Deployment and Initialization: 1. Sensing layer deployment: 12 wireless temperature and humidity sensors are evenly deployed in various environmental areas of the cold and hot computer room. 8 wireless pressure / flow sensors are deployed at the inlet and outlet of the cold and hot water pipes and in key branches. 16 equipment status sensors are deployed on core equipment such as chillers, heaters, fans, and water pumps. All sensors support hot-swapping and are connected to the edge nodes of the edge computing layer via LoRa wireless communication. 2. Edge Computing Layer Configuration: Deploy 3 edge computing nodes, equipped with LSTM neural network operation units, DQN algorithm operation units, PID control units, and anomaly detection modules. Configure the sliding window length m=8 acquisition cycles (acquisition cycle is 5 minutes), and preset the trend threshold. =0.05.

[0035] 3. Parameter initialization: Set the initial proportional coefficient of the PID execution unit. =2.5, setting the weights of the DQN multi-objective reward function. =0.4, = value 0.4, =0.2, enter the nominal accuracy error value of each sensor, initialize the sensor fusion weight, and set the preset thresholds for computer room environmental parameters (temperature and humidity), pipeline parameters (pressure / flow), and equipment operating parameters; Cloud-to-edge connectivity: The edge computing layer establishes communication with the data center cloud management platform, configures a breakpoint resume protocol, and sets the data synchronization frequency to 10 minutes / time. Example 2

[0036] System operation and control process 1. Parameter acquisition and preprocessing: The sensors in the sensing layer acquire the original multi-dimensional operating parameters (P, Q, W, T, H) of the computer room at a 5-minute interval and transmit them to the edge computing layer; the edge computing layer performs preprocessing operations in sequence, including data denoising (using median filtering), outlier removal (using the 3σ principle), and data standardization (mapping to the [0,1] interval); 2. The edge computing layer uses a sensor accuracy weighted fusion algorithm to fuse preprocessed parameters of the same type, and calculates the fused multidimensional operating parameter sequence State=[P,Q,W,T,H]; 3. Load prediction: The LSTM neural network computing unit is trained and outputs the predicted hot and cold load values ​​of the data center for the next 1.5 hours, using the fused parameter sequence of the past hour as input. 4. PID Correction and Feedforward Adjustment: Calculate the PID correction coefficient A based on the load forecast. If the predicted load value is 1.2 times the current load value, then A = 1 + (1.2 - 1) / 1 = 1.2. Increasing the proportional coefficient accelerates the adjustment speed, while using the predicted load value as a feedforward signal to raise the PID control target threshold in advance.

[0037] 5. DQN Algorithm Optimization: Based on the average energy consumption of traditional single PID control, calculate the current data center energy consumption, parameter compliance, and equipment stability to obtain... , , The total reward is obtained through a multi-objective weighted reward function, which guides the DQN algorithm to optimize the output of device operating parameters.

[0038] 6. The DQN algorithm optimization results are combined with the corrected PID control to output precise operation control commands to the execution layer. The refrigeration execution unit increases the operating power of the refrigeration unit by 10%, the valve execution unit increases the opening of the cold water pipeline valve by 15%, and the ventilation execution unit adjusts the fan speed to 1200 r / min to achieve coordinated control.

[0039] 7. Feedback and Adaptive Adjustment: The execution layer feedback module collects the equipment operating status and changes in computer room parameters, and transmits them to the collaborative control layer; the collaborative control layer calculates the sensor fusion weight update, increases the weight of temperature sensors with small errors by 5%, and iteratively optimizes the learning rate of the LSTM neural network and the exploration probability of the DQN algorithm; the edge computing layer optimizes the local control logic based on feedback, and adjusts the step size of the chiller power adjustment from 5% to 3% to improve the control accuracy. Example 3

[0040] Anomaly Detection and Emergency Response When a chiller in the computer room experiences a slight overload, the equipment status sensor detects that its operating power exceeds a preset threshold. Simultaneously, the edge computing layer performs trend analysis on the equipment's power parameters, and the slope of the first-order linear fit of the power parameters within the sliding window is k=0.6. The system immediately triggers a local audible and visual warning and generates an emergency control command: the refrigeration unit reduces the operating power of the refrigeration unit by 20% and starts the backup refrigeration unit to prevent further overload of the equipment. At the same time, the abnormal data (overload time, power value, trend curve) and the key operating data of the data center are synchronized to the cloud management platform. The operation and maintenance personnel can trace the source of the anomaly through the cloud, determine that the overload is caused by the blockage of the chiller pipeline, and carry out operation and maintenance in a timely manner. Example 4

[0041] Independent edge operation during network outages When the data center network is interrupted, the edge computing layer immediately starts the local independent control mode, stops data synchronization with the cloud, and continues to complete the entire process of parameter collection, fusion, load prediction, PID correction, DQN optimization and execution control. During the network outage, a total of 28 records of operational data, control logs, and anomaly detection logs were recorded in the local storage room of the edge computing layer. After the network was restored, the system automatically triggered the breakpoint resume mechanism, synchronizing all stored data to the cloud management platform. The cloud enabled historical data tracing and operational status analysis, and no data center parameters exceeded the standard or equipment malfunctions were found.

[0042] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing, characterized in that: Intelligent control of multiple parameters in the data center is achieved through the PID correction coefficient calculation formula and the DQN multi-objective weighted reward function based on traditional PID energy consumption benchmark, including: A formula for calculating the PID correction coefficient is constructed, and the PID proportional coefficient is dynamically adjusted based on the predicted values ​​of the cooling and heating loads of the cooling and heating room. The formula is as follows: =1 + (Predicted load - Current load) / Current load ;in, For PID correction coefficients, This is the initial proportional coefficient of the PID execution unit. This is the corrected PID proportional coefficient; when When >1, it increases proportionally. To speed up the adjustment process, when When it is less than 1, it decreases proportionally. To avoid over-adjustment, the predicted cold and hot load values ​​are used as feedforward signals for the PID execution unit to adjust the PID control target threshold in advance. Construct a DQN multi-objective weighted reward function with traditional single PID control as the energy consumption benchmark, the expression of which is: ,and ;in, For energy consumption weight, For parameter compliance weights, To ensure stable weighting of the equipment, This is an energy consumption reward item, calculated as follows: =1-(Current data center energy consumption / Baseline energy consumption), where the baseline energy consumption is the average energy consumption of the cold and hot data center under the traditional single PID control mode; Rewards for meeting parameter targets. This is a reward for equipment stability. The reward function guides the DQN algorithm to discover the correlation patterns of data center parameters, outputs the optimization direction of equipment operating parameters, and combines it with the PID control after being corrected by the above formula to output precise operating control commands for each execution device in the cold and hot data centers, so as to achieve multi-objective coordinated regulation of data center energy consumption, environmental parameter compliance, and stable equipment operation.

2. The multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing as described in claim 1, characterized in that, The value is 0.

4. The value is 0.

4. The value is 0.2; The value selection rule is: when the computer room environmental parameters and pipeline operation parameters are all within the preset threshold range. =1, otherwise =0.1; the The value selection rule is: when the core equipment in the computer room is not overloaded or faulty. =1, when the core equipment is slightly overloaded =0.5, when the core equipment fails. =0.

3. The multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing according to claim 2, characterized in that, The predicted value of the heating and cooling load is output by the LSTM neural network operation unit. The LSTM neural network operation unit takes the multi-dimensional operating parameter sequence of the heating and cooling room after fusion processing in the past hour as input and outputs the predicted value of the heating and cooling load of the room in the next 1-2 hours. The multidimensional operating parameters include pipeline medium pressure P, pipeline medium flow rate Q, core equipment operating power W, computer room ambient temperature T, and computer room ambient humidity H. Each parameter is standardized and mapped to the [0,1] interval. The multidimensional operating parameters serve as the state space State of the DQN algorithm, i.e., State=[P,Q,W,T,H].

4. The multi-parameter collaborative intelligent control system for cold and hot computer rooms based on edge computing according to claim 3, characterized in that, It also includes a perception layer and an edge computing layer. The perception layer collects raw multi-dimensional operating parameters of the computer room and transmits them to the edge computing layer. The edge computing layer uses a sensor accuracy weighted fusion algorithm to fuse the raw multi-dimensional operating parameters. The weighting calculation process of this sensor accuracy weighted fusion algorithm is as follows: based on the nominal accuracy error value of each type of sensor, the accuracy weight coefficient of each type of sensor is calculated. ,in, For the first The nominal accuracy error value of the sensor class; the accuracy weight coefficients of all single-class sensors are normalized to obtain the normalized weights. , The number of similar sensors participating in the fusion; based on The raw parameters collected by multiple sensors of the same type are weighted and summed to obtain the fused value of a single type of parameter.

5. The multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing according to claim 4, characterized in that, It also includes an execution layer, which is communicatively connected to the PID execution unit and each execution unit is equipped with a feedback module. The feedback module collects the operating status and changes in the machine room parameters after the equipment performs the control operation and generates feedback data, which is then transmitted to the collaborative control layer. The system constructs a triple adaptive mechanism. The collaborative control layer calculates the normalized weight update amount of the sensor accuracy weighted fusion algorithm based on feedback data, thereby realizing the adaptive adjustment of the sensor fusion weights. The specific update algorithm is as follows: ,in, For the updated number Sensor-like normalized weights, For the previous version Sensor-like normalized weights, For the first The difference between the absolute value of the error between the parameters acquired by the sensor class and the fused parameters and the average error of all sensors. For the first The absolute value of the error between the parameters acquired by the sensor and the fused parameters. This represents the average error of sensors of the same type.

6. The multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing according to claim 5, characterized in that, The triple adaptive mechanism also includes: the collaborative control layer iteratively optimizes the training parameters of the LSTM neural network operation unit and the reward function weights of the DQN algorithm based on the feedback data, so as to realize the dual adaptive adjustment of algorithm parameters and sensor fusion weights; at the same time, the edge computing layer optimizes the local collaborative decision-making logic in combination with the feedback data, forming a triple adaptive control system of "adaptive algorithm parameters, adaptive sensor fusion weights, and adaptive local decision-making logic".

7. The multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing according to claim 6, characterized in that, The edge computing layer performs dual anomaly detection on the preprocessed parameters using thresholding and trend analysis. The trend analysis employs a mathematical model combining first-order linear fitting and a sliding window, specifically implemented by setting the sliding window length to... , For 5-10 consecutive acquisition cycles, the parameter sequence within the window... Perform a first-order linear fit, the fitting formula is as follows: ,in, The slope is the fitted slope. The intercept is used to calculate the slope. The absolute value and the preset trend threshold The ratio when When parameters show abnormal change trends, a local early warning is immediately triggered and an emergency control command is generated when the parameter data exceeds the preset value threshold or is determined to be an abnormal change trend by the above trend analysis model. At the same time, the abnormal data and the key operating data of the computer room are synchronized to the cloud management platform.

8. The multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing according to claim 7, characterized in that, Before fusing the original multidimensional operating parameters, the edge computing layer sequentially performs preprocessing operations such as data denoising, outlier removal, and data standardization. The sensing layer includes temperature sensors, humidity sensors, pressure sensors, flow sensors, and equipment status sensors. All sensors are connected to the edge nodes of the edge computing layer via wireless communication and support hot-swapping. Temperature and humidity sensors are evenly deployed in various areas of the computer room, pressure and flow sensors are deployed at key nodes of the hot and cold pipelines, and equipment status sensors are deployed on core equipment such as chillers, heaters, fans, and water pumps to collect the operating power, operating time, and fault signals of the core equipment.

9. The multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing according to claim 8, characterized in that, The edge computing layer and the cloud management platform adopt a breakpoint resume mechanism. When the network is interrupted, the edge computing layer independently completes the local control of the data center. When the network is restored, the data center operation data, system control records and anomaly detection records are automatically synchronized to the cloud management platform, realizing remote monitoring of the data center, historical data tracing and anomaly tracing by the cloud.

10. The multi-parameter collaborative intelligent control system for cold and hot data centers based on edge computing according to claim 9, characterized in that, The execution layer includes a refrigeration execution unit, a heating execution unit, a ventilation execution unit, and a valve execution unit; the refrigeration execution unit is connected to the chiller and the cold water pump, the heating execution unit is connected to the heater and the hot water pump, the ventilation execution unit is connected to the fan, and the valve execution unit is connected to various valves in the cold and hot pipelines; The cooling and heating actuators regulate the operation by adjusting the start / stop status and operating power of the corresponding equipment. The ventilation actuators adjust the ventilation volume by adjusting the fan speed. The valve actuators control the flow rate of the medium in the pipeline by adjusting the valve opening. Each actuator completes coordinated regulation operation based on the precise operation control commands output by the PID actuator.

Images