An Energy Efficiency Optimization System and Method for IDC Intelligent Computing Centers Based on Intelligent Algorithms

Abstract

This invention discloses an energy efficiency optimization system and method for IDC intelligent computing centers based on intelligent algorithms. This invention relates to the field of computer system resource collaborative management technology, and includes a swarm intelligence decision-making central module, an ant colony optimization LSTM temperature prediction module, an ant colony collaborative task scheduling module, a path optimization cooling control module, and a fault energy consumption linkage sensing module. The energy efficiency optimization system and method for IDC intelligent computing centers described in this invention uses a swarm intelligence decision-making central module as its core, integrating ant colony algorithms throughout the entire process of temperature prediction, task scheduling, and cooling control. This breaks through the limitations of local optimization by a single algorithm in traditional solutions. It integrates data from various modules through a pheromone transmission mechanism, achieving global strategy coordination for prediction, scheduling, and cooling. This avoids decision biases caused by data fragmentation in independent algorithms, improving the overall integrity and coordination of energy efficiency management from the top-level design level of computer system resource allocation.

Description

Technical Field

[0001] This invention relates to the field of collaborative management technology for computer system resources, and in particular to an integrated method for optimizing CPU / GPU computing power allocation, temperature status monitoring, and energy efficiency control within an IDC intelligent computing center using swarm intelligence algorithms. Background Technology

[0002] With the explosive growth of the digital economy, IDC intelligent computing centers, as the core of computing infrastructure, have seen an exponential increase in scale and power density. However, the coordinated optimization of computing power, energy efficiency, and stability of intelligent computing centers has become a core technological bottleneck. Existing energy efficiency management solutions at the computer system level have the following fundamental defects: Existing intelligent computing centers rely heavily on single deep learning models such as LSTM and GRU for temperature prediction. However, their core parameters are mostly set manually or optimized locally through gradient descent, which cannot adapt to the dynamic coupling changes of rack load, ambient temperature, and fault type. At the same time, task scheduling often adopts static rule chains or simple federated learning frameworks, without incorporating global energy efficiency goals, which leads to high-load tasks being concentrated in adjacent racks.

[0003] Existing fault handling solutions mostly focus on hardware repair, failing to integrate faults, computing tasks, and energy efficiency adjustments into a unified decision-making loop. When a server experiences a hard drive failure, traditional systems only trigger an alarm without actively migrating the computing tasks on that rack. This results in the failed rack operating at low load during maintenance, while other racks become overloaded due to the additional tasks, leading to a significant loss of overall system computing power. Simultaneously, the cooling system fails to adjust cooling capacity distribution according to load changes after the fault, causing overcooling and increased air conditioning energy consumption.

[0004] Therefore, it is necessary to propose an energy efficiency optimization system and method for IDC intelligent computing centers based on intelligent algorithms to solve the above problems. Summary of the Invention

[0005] The main objective of this invention is to provide an energy efficiency optimization system and method for IDC intelligent computing centers based on intelligent algorithms, which can effectively solve the problems in the background technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An intelligent computing center energy efficiency optimization system based on intelligent algorithms includes a swarm intelligence decision-making central module, an ant colony optimization LSTM temperature prediction module, an ant colony collaborative task scheduling module, a path optimization cooling control module, and a fault energy consumption linkage sensing module. The swarm intelligence decision-making central module integrates data from the other modules based on the ant colony algorithm, achieves global optimization through pheromone updates, and outputs temperature prediction parameter correction, task scheduling, and cooling control commands. The ant colony-optimized LSTM temperature prediction module optimizes LSTM parameters through the ant colony algorithm to accurately predict the peak temperature and diffusion rate when maintenance is completed. The ant colony collaborative task scheduling module matches tasks with low-heat, low-load, and low-power servers through ant colony optimization, selects nodes according to the fitness function, dynamically migrates tasks on high-temperature nodes, and optimizes scheduling through pheromone updates. The path optimization cooling control module is used to optimize the cooling delivery path, calculate the flow rate of a single cabinet according to the formula, and perform three-stage control: advance regulation, dynamic balance and emergency intervention, and supply cooling on demand to match the task and temperature. The fault energy consumption linkage sensing module is used to collect data and transmit it to the swarm intelligence decision-making center module after preprocessing. The output of the swarm intelligence decision-making central module is connected to the parameter correction interface of the ant colony optimization LSTM temperature prediction module, the instruction input interface of the ant colony collaborative task scheduling module, and the flow setting interface of the path optimization cooling control module; the output of the fault energy consumption linkage sensing module is connected to the decision data input interface of the swarm intelligence decision-making central module.

[0007] Preferably, the swarm intelligence decision-making central module is used to abstract the system decision-making objective into an ant colony optimization problem, specifically including candidate decision schemes, paths, and pheromone concentrations. That is, each ant represents a set of candidate decision schemes, including temperature prediction parameters, task allocation paths, and cold energy delivery paths; paths correspond to strategy combinations of different modules; and pheromone concentrations correspond to the superiority or inferiority of the schemes.

[0008] Preferably, the ant colony-optimized LSTM temperature prediction module includes a parameter optimization objective function, the formula of which is: ; in To comprehensively optimize the target value; Weights for prediction errors; This is the temperature prediction error; To calculate the cost weights; Calculate the overhead for the LSTM model.

[0009] Preferably, the ant colony collaborative task scheduling module includes a task scheduling fitness function, used to define the fitness of ants selecting server nodes as the optimal value comprehensively considering load, temperature, and energy consumption, as follows: ; in This is the fitness value; Weighted by load rate; This refers to the load rate of the server's CPU / GPU. Temperature weighting; The current temperature of the server; This refers to the server's safe temperature threshold. Power consumption weight; Real-time power consumption of the server; This refers to the server's rated power consumption. .

[0010] Preferably, the ant colony collaborative task scheduling module includes a problem modeling submodule, a node selection submodule, a pheromone update submodule, and a task migration mechanism submodule. The problem modeling submodule is used to abstract the server cluster of the intelligent computing center into a node graph, where each node is a server and the edges between nodes are task transmission paths. Each ant represents a task to be scheduled, and the task allocation process is the process of ants finding the optimal node with the highest fitness value. Node selection submodule: Ants select nodes based on their fitness values. With pheromone concentration Select a node, select a probability formula: ; in , representing the importance of pheromones; , representing the importance of fitness; To remove Other optional nodes; Pheromone Update Submodule: After the task is completed, if the node's actual load rate is ≤70%, temperature is ≤server safe temperature threshold -2℃, and power consumption is ≤80% of the rated value, then the pheromone of that node is enhanced. Otherwise, pheromones will be released. ; Task migration mechanism submodule: When the temperature of a server is greater than or equal to the server's safe temperature threshold plus 1℃, the swarm intelligence decision-making center module triggers the ant colony to re-optimize and schedule the tasks on the server to the nodes with the top 20% fitness at the current moment. Task scheduling rules include: Priority scheduling rule: GPU-intensive tasks are assigned to nodes with GPU utilization ≤60% and temperature ≤28℃; Load balancing rule: Number of tasks per node ≤ Number of CPU cores × 2; Fault avoidance rule: Do not assign high-load tasks to the three racks surrounding the faulty rack.

[0011] Preferably, the high-load task is a task with CPU utilization ≥80% and GPU memory usage ≥90%.

[0012] Preferably, the path optimization cooling control module is used to calculate the liquid flow rate of the cold plate in a single cabinet, using the following formula: ; in For the first Cold plate liquid flow rate of each cabinet; This represents the total flow rate of the cooling system. For the first Pheromon concentration along the cooling transport path of each cabinet; The sum of pheromone concentrations along all cooling load delivery paths of the server racks; For flow coefficient; For the first Real-time power consumption of each rack.

[0013] Preferably, the path optimization cooling control module further includes: The path modeling submodule abstracts the cooling system's pipe network into a path graph, where nodes are the cold plate interfaces of the cabinets and edges are pipe segments; each ant represents a cooling capacity delivery path, and the path cost is the combined value of pipe pressure loss and delivery distance. Path selection submodule: Ants select paths based on pheromone concentration. With heuristic functions Choose a path; Pheromone Update Submodule: Enhances pheromones along the optimal path with pressure loss ≤0.1MPa and delivery distance ≤5m. For pathways with pressure loss ≥0.3MPa, pheromones are volatile. .

[0014] Preferably, the fault energy consumption linkage sensing module includes a data acquisition submodule, a data preprocessing module, and an energy consumption feedback submodule, wherein: The data acquisition submodule is used for fault diagnosis data collection: each server is equipped with a fault sensor to monitor hard drive, power supply, and motherboard faults, with a response time of ≤0.1s. The fault type and estimated maintenance duration are transmitted via RS485 bus. Energy consumption monitoring data collection: each rack is equipped with a smart meter to collect total power consumption and single-phase power consumption in real time. Temperature and humidity monitoring: infrared temperature and humidity sensors are installed inside the rack, sampling once every 10s. Load monitoring: server CPU / GPU load rate and memory usage are read via SNMP protocol, with a sampling interval of 5s. Flow monitoring: cold plate flow is collected in real time via an electromagnetic flow meter. The data preprocessing module is used to perform outlier removal, data normalization, and data fusion on the data collected by the data acquisition submodule.

[0015] An energy efficiency optimization method for IDC intelligent computing centers based on intelligent algorithms includes the following steps: S1: Fault perception and data acquisition. Data is collected through the data acquisition submodule. The collected data is preprocessed and then transmitted to the swarm intelligence decision-making center module. S2: Ant colony decision-making, specifically: Initialization: Set the number of ants to 50 and the initial pheromone concentration. pheromone volatility coefficient ; Solution generation: Each ant selects a decision based on pheromone concentration and a heuristic function. Heuristic function: ; in The estimated energy consumption of the plan, For temperature prediction error, Delayed task scheduling; Pheromone Update: The optimal solution, i.e., the solution with the lowest energy consumption, smallest prediction error, and shortest scheduling delay, is enhanced with pheromone. The update formula is: ; in ; The concentration of pheromones is used as the basis for determining the degradation scheme, thereby avoiding repeated selection. Convergence judgment: When the overall cost fluctuation of the optimal solution after 10 consecutive iterations is ≤0.5%, the final decision instruction is output; S3: Ant colony optimization of LSTM parameters, specifically: Parameter encoding: Each ant represents a set of LSTM parameter combinations: The number of hidden layer nodes Learning rate: , ; Fitness calculation: Input each set of parameters into the LSTM model and calculate the objective function. fitness The higher the fitness, the better the parameters; Pheromones are updated: Pheromones are enhanced for the parameter combinations corresponding to the top 10% of ants in fitness, and the updated formula is the same as that of the swarm intelligence decision-making center module to ensure that the ant colony prioritizes searching for high-quality parameter areas. Optimal parameter output: After 50 iterations, output the parameter combination with the highest fitness to construct the final ant colony-LSTM fusion prediction model; Based on this ant colony optimization LSTM temperature prediction module, the following inputs are required: historical temperature sequence of the server rack; fault type: hard drive failure, power supply failure, motherboard failure; estimated maintenance time; real-time power consumption of the server; rack spacing; ambient humidity; current cooling system flow rate. S4: Ant colony task scheduling process, specifically: The server cluster of the intelligent computing center is abstracted as a node graph, where each node is a server and the edges between nodes are task transmission paths; each ant represents a task to be scheduled. Ants determine the fitness value of a node. With pheromone concentration Select a node and calculate the probability. ; After the task is completed, if the actual load rate of the node is ≤70%, the temperature is ≤2℃ below the server's safe temperature threshold, and the power consumption is ≤80% of the rated value, then the pheromone of the node is enhanced; otherwise, the pheromone is released. When the temperature of a server is greater than or equal to the server's safe temperature threshold plus 1°C, the tasks on that server are scheduled to the top 20% of nodes in terms of fitness through the swarm intelligence decision-making central module. S5: Three-stage cooling control, specifically: Advanced control phase: Based on the highest cabinet temperature and temperature diffusion rate at the time of maintenance completion output by the ant colony-optimized LSTM temperature prediction module, cooling capacity is delivered 20 minutes in advance via the optimal path, with the flow rate adjusted according to the formula: ; Where 1.2 is the precooling coefficient; Dynamic balancing phase: Based on task scheduling results, if the server load rate increases by 10%, traffic will be increased by 8% through ant colony optimization to ensure... Positively correlated with load; Emergency Intervention Phase: When the highest temperature of the server rack is ≥ server safety temperature threshold + 3℃ after maintenance is completed, a 30% increase in total traffic is triggered. Simultaneously, cooling is delivered to the hotspot rack via the shortest path optimized by ant colony optimization. Traffic is adjusted according to the following formula: ; The value of 1.5 represents the urgency factor, which is used to reduce the load in conjunction with task migration.

[0016] Compared with existing technologies, this invention provides an energy efficiency optimization system and method for IDC intelligent computing centers based on intelligent algorithms, which has the following beneficial effects: This intelligent algorithm-based energy efficiency optimization system and method for IDC intelligent computing centers, with a swarm intelligence decision-making center at its core, integrates ant colony optimization throughout the entire process of temperature prediction, task scheduling, and cooling control, breaking through the limitations of local optimization by a single algorithm in traditional solutions. The swarm intelligence decision-making center module integrates data from various modules through a pheromone transmission mechanism, achieving global strategy coordination for prediction, scheduling, and cooling. This avoids decision-making biases caused by data fragmentation in independent algorithms, improving the overall integrity and coordination of energy efficiency management from the top-level design level of computer system resource allocation.

[0017] This intelligent algorithm-based energy efficiency optimization system and method for IDC intelligent computing centers overcomes the limitations of manual settings or single optimization methods by optimizing the LSTM temperature prediction module through ant colony optimization. It incorporates multi-dimensional data such as fault type, operational status, and environmental parameters into the prediction model, enabling it to not only accurately output temperature change trends but also predict temperature diffusion characteristics. This data processing approach based on swarm intelligence makes temperature prediction more aligned with the dynamic operating scenarios of intelligent computing centers, providing a reliable basis for subsequent resource scheduling and cooling control, and significantly improving the accuracy of computer system status perception.

[0018] This intelligent algorithm-based energy efficiency optimization system and method for IDC intelligent computing centers utilizes the path optimization characteristics of the ant colony collaborative task scheduling module. It incorporates server load, temperature status, and energy consumption levels into a unified fitness evaluation framework to achieve dynamic matching of tasks with optimal server nodes. Compared to traditional static scheduling strategies, this system can adjust task allocation paths based on real-time operational data, preventing high-load tasks from concentrating on hot servers. Furthermore, through collaboration with a temperature prediction module, it proactively mitigates potential thermal risks, effectively improving the utilization rate and operational stability of computer system resources.

[0019] This intelligent algorithm-based energy efficiency optimization system and method for IDC (Internet Data Center) computing centers utilizes an ant colony optimization cooling control module to optimize the cooling energy delivery path. It dynamically adjusts the cooling energy allocation strategy based on real-time task allocation results and temperature data to achieve on-demand cooling. This design, which deeply integrates cooling control into the computer system's resource scheduling framework, completely changes the traditional disconnect between cooling systems and computing loads. It avoids energy waste caused by overcooling and precisely suppresses localized overheating, reducing the overall energy consumption of the intelligent computing center from a resource coordination perspective.

[0020] This intelligent algorithm-based energy efficiency optimization system and method for IDC intelligent computing centers addresses unforeseen circumstances such as hardware failures. The system uses a fault energy consumption linkage sensing module to collect data in real time, and a swarm intelligence decision-making central module drives an ant colony algorithm to rapidly reorganize task scheduling strategies and cooling control schemes, achieving coordinated responses from all modules. This swarm intelligence-based fault handling mode eliminates the lag in independent module responses found in traditional solutions, significantly improving the flexibility and reliability of the computer system in responding to emergencies and ensuring the continuous and stable operation of the intelligent computing center. Attached Figure Description

[0021] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0023] Example 1: like Figure 1 As shown, an IDC intelligent computing center energy efficiency optimization system based on intelligent algorithms includes a swarm intelligence decision-making central module, an ant colony optimization LSTM temperature prediction module, an ant colony collaborative task scheduling module, a path optimization cooling control module, and a fault energy consumption linkage sensing module. The swarm intelligence decision-making central module integrates data from the other modules based on the ant colony algorithm, achieves global optimization through pheromone updates, and outputs temperature prediction parameter correction, task scheduling, and cooling control commands. The swarm intelligence decision-making hub module is used to abstract the system's decision-making objective into an ant colony optimization problem. Specifically, it includes candidate decision schemes, paths, and pheromone concentrations. Each ant represents a set of candidate decision schemes, including temperature prediction parameters, task allocation paths, and cold energy delivery paths. Paths correspond to the strategy combinations of different modules. The pheromone concentration corresponds to the quality of the scheme, which is determined by a combination of energy consumption, temperature control accuracy, and task response speed.

[0024] The ant colony-optimized LSTM temperature prediction module optimizes LSTM parameters through the ant colony algorithm to accurately predict the peak temperature and diffusion rate when maintenance is completed. The ant colony-optimized LSTM temperature prediction module includes a parameter optimization objective function, the formula of which is: ; in To optimize the overall target value, the smaller it is, the better; Weights for prediction errors; For temperature prediction error, ,in To predict temperature, This is the actual temperature; To calculate the cost weights; Calculate the overhead for the LSTM model. , This represents the number of hidden layer nodes. The number of nodes in the input layer. is the number of iterations; where , The study, through variance analysis of 100 sets of experiments, determined that by setting up fault simulation scenarios (hard drive failure / power supply failure / motherboard failure), and using mean squared error and inference latency as indicators, the results were verified through grid search. The Pareto optimality of prediction accuracy and computational cost is achieved.

[0025] The ant colony collaborative task scheduling module uses ant colony optimization to match tasks with low-heat, low-load, and low-power servers, selects nodes according to the fitness function, dynamically migrates tasks on high-temperature nodes, and optimizes scheduling through pheromone updates. The ant colony collaborative task scheduling module includes a task scheduling fitness function, which defines the fitness of ants selecting server nodes as the optimal value based on a combination of load, temperature, and energy consumption. The formula is as follows: ; in The fitness value ranges from 0 to 3. , , At that time, its theoretical upper limit is 3.0, and the lower limit approaches 0. The larger the value, the better the server node. The load factor is weighted at 0.4 in this case. This refers to the load rate of the server's CPU / GPU. The weight is based on temperature, and is 0.3 here. The current temperature of the server; The server's safe temperature threshold is calibrated using an infrared temperature sensor installed inside the server rack. The power consumption weight is 0.3 here; Real-time power consumption of the server; This refers to the server's rated power consumption, configured according to the server model. The ant colony collaborative task scheduling module includes a problem modeling submodule, a node selection submodule, a pheromone update submodule, and a task migration mechanism submodule. The problem modeling submodule is used to abstract the server cluster of the intelligent computing center into a node graph, where each node is a server and the edges between nodes are task transmission paths. Each ant represents a task to be scheduled, and the task allocation process is the process of ants finding the optimal node. Node selection submodule: Ants select nodes based on their fitness values. With pheromone concentration Select a node, select a probability formula: ; in , representing the importance of pheromones; , representing the importance of fitness; To remove Other optional nodes; Pheromone Update Submodule: After the task is completed, if the node's actual load rate is ≤70%, temperature is ≤server safe temperature threshold -2℃, and power consumption is ≤80% of the rated value, then the pheromone of that node is enhanced. Otherwise, pheromones will be released. ; Task migration mechanism submodule: When the temperature of a server is greater than or equal to the server's safe temperature threshold plus 1℃, the swarm intelligence decision-making center module triggers the ant colony to re-optimize and schedule the tasks on the server to the nodes with the top 20% fitness at the current moment. Task scheduling rules include: Priority scheduling rule: GPU-intensive tasks are assigned to nodes with GPU utilization ≤60% and temperature ≤28℃; Load balancing rule: Number of tasks per node ≤ number of CPU cores × 2, to avoid overload; Fault avoidance rules: Do not assign high-load tasks to the three racks surrounding the faulty rack to prevent a cascading increase in temperature; High-load tasks are those with CPU utilization ≥80% and GPU memory usage ≥90%.

[0026] The path optimization cooling control module is used to optimize the cooling delivery path, calculate the flow rate of a single cabinet according to the formula, and perform three-stage control: advance regulation, dynamic balance and emergency intervention, and supply cooling on demand to match the task and temperature. The path optimization cooling control module is used to calculate the liquid flow rate of the cold plate in a single cabinet. The formula is: ; in For the first The cold plate liquid flow rate of each cabinet, in L / min; This represents the total flow rate of the cooling system, expressed in L / min. For the first Pheromon concentration along the cooling transport path of each cabinet; The sum of pheromone concentrations along all cooling load delivery paths of the server racks; For flow coefficient, The coefficient is calibrated through a cold plate heat exchange experiment; For the first Real-time power consumption of each rack; The path optimization cooling control module also includes: The path modeling submodule abstracts the cooling system's pipe network into a path graph, where nodes are the cold plate interfaces of the cabinets and edges are pipe segments; each ant represents a cooling capacity delivery path, and the path cost is the combined value of pipe pressure loss and delivery distance. Path selection submodule: Ants select paths based on pheromone concentration. With heuristic functions Choose a path; Pheromone Update Submodule: Enhances pheromones along the optimal path with pressure loss ≤0.1MPa and delivery distance ≤5m. For pathways with pressure loss ≥0.3MPa, pheromones are volatile. .

[0027] The fault energy consumption linkage sensing module is used to collect data and transmit it to the swarm intelligence decision-making center module after preprocessing. The fault energy consumption linkage sensing module includes a data acquisition submodule, a data preprocessing module, and an energy consumption feedback submodule, among which: The data acquisition submodule is used for fault diagnosis data collection: each server is equipped with a fault sensor to monitor hard drive, power supply, and motherboard faults, with a response time of ≤0.1s. The fault type and estimated maintenance duration are transmitted via RS485 bus. Energy consumption monitoring data collection: each rack is equipped with a smart meter to collect total power consumption and single-phase power consumption in real time. Temperature and humidity monitoring: infrared temperature and humidity sensors are installed inside the rack, sampling once every 10s. Load monitoring: server CPU / GPU load rate and memory usage are read via SNMP protocol, with a sampling interval of 5s. Flow monitoring: cold plate flow is collected in real time via an electromagnetic flow meter. The data preprocessing module is used to perform outlier removal, data normalization, and data fusion on the data collected by the data acquisition submodule.

[0028] The output of the swarm intelligence decision-making central module is connected to the parameter correction interface of the ant colony optimization LSTM temperature prediction module, the command input interface of the ant colony collaborative task scheduling module, and the flow setting interface of the path optimization cooling control module. The output of the fault energy consumption linkage sensing module is connected to the decision data input interface of the swarm intelligence decision-making central module.

[0029] Example 2: An energy efficiency optimization method for IDC intelligent computing centers based on intelligent algorithms includes the following steps: S1: Fault perception and data acquisition. Data is collected through the data acquisition submodule. The collected data is preprocessed and then transmitted to the swarm intelligence decision-making center module. S2: Ant colony decision-making, specifically: Initialization: Set the number of ants to 50, the optimal population size verified experimentally, and the initial pheromone concentration. pheromone volatility coefficient This is used to balance global optimization and local convergence. Solution generation: Each ant selects a decision based on pheromone concentration and a heuristic function. Heuristic function: ; in The estimated energy consumption of the plan, For temperature prediction error, Delayed task scheduling; Pheromone Update: The optimal solution, i.e., the solution with the lowest energy consumption, smallest prediction error, and shortest scheduling delay, is enhanced with pheromone. The update formula is: ; in ; The pheromone concentration is used as a basis for determining the degradation scheme, which involves pheromone evaporation to avoid repeated selection. The overall cost is: ; in , , These are all weighting coefficients; Convergence judgment: When the overall cost fluctuation of the optimal solution after 10 consecutive iterations is ≤0.5%, the final decision instruction is output; S3: Ant colony optimization of LSTM parameters, specifically: Parameter encoding: Each ant represents a set of LSTM parameter combinations: The number of hidden layer nodes , where is an integer, and the learning rate is: Step size 0.001 , which is an integer; Fitness calculation: Input each set of parameters into the LSTM model and calculate the objective function. fitness The higher the fitness, the better the parameters; Pheromones are updated: Pheromones are enhanced for the parameter combinations corresponding to the top 10% of ants in fitness, and the updated formula is the same as that of the swarm intelligence decision-making center module to ensure that the ant colony prioritizes searching for high-quality parameter areas. Optimal parameter output: After 50 iterations, output the parameter combination with the highest fitness to construct the final ant colony-LSTM fusion prediction model.

[0030] Based on this ant colony optimization LSTM temperature prediction module, the input needs to be: the historical temperature sequence of the server rack. Sampling, a total of 60 time steps; Fault types: hard drive failure, power supply failure, motherboard failure; Estimated maintenance time; Real-time server power consumption; Rack spacing; Ambient humidity; Current cooling system traffic; S4: Ant colony task scheduling process, specifically: The server cluster of the intelligent computing center is abstracted as a node graph, where each node is a server and the edges between nodes are task transmission paths; each ant represents a task to be scheduled. Ants determine the fitness value of a node. With pheromone concentration Select a node and calculate the probability. ; After the task is completed, if the actual load rate of the node is ≤70%, the temperature is ≤2℃ below the server's safe temperature threshold, and the power consumption is ≤80% of the rated value, then the pheromone of the node is enhanced; otherwise, the pheromone is released. When the temperature of a server is greater than or equal to the server's safe temperature threshold plus 1°C, the tasks on that server are scheduled to the top 20% of nodes in terms of fitness through the swarm intelligence decision-making central module. S5: Three-stage cooling control, specifically: Advanced control phase: Based on the highest cabinet temperature and temperature diffusion rate at the time of maintenance completion output by the ant colony-optimized LSTM temperature prediction module, cooling capacity is delivered 20 minutes in advance via the optimal path, with the flow rate adjusted according to the formula: ; Where 1.2 is the precooling coefficient; Dynamic balancing phase: Based on task scheduling results, if the server load rate increases by 10%, traffic will be increased by 8% through ant colony optimization to ensure... Positively correlated with load; Emergency Intervention Phase: When the highest temperature of the server rack is ≥ server safety temperature threshold + 3℃ after maintenance is completed, a 30% increase in total traffic is triggered. Simultaneously, cooling is delivered to the hotspot rack via the shortest path optimized by ant colony optimization. Traffic is adjusted according to the following formula: ; The value of 1.5 represents the urgency factor, which is used to reduce the load in conjunction with task migration.

[0031] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An energy efficiency optimization system for an IDC intelligent computing center based on intelligent algorithms, comprising a swarm intelligence decision-making central module, an ant colony optimization LSTM temperature prediction module, an ant colony collaborative task scheduling module, a path optimization cooling control module, and a fault energy consumption linkage sensing module, characterized in that: The swarm intelligence decision-making central module integrates data from other modules based on the ant colony algorithm, achieves global optimization through pheromone updates, and outputs temperature prediction parameter correction, task scheduling, and cooling control commands. The ant colony-optimized LSTM temperature prediction module optimizes LSTM parameters through the ant colony algorithm to accurately predict the peak temperature and diffusion rate when maintenance is completed. The ant colony collaborative task scheduling module matches tasks with low-heat, low-load, and low-power servers through ant colony optimization, selects nodes according to the fitness function, dynamically migrates tasks on high-temperature nodes, and optimizes scheduling through pheromone updates. The path optimization cooling control module is used to optimize the cooling delivery path, calculate the flow rate of a single cabinet according to the formula, and perform three-stage control: advance regulation, dynamic balance and emergency intervention, and supply cooling on demand to match the task and temperature. The fault energy consumption linkage sensing module is used to collect data and transmit it to the swarm intelligence decision-making center module after preprocessing. The output of the swarm intelligence decision-making central module is connected to the parameter correction interface of the ant colony optimization LSTM temperature prediction module, the instruction input interface of the ant colony collaborative task scheduling module, and the flow setting interface of the path optimization cooling control module; the output of the fault energy consumption linkage sensing module is connected to the decision data input interface of the swarm intelligence decision-making central module. The ant colony collaborative task scheduling module includes a task scheduling fitness function, which defines the fitness of ants selecting server nodes as the optimal value based on a combination of load, temperature, and energy consumption. The formula is as follows: ； in This is the fitness value; Weighted by load rate; This refers to the load rate of the server's CPU / GPU. Temperature weighting; The current temperature of the server; This refers to the server's safe temperature threshold. Power consumption weight; Real-time power consumption of the server; This refers to the server's rated power consumption. ; The ant colony collaborative task scheduling module includes a problem modeling submodule, a node selection submodule, a pheromone update submodule, and a task migration mechanism submodule. The problem modeling submodule is used to abstract the server cluster of the intelligent computing center into a node graph, where each node is a server and the edges between nodes are task transmission paths. Each ant represents a task to be scheduled, and the task allocation process is the process of ants finding the optimal node with the highest fitness value. Node selection submodule: Ants select nodes based on their fitness values. With pheromone concentration Select a node, select a probability formula: ； in , representing the importance of pheromones; , representing the importance of fitness; To remove Other optional nodes; Pheromone Update Submodule: After the task is completed, if the node's actual load rate is ≤70%, temperature is ≤server safe temperature threshold -2℃, and power consumption is ≤80% of the rated value, then the pheromone of that node is enhanced. ; Otherwise, pheromones will be released: ; Task migration mechanism submodule: When the temperature of a server is greater than or equal to the server's safe temperature threshold plus 1℃, the swarm intelligence decision-making center module triggers the ant colony to re-optimize and schedule the tasks on the server to the nodes with the top 20% fitness at the current moment. Task scheduling rules include: Priority scheduling rule: GPU-intensive tasks are assigned to nodes with GPU utilization ≤60% and temperature ≤28℃; Load balancing rule: Number of tasks per node ≤ Number of CPU cores × 2; Fault avoidance rules: Do not assign high-load tasks to the three racks surrounding the faulty rack; The high-load tasks are those with CPU utilization ≥80% and GPU memory usage ≥90%. The path optimization cooling control module is used to calculate the liquid flow rate of the cold plate in a single cabinet, using the following formula: ； in For the first Cold plate liquid flow rate of each cabinet; This represents the total flow rate of the cooling system. For the first Pheromon concentration along the cooling transport path of each cabinet; The sum of pheromone concentrations along all cooling load delivery paths of the server racks; For flow coefficient; For the first Real-time power consumption of each rack; The path optimization cooling control module also includes: The path modeling submodule abstracts the cooling system's pipe network into a path graph, where nodes are the cold plate interfaces of the cabinets and edges are pipe segments; each ant represents a cooling capacity delivery path, and the path cost is the combined value of pipe pressure loss and delivery distance. Path selection submodule: Ants select paths based on pheromone concentration. With heuristic functions Choose a path; Pheromone Update Submodule: Enhances pheromones along the optimal path with pressure loss ≤0.1MPa and delivery distance ≤5m. For pathways with pressure loss ≥0.3MPa, pheromones are volatile. ; The fault energy consumption linkage sensing module includes a data acquisition submodule, a data preprocessing module, and an energy consumption feedback submodule, wherein: The data acquisition submodule is used for fault diagnosis data collection: each server is equipped with a fault sensor to monitor hard drive, power supply, and motherboard faults, with a response time of ≤0.1s. The fault type and estimated maintenance duration are transmitted via RS485 bus. Energy consumption monitoring data collection: each rack is equipped with a smart meter to collect total power consumption and single-phase power consumption in real time. Temperature and humidity monitoring: infrared temperature and humidity sensors are installed inside the rack, sampling once every 10s. Load monitoring: server CPU / GPU load rate and memory usage are read via SNMP protocol, with a sampling interval of 5s. Flow monitoring: cold plate flow is collected in real time via an electromagnetic flow meter. The data preprocessing module is used to perform outlier removal, data normalization, and data fusion on the data collected by the data acquisition submodule.

2. The energy efficiency optimization system for IDC intelligent computing centers based on intelligent algorithms according to claim 1, characterized in that: The swarm intelligence decision-making central module is used to abstract the system decision-making objective into an ant colony optimization problem, which specifically includes candidate decision schemes, paths, and pheromone concentrations. That is, each ant represents a set of candidate decision schemes, including temperature prediction parameters, task allocation paths, and cold energy delivery paths. Path: The strategy combination corresponding to different modules; pheromone concentration corresponds to the quality of the scheme.

3. The IDC intelligent computing center energy efficiency optimization system based on intelligent algorithms according to claim 1, characterized in that: The ant colony-optimized LSTM temperature prediction module includes a parameter optimization objective function, the formula of which is: ； in To comprehensively optimize the target value; Weights for prediction errors; This is the temperature prediction error; To calculate the cost weights; Calculate the overhead for the LSTM model.

4. A method for optimizing the energy efficiency of an IDC intelligent computing center based on intelligent algorithms, employing an IDC intelligent computing center energy efficiency optimization system based on intelligent algorithms as described in any one of claims 1-3, characterized in that: The following steps are included: S1: Fault perception and data acquisition. Data is collected through the data acquisition submodule. The collected data is preprocessed and then transmitted to the swarm intelligence decision-making center module. S2: Ant colony decision-making, specifically: Initialization: Set the number of ants to 50 and the initial pheromone concentration. pheromone volatility coefficient ; Solution generation: Each ant selects a decision based on pheromone concentration and a heuristic function. Heuristic function: ； in The estimated energy consumption of the plan, For temperature prediction error, Delayed task scheduling; Pheromone Update: The optimal solution, i.e., the solution with the lowest energy consumption, smallest prediction error, and shortest scheduling delay, is enhanced with pheromone. The update formula is: ； in ; The concentration of pheromones is used as the basis for determining the degradation scheme, thereby avoiding repeated selection. Convergence judgment: When the overall cost fluctuation of the optimal solution after 10 consecutive iterations is ≤0.5%, the final decision instruction is output; S3: Ant colony optimization of LSTM parameters, specifically: Parameter encoding: Each ant represents a set of LSTM parameter combinations: The number of hidden layer nodes Learning rate: , ; Fitness calculation: Input each set of parameters into the LSTM model and calculate the objective function. fitness The higher the fitness, the better the parameters; Pheromones are updated: Pheromones are enhanced for the parameter combinations corresponding to the top 10% of ants in fitness, and the updated formula is the same as that of the swarm intelligence decision-making center module to ensure that the ant colony prioritizes searching for high-quality parameter areas. Optimal parameter output: After 50 iterations, output the parameter combination with the highest fitness to construct the final ant colony-LSTM fusion prediction model; Based on this ant colony optimization LSTM temperature prediction module, the following inputs are required: historical temperature sequence of the server rack; fault type: hard drive failure, power supply failure, motherboard failure; estimated maintenance time; real-time power consumption of the server; rack spacing; ambient humidity; current cooling system flow rate. S4: Ant colony task scheduling process, specifically: The server cluster of the intelligent computing center is abstracted as a node graph, where each node is a server and the edges between nodes are task transmission paths; each ant represents a task to be scheduled. Ants determine the fitness value of a node. With pheromone concentration Select a node and calculate the probability. ; After the task is completed, if the actual load rate of the node is ≤70%, the temperature is ≤2℃ below the server's safe temperature threshold, and the power consumption is ≤80% of the rated value, then the pheromone of the node is enhanced; otherwise, the pheromone is released. When the temperature of a server is greater than or equal to the server's safe temperature threshold plus 1°C, the tasks on that server are scheduled to the top 20% of nodes in terms of fitness through the swarm intelligence decision-making central module. S5: Three-stage cooling control, specifically: Advanced control phase: Based on the highest cabinet temperature and temperature diffusion rate at the time of maintenance completion output by the ant colony-optimized LSTM temperature prediction module, cooling capacity is delivered 20 minutes in advance via the optimal path, with the flow rate adjusted according to the formula: ； Where 1.2 is the precooling coefficient; Dynamic balancing phase: Based on task scheduling results, if the server load rate increases by 10%, traffic will be increased by 8% through ant colony optimization to ensure... Positively correlated with load; Emergency Intervention Phase: When the highest temperature of the server rack is ≥ server safety temperature threshold + 3℃ after maintenance is completed, a 30% increase in total traffic is triggered. Simultaneously, cooling is delivered to the hotspot rack via the shortest path optimized by ant colony optimization. Traffic is adjusted according to the following formula: ； The value of 1.5 represents the urgency factor, which is used to reduce the load in conjunction with task migration.

Images