Energy-saving control system for remote intelligent hard turn-off of base station machine room

Abstract

The invention discloses an energy-saving control system for remote intelligent hard turn-off of a base station machine room, which relates to the technical field of energy conservation of communication base stations, and comprises a station group portrait module for acquiring and processing running states, energy consumption data, service flow and environmental parameters of a plurality of base stations, and analyzing service association, energy consumption complementary relationship and geographical proximity characteristics among the stations, establishing a station group association matrix, and generating a station group portrait for subsequent analysis and control. In the invention, the linkage arrangement module carries out strict semantic verification and safe arrangement on the turn-off plan, so that the issuing of a wrong instruction is effectively prevented, and the execution safety is improved; the real-time monitoring and automatic recovery are realized by the exception verification module, the potential risk is obviously reduced, the execution safety and reliability are ensured, the auditing display module provides comprehensive operation logs, energy efficiency data and reputation calculation, and the transparency, traceability and reliability of the energy-saving process are realized.

Description

Technical Field

[0001] This invention relates to the field of energy-saving technology for communication base stations, and in particular to an energy-saving control system for remote intelligent hard shutdown of base station equipment rooms. Background Technology

[0002] The system runs in a cloud or local server environment. The seven modules work together sequentially according to the data flow and control flow during system operation. The station group profiling module provides structural feature input, the load thermal assessment module outputs thermal safety boundaries, the life assessment module provides reliability constraints, the game decision module generates energy-saving strategies, the linkage orchestration module executes strategy verification and distribution, the anomaly verification module realizes operation monitoring and adaptive recovery, and the audit display module completes data storage and result display. The modules exchange data and provide status feedback through standard interfaces, forming a closed loop of intelligent energy-saving control that enables sustainable self-learning and parameter self-calibration.

[0003] Existing base station energy-saving control systems typically rely on preset simple thresholds or fixed schedules when performing shutdown operations. They lack the ability to deeply perceive and predict inter-base station service interactions, dynamic load changes, equipment lifespan impacts, and the thermal safety of the equipment room. Frequent power-on and power-off operations can negatively affect equipment lifespan, and existing systems often lack the assessment and prediction of equipment health status, failing to achieve an optimal balance between energy saving and equipment lifespan. In actual operation, even when shutdown operations are performed, there is a lack of effective real-time monitoring and anomaly handling mechanisms. In the event of unexpected situations (such as abnormal temperature increases or sharp declines in service quality), the system's recovery capabilities are insufficient, potentially leading to service interruptions or equipment damage. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an energy-saving control system for remote intelligent hard shutdown of base station equipment rooms.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an energy-saving control system for remote intelligent hard shutdown of base station equipment rooms, comprising: a station group profiling module for collecting and processing the operating status, energy consumption data, service traffic, and environmental parameters of multiple base stations, analyzing the service correlation, energy consumption complementarity, and geographical proximity characteristics between stations, establishing a station group correlation matrix, and generating a station group profile for subsequent analysis and control; a load thermal assessment module for performing load prediction and thermal safety analysis based on the station group profile and external environmental data, determining the temperature change trend, thermal safety level, and shutdown duration of the equipment room by integrating service data and meteorological parameters; a lifespan assessment module for calculating the equipment lifespan distribution and reliability range based on equipment operating time, number of thermal cycles, power fluctuations, and maintenance records, generating a remaining lifespan estimate and recovery interval; and a game theory decision-making module for determining the equipment lifespan distribution and reliability range based on the load thermal assessment module. The evaluation and lifespan assessment modules provide data to establish an optimization decision-making model among multiple base stations, generating shutdown and recovery plans that comply with thermal safety and lifespan constraints. The linkage orchestration module performs semantic verification and secure orchestration on the shutdown plans output by the game decision-making module, performs consistency checks and task scheduling based on the rule base, and generates securely executable control commands. The anomaly verification module monitors the system's operating status during the shutdown plan execution, identifies anomalies in parameters such as temperature, energy consumption, and service quality, and performs automatic recovery based on the anomaly level. The audit display module records system operation logs and energy efficiency data, performs encrypted evidence storage and reputation calculation, and generates audit reports and operational visualizations. All modules form a closed-loop operation through data and control channels, realizing an energy-saving control process based on data perception, predictive evaluation, optimization decision-making, secure execution, anomaly verification, and result auditing.

[0006] As a further description of the above technical solution: The site group profiling module includes a data access layer, a data processing layer, a relationship analysis layer, a structure generation layer, and an output layer. The data access layer is used to collect base station power, service, and environmental data through the communication interface and perform time synchronization. The data processing layer is used to clean outliers and fill in missing data. The relationship analysis layer is used to calculate the similarity and correlation of each site. The structure generation layer is used to establish a site group relationship model. The output layer is used to generate a site group profiling result file.

[0007] As a further description of the above technical solution: The load thermal assessment module includes a feature fusion layer, a prediction layer, a thermal analysis layer, and a security assessment layer. The feature fusion layer is used to integrate business data and external environmental parameters, the prediction layer is used to calculate future load trends, the thermal analysis layer is used to simulate temperature change processes, and the security assessment layer is used to determine the thermal safety level and shutdown time.

[0008] As a further description of the above technical solution: The life assessment module includes a feature extraction layer, a life calculation layer, a history correction layer, and an update layer. The feature extraction layer is used to extract life-related features from power fluctuations, thermal cycling, and maintenance information. The life calculation layer is used to extrapolate the life distribution based on multidimensional data. The history correction layer is used to correct estimation errors. The update layer is used to adjust the life prediction results when the operating data changes.

[0009] As a further description of the above technical solution: The game decision-making module includes a state modeling layer, an optimization calculation layer, and a strategy generation layer. The state modeling layer is used to describe the operating state of the base station group, the optimization calculation layer is used to calculate the coordinated shutdown scheme of each site, and the strategy generation layer is used to output energy-saving strategies that meet security and reliability constraints.

[0010] As a further description of the above technical solution: The linkage orchestration module includes a rule engine layer, a verification layer, a scheduling layer, and a feedback layer. The rule engine layer is used to store security constraints, the verification layer is used to detect the logical consistency between control commands, the scheduling layer is used to adjust the execution order according to the risk level, and the feedback layer is used to record the execution results and update the rule base.

[0011] As a further description of the above technical solution: The anomaly verification module includes a sampling layer, an analysis layer, an identification layer, and a recovery layer. The sampling layer is used to collect operating parameters in real time, the analysis layer is used to evaluate the stability of the operating state, the identification layer is used to detect the anomaly type and range, and the recovery layer is used to generate and execute adaptive recovery instructions.

[0012] As a further description of the above technical solution: The audit display module includes a log recording layer, an evidence storage layer, a reputation assessment layer, and a display layer. The log recording layer is used to store operation logs, the evidence storage layer is used for data encryption and tamper-proof storage, the reputation assessment layer is used to calculate the operation reputation score, and the display layer is used to generate audit reports and graphical display interfaces.

[0013] As a further description of the above technical solution: The system has a periodic update mechanism. When a new base station is put into operation or an existing base station is decommissioned, the system automatically updates the base station group relationship matrix and base station group profile data to ensure that the model is synchronized with the actual operating status.

[0014] As a further description of the above technical solution: The system runs in a cloud or local server environment. The seven modules work together sequentially according to the data flow and control flow during system operation. The station group profiling module provides structural feature input, the load thermal assessment module outputs thermal safety boundaries, the life assessment module provides reliability constraints, the game decision module generates energy-saving strategies, the linkage orchestration module executes strategy verification and distribution, the anomaly verification module realizes operation monitoring and adaptive recovery, and the audit display module completes data storage and result display. The modules exchange data and provide status feedback through standard interfaces, forming a closed loop of intelligent energy-saving control that enables sustainable self-learning and parameter self-calibration.

[0015] The present invention has the following beneficial effects: 1. In this invention, the base station profiling module performs in-depth analysis of the base station group, establishes the correlation between sites, and provides a global perspective for subsequent decision-making, avoiding the limitations of single-point independent control, thereby improving the level of intelligence. The load assessment module integrates business and meteorological data, which can accurately predict load trends and changes in data center temperature, improving the accuracy and security of energy-saving strategies and enhancing prediction capabilities. The lifespan assessment module introduces equipment lifespan management, incorporates the health status of equipment into shutdown decision considerations, effectively balances energy saving and equipment reliability, extends equipment lifespan, and thus protects equipment lifespan. The game theory decision module optimizes based on multi-dimensional constraints (thermal safety, lifespan, business traffic, etc.), and the generated shutdown and recovery plans have higher global optimality, ensuring service quality and system stability, making decision optimization more comprehensive.

[0016] 2. In this invention, the linkage orchestration module performs strict semantic verification and secure orchestration of the shutdown plan, effectively preventing the issuance of erroneous instructions and improving execution security. The anomaly verification module realizes real-time monitoring and automatic recovery, significantly reducing potential risks and ensuring execution security and reliability. The audit display module provides comprehensive operation logs, energy efficiency data, and reputation calculations, realizing transparency, traceability, and trustworthiness of the energy-saving process, thereby making management more transparent and reliable. The system operates in a closed loop through data and control channels, enabling continuous perception, evaluation, decision-making, and verification. It can also continuously learn and self-calibrate parameters based on operational data, continuously optimizing energy-saving strategies, improving system performance, and demonstrating adaptive and self-learning capabilities. The system can be deployed in the cloud or on a local server environment, and its modular design makes it easy to expand and maintain, adapting to base station networks of different sizes and needs, demonstrating flexibility and scalability. Attached Figure Description

[0017] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Reference Figure 1 This invention provides an embodiment of an energy-saving control system for remote intelligent hard shutdown of base station equipment rooms, comprising: a station group profiling module for collecting and processing the operating status, energy consumption data, service traffic, and environmental parameters of multiple base stations, analyzing the service correlation, energy consumption complementarity, and geographical proximity characteristics between stations, establishing a station group correlation matrix, and generating a station group profile for subsequent analysis and control; a load thermal assessment module for performing load prediction and thermal safety analysis based on the station group profile and external environmental data, determining the temperature change trend, thermal safety level, and shutdown duration of the equipment room by fusing service data and meteorological parameters; a lifespan assessment module for calculating the equipment lifespan distribution and reliability range based on equipment operating time, thermal cycle count, power fluctuation, and maintenance records, generating a remaining lifespan estimate and recovery interval; and a game theory decision-making module for using the load thermal assessment module and... The lifespan assessment module provides data to establish an optimization decision-making model among multiple base stations, generating shutdown and recovery plans that comply with thermal safety and lifespan constraints. The linkage orchestration module performs semantic verification and secure orchestration on the shutdown plans output by the game decision-making module, performs consistency checks and task scheduling based on the rule base, and generates securely executable control commands. The anomaly verification module monitors the system's operating status during the shutdown plan execution, identifies anomalies in parameters such as temperature, energy consumption, and service quality, and performs automatic recovery based on the anomaly level. The audit display module records system operation logs and energy efficiency data, performs encrypted evidence storage and reputation calculation, and generates audit reports and operational visualizations. All modules form a closed-loop operation through data and control channels, realizing an energy-saving control process based on data perception, predictive evaluation, optimization decision-making, secure execution, anomaly verification, and result auditing.

[0020] The station group profiling module comprises a data access layer, a data processing layer, a relationship analysis layer, a structure generation layer, and an output layer. The data access layer collects base station power, service, and environmental data through communication interfaces and performs time synchronization. The data processing layer cleans outliers and fills in missing data. The relationship analysis layer calculates the similarity and correlation between stations. The structure generation layer establishes a station group relationship model. The output layer generates the station group profiling result file. The station group profiling module is the system's perception layer, and its main function is to comprehensively collect and process the operating status, energy consumption data, service traffic, and environmental parameters of multiple base stations. This module establishes a refined station group correlation matrix by deeply analyzing the service correlations, energy consumption complementarity, and geographical proximity characteristics between stations, and generates station group profiling data for subsequent analysis and control. The station group profiling module includes a data access layer, a data processing layer, a relationship analysis layer, a structure generation layer, and an output layer. The data access layer is responsible for collecting raw data from the base station through various communication interfaces (such as SNMP, ModbusTCP, RESTfulAPI, etc.), including but not limited to base station power (transmit power, power consumption of each device), service data (number of users, traffic, drop rate, latency, and other KPIs), and environmental data (room temperature, humidity, external ambient temperature, solar radiation intensity, etc.). To ensure data accuracy and availability, the data access layer also performs strict time synchronization to eliminate inconsistencies in data collection timing. A unified timestamp for data from each base station can be achieved through the NTP protocol or GPS clock synchronization technology. After receiving the raw data, the data processing layer preprocesses it. This includes using statistical methods (such as the 3σ principle, IQR method) or machine learning algorithms (such as IsolationForest, One-ClassSVM) to detect and remove outliers or erroneous data in the collected data to improve data quality. For missing data caused by data transmission interruptions, equipment failures, etc., interpolation methods (such as linear interpolation, spline interpolation), mean imputation, forward / backward imputation, or more complex model predictions (such as the ARIMA time series model) are used to supplement the data and ensure its integrity. Data of different dimensions are processed to make them comparable; for example, power consumption, traffic, and temperature data are mapped to a unified numerical range. The relationship analysis layer is the core of the site group profiling. Based on the processed data, it calculates the similarity and correlation between each site. Business correlation analysis is performed by analyzing the peak business periods of adjacent base stations, the overlap of business types (such as voice, video, IoT), and whether the business volume of related base stations increases when the business volume of one base station decreases, to assess the potential for business transfer and complementarity. Methods such as correlation coefficients, mutual information, and Granger causality tests can be used. Linear interpolation: , Given data points, : Corresponding time, :exist Estimated time value. Spline interpolation (cubic spline): , , The coefficients of the piecewise polynomial are determined by the boundary continuity condition. Adjacent time points. ARIMA time series forecasting model: , Time series values : shift operator, , Autoregressive polynomial, : Moving average polynomial : Difference order, White noise error term. Pearson correlation coefficient: , Site and The linear correlation coefficient between them :time The observation values ​​at the two stations at that time Mean : Observation time length. Mutual information: , Two random variables (such as the business sequences of two sites). Joint probability distribution Marginal distribution; mutual information characterizes non-linear dependence; a larger value indicates a stronger correlation. Granger causality test: , Time series data from two sites Lag order : Regression coefficient, : Residual term, if If at least one of them is significantly nonzero, then it is called right It exhibits Granger causality. Energy consumption complementarity analysis is performed to evaluate the energy consumption characteristics of different base stations at different time periods, identifying which base stations can be shut down during a certain period, while their services can be taken over by other base stations, with the energy increase of the taking-over base station being less than the energy saving of the shut-down base station, achieving overall energy consumption optimization. Geographic proximity analysis is performed, calculating the distance between sites based on the geographical location information of the base stations, and determining the base station group topology and its impact on service carrying and handover by combining wireless signal coverage and interference conditions. The structure generation layer establishes a base station group relationship model based on the results of the relationship analysis layer. A base station group association matrix is ​​constructed, where rows and columns represent different base stations, and the elements in the matrix represent the association strength or type between base station i and base station j (such as service transfer coefficient, signal coverage overlap, geographical distance, etc.). A base station group topology graph structure is constructed to represent the connections and dependencies between base stations, facilitating intuitive understanding and subsequent algorithm processing. Base stations can be treated as nodes, and association as edge weights. The output layer encapsulates the generated base station group association matrix and related analysis results into a standard-format base station profile result file for subsequent modules to use. This profile contains a comprehensive and structured description of the current and historical operational status of the base station cluster. When a new base station is put into operation or an existing base station is decommissioned, the system, through a periodic update mechanism, can automatically trigger the base station cluster profile module to re-execute the above process, updating the base station cluster relationship matrix and base station cluster profile data, ensuring that the model is synchronized with the actual operational status, and ensuring the real-time nature and accuracy of decision-making. Energy Complementarity Index: , Base station and Energy complementarity : The average energy consumption of the corresponding base station in the same period. The closer the value is to 1, the stronger the energy complementarity. Geographic distance calculation (spherical distance formula): , Base station Geographical distance between them :latitude, :longitude, Earth's radius. Definition of a site cluster association matrix: , Website cluster association matrix Base station and The overall correlation strength The weighting coefficients for business relevance, energy complementarity, and geographical proximity are as follows: , Distance attenuation parameter: This matrix is ​​the core data structure of the station cluster structure generation layer. Edge weight normalization: , Normalized correlation strength : Number of sites. The normalized matrix can be directly used for subsequent topology construction and visualization.

[0021] The load assessment module comprises a feature fusion layer, a prediction layer, a thermal analysis layer, and a security assessment layer. The feature fusion layer integrates business data with external environmental parameters; the prediction layer calculates future load trends; the thermal analysis layer simulates temperature changes; and the security assessment layer determines the thermal safety level and the acceptable shutdown time. The load assessment module is one of the system's prediction and assessment layers. Its function is to perform refined load prediction and thermal safety analysis based on site group profiles and external environmental data. By fusing business data and meteorological parameters, this module can accurately determine the temperature change trend, thermal safety level, and the safe shutdown duration for base station equipment rooms. The feature fusion layer is responsible for integrating business data from the site group profile module (such as predicted business traffic, number of users, activity levels, etc.) and external environmental parameters (such as historical and predicted meteorological data: ambient temperature, humidity, wind speed, solar radiation intensity, etc.). Data fusion algorithms are employed to fuse multi-source data, such as Kalman filtering, deep learning models (e.g., LSTM), or neural network autoencoders. This aligns, cleans, and performs feature engineering on heterogeneous data, generating a unified, high-dimensional feature vector. Based on different scenarios (e.g., day / night, sunny / rainy) or seasons, the weights of various features in the prediction model are dynamically adjusted to improve prediction accuracy. The Kalman filtering fusion update equation... , , , , , :time The predicted state vector, :time The updated state vector (fusion result). The state transition matrix describes how the system changes over time. : Control input matrix, : Control input vector, Predict the covariance matrix. Update the covariance matrix. Process noise covariance matrix : Observation noise covariance matrix Observation matrix Kalman gain, Observation, : Identity matrix. By fusing business and environmental data from different sources, the data are aligned temporally and denoised to obtain a continuous and stable state estimation vector. Autoencoder feature compression model. , , Input feature vector (result of multi-source operational and meteorological data splicing). Hidden layer representation, compressed features. : Reconstructed vector, used to calculate reconstruction error. Weight matrix Bias term, Non-linear activation functions (commonly ReLU or Sigmoid). Enables unsupervised feature extraction from high-dimensional heterogeneous data, compressing the original input into a low-dimensional, high-information-density fused vector. LSTM temporal fusion unit (Long Short-Term Memory network). , , , , , , :time Input features, Hidden state vector : Memory unit state, : Gating weight matrices, Bias term, Sigmoid activation function Hyperbolic tangent function Hadamard element-wise multiplication fuses data features from different time steps while preserving long-term dependency information to generate time-consistent feature sequences. The prediction layer calculates load trends over a future period based on the fused feature data. It uses time series prediction models (such as ARIMA and Prophet), machine learning models (such as XGBoost and LightGBM), or deep learning models (such as RNN and Transformer) to predict the service load (such as traffic demand, CPU utilization, and memory utilization) of each base station and base station cluster over the next few hours or days. Based on historical energy consumption data, service load prediction results, and external environmental factors, it predicts the future energy consumption curve of the base stations. ARIMA time series prediction model: , Time series variables (business load or energy consumption). Lag operator ( ), Autoregressive polynomial, order , Moving average polynomial, order , Difference degree White noise term. XGBoost weighted tree model: , :sample The prediction results : No. Tree regression function, The function space of all trees. The total number of trees Input feature vector. Objective function: , Loss function (e.g., squared error) Regularization term, controlling the complexity of the tree. Core computation of the Transformer prediction network (multi-head attention): , : Query, key-value matrix, : Key vector dimension Normalization function. Implements dependency weight allocation across multiple time steps, capturing complex load variation patterns over long time intervals. Thermal analysis layer simulates temperature changes within the base station equipment room. A physical thermal model of the base station equipment room is established, fusing service and environmental data from different sources to align the data temporally and remove noise, resulting in a continuous and stable state estimation vector. Equipment room thermal balance equation (steady-state thermal resistance network model): ,in: , The heat output of the equipment varies with the load. Heat dissipation through the building envelope. The heat removed by the cooling system Average temperature inside the computer room Outdoor temperature, Total thermal resistance (resulting from the combined effects of wall structure, airflow, and heat conduction path). Dynamic temperature change differential equation (heat capacity model): , Equivalent heat capacity of the computer room :time The temperature of the computer room Equipment heating power : Cooling equipment heat dissipation power Equivalent thermal resistance External temperature. Key point temperature rise simulation function: , The first in the computer room Temperature at key points (such as the power supply area and the server area). Average temperature in the computer room Localized temperature rise depends on heat flow distribution. Data features from different time steps are integrated, and long-term dependency information is retained to generate time-consistent feature sequences, insulation performance, efficiency of heat dissipation equipment (air conditioners, fans), equipment heat generation (load-related), and external ambient temperature. This can be constructed using CFD (Computational Fluid Dynamics) simulation, finite element analysis, or simplified thermal resistance network models. Combining future service load and energy consumption data provided by the prediction layer, as well as meteorological parameters, real-time temperature change curves of key locations inside the equipment room (such as core equipment areas and battery compartments) are simulated under different shutdown strategies. Based on the simulation results from the thermal analysis layer, the security assessment layer determines the thermal safety level and shutdown time of the base station equipment room. According to equipment manufacturer specifications, industry standards, and operator operating experience, the maximum allowable operating temperature for each device and the overall thermal safety threshold of the equipment room are set. By comparing the simulated temperature curves with the safety thresholds, the thermal state of the equipment room is divided into multiple levels, such as "Safe," "Alert," "Warning," and "Danger." Based on the current thermal safety level and predicted temperature change trends, the maximum sustainable shutdown duration for each base station equipment room is calculated to ensure that the room temperature does not exceed the safety threshold during the shutdown period. For example, if the simulation shows that the temperature will reach the warning line after 1 hour of shutdown, then its shutdown duration is 1 hour. The thermal safety boundary output by this layer provides important constraints for the game-theoretic decision-making module. Thermal safety level determination criteria: , Thermal safety rating The highest temperature obtained from the simulation. Safety threshold Warning threshold Danger threshold. Formula for calculating shutdown duration: , Maximum shutdown duration Temperature curves obtained from thermal analysis model calculations. Warning temperature threshold: When the temperature reaches the warning threshold, the system determines that the shutdown time limit has been reached. Thermal safety boundary function: , Thermal safety boundary curve Load power, External temperature The highest temperature function output by the thermal analysis model. Integrated model of energy consumption prediction and temperature response: , Predicted energy consumption value Predict load power. External temperature : Empirical coefficients, obtained by fitting historical data. Temperature and energy consumption constraints: , , Real-time temperature inside the computer room. Cumulative energy consumption of the computer room Maximum permissible temperature Energy consumption limit threshold. Thermal safety optimization objective: , , , :time energy consumption :time Load power, : Duration of shutdown Computer room temperature Safe upper temperature limit Maximum load power.

[0022] The lifespan assessment module comprises a feature extraction layer, a lifespan calculation layer, a history correction layer, and an update layer. The feature extraction layer extracts lifespan-related features from power fluctuations, thermal cycling, and maintenance information. The lifespan calculation layer extrapolates the lifespan distribution based on multidimensional data. The history correction layer corrects estimation errors, and the update layer adjusts the lifespan prediction results when operational data changes. The lifespan assessment module is the second predictive assessment layer of the system. Its main function is to accurately calculate the equipment's lifespan distribution and reliability range based on the equipment's operational lifespan data. This module considers key factors such as equipment operating time, thermal cycle count, power fluctuations, and maintenance records to generate an estimate of the equipment's remaining lifespan and a suggested recovery interval. The feature extraction layer identifies and extracts features directly related to equipment lifespan degradation from a large amount of operational data. It records the equipment's cumulative operating time, a fundamental indicator of lifespan loss. It monitors the equipment's start / stop cycles and calculates the number of temperature changes (thermal cycles) the equipment experiences, as frequent temperature fluctuations can lead to material fatigue. This can be indirectly estimated by monitoring the number of times the upper and lower limits of the equipment room temperature sensor are crossed. Analyze the fluctuations and frequency of equipment power output or power consumption during operation. Large power fluctuations may indicate changes in internal stress, accelerating aging. For example, calculate the variance, peak-to-valley difference, or rate of change of the power curve. Collect historical records of equipment maintenance, component replacement, firmware upgrades, etc., as discrete event features affecting equipment lifespan. In addition to the ambient temperature of the computer room, the impact of environmental factors such as humidity, vibration, and dust on equipment lifespan can be considered and quantified as features. The lifespan calculation layer extrapolates the equipment's lifespan distribution based on the extracted multidimensional feature data. Reliability statistical models (such as Weibull distribution, exponential distribution, log-normal distribution, etc.) or physics-based degradation models (such as Arrhenius model, Eyring model) are used to describe the probability distribution of equipment lifespan. Regression models (such as Support Vector Regression SVR, Random Forest RF), neural networks (such as LSTM for predicting degradation trends), etc., are used to establish a mapping relationship between features and remaining lifespan. Based on the current equipment state and the above models, the remaining usable lifespan of the equipment is calculated in real time. For example, it can predict how long the equipment can continue to operate before reaching a certain failure threshold (such as performance degradation of x%). In addition to point estimation, this layer also provides confidence intervals for remaining lifetime, representing the uncertainty of the prediction results and providing more comprehensive information for game theory decision-making. The historical correction layer is used to correct errors in lifetime estimation and improve prediction accuracy. It analyzes historical failure data, identifies common failure modes and causes, and feeds this information back into the lifetime calculation model to optimize parameters. After each maintenance or significant change in equipment condition, the lifetime calculation model is retrospectively corrected and calibrated using new data, for example, through Bayesian update methods. The actual lifecycle data of the equipment is compared with the predicted lifetime in real time to identify and reduce systematic biases.The update layer can trigger adjustments to lifetime prediction results when operational data changes. New operational data is injected in real time, and when new operational data (such as updated power fluctuation data or new maintenance records) flows in, it automatically reruns part or all of the lifetime calculation process. Lifetime prediction is re-performed at set intervals (e.g., daily, weekly) to provide the latest remaining lifetime estimate, ensuring timely and accurate decision-making. The lifetime assessment module provides the game-theoretic decision-making module with key reliability constraint information, including the equipment's remaining lifetime estimate and the recommended minimum recovery interval (i.e., how long the equipment must operate after shutdown before being shut down again to avoid damage caused by frequent power-on / off cycles). Weibull distribution model: , This refers to the operating time of the equipment. It is a scale parameter (the average lifespan of the equipment). It is a shape parameter that determines the shape of the distribution. Indicates an early failure. This indicates aging failure. For the device in time The failure probability density at time step 1. The probability density function (PDF) of the exponential distribution model is: , Failure rate represents the frequency of equipment failure. It is time. It is the device in time The failure probability density at time t. The probability density function of the log-normal distribution model is: , The operating time of the equipment. The mean of a log-normal distribution. The standard deviation is given. Arrhenius model: , It is a constant. It refers to activation energy, which indicates the effect of temperature on equipment lifespan. It is Boltzmann's constant. It's temperature. Eyring model: , and It is a constant. It's temperature. Support Vector Regression Model: , It is a weight vector. These are input features. It is a bias term, which is applied by non-linearly mapping the input features and using a loss function. -Insensitive loss is used to avoid overfitting. The random forest model derives its final output by combining the results of multiple decision trees. , For the first The predicted values ​​of each decision tree. Let be the number of trees. The recursive formula for LSTM is: , , , , , , It's the forgetting gate, which determines how much information from the previous state is forgotten. It's the input gate, which determines how much of the currently input information is stored in the state. It is a candidate state, representing the impact of the current input. This is the current state of the cell. It's the output gate, which determines how much information is output from the cell state. This is the current hidden state. The Bayesian update formula is: , These are model parameters (such as lifetime prediction parameters). It is the observed data. It is the posterior probability. It is the likelihood function. It is the prior probability. This is evidence. Remaining life expectancy prediction formula: , It is the expected failure time of the equipment. This is the current time.

[0023] The game theory decision-making module comprises a state modeling layer, an optimization calculation layer, and a strategy generation layer. The state modeling layer describes the operational status of the base station cluster, the optimization calculation layer calculates the coordinated shutdown schemes for each site, and the strategy generation layer outputs energy-saving strategies that comply with security and reliability constraints. The game theory decision-making module is the core decision-making layer of the system. Its main function is to establish an optimization decision-making model among multiple base stations based on the thermal safety boundary data provided by the load assessment module and the equipment reliability constraint data provided by the lifetime assessment module, and to generate shutdown and recovery plans that comply with thermal safety and lifetime constraints. This module aims to maximize the overall energy saving of the base station cluster while ensuring service quality and equipment health. The state modeling layer comprehensively describes the current operational status of the base station cluster, providing input for optimization calculations. The state of each base station can be abstracted as a multi-dimensional vector to represent the base station state, including: current service load (traffic, number of users), equipment room temperature, remaining equipment lifetime, historical shutdown records, adjacent base station states, and service quality requirements from the upper-level network. The system integrates the site cluster association matrix generated by the site cluster profiling module to establish site cluster collaboration relationships and clarify the service carrying capacity and service transfer paths between each base station. The system quantifies the thermal safety level (e.g., maximum allowable shutdown duration) output by the load assessment module, the remaining lifetime estimate and recovery interval output by the lifetime assessment module, and other service level agreement (SLA) requirements into constraints in the model. The objective function of the game-theoretic decision is defined, typically maximizing the total energy saving benefit (i.e., energy savings from shutting down base stations) while penalizing violations of service SLAs, thermal safety exceeding limits, or accelerated equipment lifetime degradation. The optimization computation layer is the core of the game-theoretic decision, used to calculate the collaborative shutdown scheme for each site. Since energy saving, service quality, and equipment lifetime are multi-objectives and may conflict, a multi-objective optimization algorithm is required. For example, evolutionary algorithms such as NSGA-II (Non-dominated Sorting Genetic Algorithm II) or MOEA / D (Multi-Objective Evolutionary Algorithm based on Decomposition) can be used to find a series of trade-off solutions on the Pareto optimal frontier. By employing game theory models such as Nash equilibrium, Stackelberg game, or Shapley value, the shutdown and restoration of a base station cluster is regarded as a non-cooperative or cooperative game process. Each base station (or base station cluster) is a "player", and its decision (shutdown or operation) will affect the benefits (energy saving, service quality) of other players, thereby finding the optimal strategy.In complex and dynamic environments, Q-learning, SARSA, or deep reinforcement learning models (such as DQN and PPO) can be used to enable agents to learn optimal shutdown / restore decision strategies under different states through interaction with the environment, maximizing long-term cumulative rewards (energy-saving benefits). The switching states of base stations are modeled as binary variables, and the constraints and objective functions are transformed into linear forms. Optimization calculations are performed using solvers such as Cplex and Gurobi, suitable for small to medium-sized networks. For time-series decision problems, dynamic programming algorithms are used to decompose the problem into sub-problems, progressively solving for the optimal shutdown sequence. The policy generation layer transforms the abstract optimization results obtained from the optimization calculation layer into specific, executable energy-saving strategies. Detailed shutdown / restore plans are generated, explicitly specifying: which base stations (or their internal devices) will begin shutdown, for how long, when to resume operation, and how adjacent base stations will take over their services during the shutdown period. For each shutdown plan, a set of backup restore instructions or alternative solutions is also generated to cope with unforeseen circumstances or uncertainties. The output includes a decision report containing information such as expected energy-saving benefits, impact assessment on service quality, and prediction of impact on equipment lifespan. The output shutdown plan is the input to the linkage orchestration module. Base station energy-saving game objective function: , Number of base stations : Decision vectors for each base station Indicates base station The off (0) or running (1) state, : Except base stations The decision set of other base stations, Base station In decision-making Energy saving benefits (unit: kWh) The service quality loss caused by this decision (such as increased disconnection rate and latency), expressed in penalty coefficients. The thermal risk costs resulting from violating thermal safety boundaries (such as penalties for exceeding the threshold temperature in the computer room). The cost of accelerated equipment lifespan deterioration. These are the dimensionless penalty weights for service quality, thermal safety, and lifetime loss, respectively. Base station revenue function (Nash equilibrium condition): , Base station The individual benefit function, and the above formula , , , Common definition, Base station The optimal strategy The set of optimal strategies for other base stations at equilibrium. Stackelberg game optimization function: , : Strategy variables of the leader (superior control center or main base station) : The policy variable of the follower (subordinate base station or lower-level node) The leader's payoff function under its optimal response. Followers in Leader Strategy The Shapley value payoff function is as follows: , Base station The Shapley value represents the fair distribution of benefits in a cooperative game. Base station set : Does not include base stations any subset, Base station subset The total benefit function (e.g., total energy savings or QoS improvement). :gather The number of elements. Multi-objective optimization fitness function (NSGA-II framework): , Multi-objective vector Energy saving benefits (to be maximized, hence the negative sign in the target). Service quality loss (to be minimized) Equipment lifespan degradation (to be minimized). Reinforcement learning state-action value function (Q-learning model): , In state Next action The expected cumulative reward value, Learning rate ( ), Discount factor, which measures the importance of future rewards. , Instant reward value (e.g., energy saving gain). Possible actions at the next time step. , The current and next state vectors contain features such as load, temperature, and lifetime. Dynamic programming recursive equation: , At any moment state The optimal value function under the given conditions, At any moment The decision to close the door, : Immediate benefit function (comprehensive energy saving, penalty, etc.) The value function at the next moment. Future return discount factor. Constraints of the linear optimization model: , , , Base station Power consumption (kW) : Maximum permissible operating power (kW) Base station In state The temperature of the computer room (°C). : The thermal safety threshold (°C) of this base station. The device is in status The remaining life estimate (in hours) is as follows. Minimum lifetime safety limit. Strategy generation layer energy efficiency prediction function: , : Expected total energy savings (kWh) : Optimal decision-making strategy. The meanings of other parameters are consistent with the objective function of the base station energy-saving game.

[0024] The linkage orchestration module comprises a rule engine layer, a verification layer, a scheduling layer, and a feedback layer. The rule engine layer stores security constraints, the verification layer checks the logical consistency between control commands, the scheduling layer adjusts the execution order based on risk levels, and the feedback layer records execution results and updates the rule base. The linkage orchestration module is the system's execution control layer, its main function being to perform rigorous semantic verification and secure orchestration of the shutdown plan output by the game theory decision-making module. This module performs consistency checks and task scheduling based on a preset rule base, ultimately generating control commands that can be executed securely and in an orderly manner. The rule engine layer is the core of the module, used to store and manage all security constraints, business rules, and operational specifications. It has a rule base containing hard rules (such as minimum coverage and maximum disconnection rate), device operating procedures (such as the minimum interval between two shutdowns), alarm thresholds, and emergency plan triggering conditions. It defines the storage format and data model of the rules; rules can be based on condition-action (IF-THEN) formats. The system selects and adopts a mature rule engine framework, such as Drools or a custom rule engine, for efficient matching and inference using the Rete algorithm. Security constraint rules are written, transforming all known security constraints into specific rules and entering them into the rule base, including SLA compliance rules, device protection rules, and operational strategy rules. The rule base is dynamically updated; through a visual interface or API, operations personnel can add, modify, or delete rules, adapting the system to constantly changing business needs and network environments. The feedback layer can also update the rule base based on execution results. The verification layer receives the shutdown plan generated by the game theory decision module and performs multiple verifications to ensure its security and compliance. Execution logic consistency checks are performed, checking for conflicting instructions in the plan and ensuring the correct sequence of all dependent actions; for example, two mutually backup core devices cannot be shut down simultaneously. The rule base is validated by comparing the shutdown plan against the security constraints stored in the rule engine layer. Each pre-defined rule is checked to ensure the shutdown plan meets all preset rules. Each instruction and related parameter of the shutdown plan is injected into the rule engine as a fact. The rule engine performs matching and reasoning based on the preset rules. If any rule is triggered, the plan is marked as non-compliant. Examples of non-compliance include: whether the service capacity of the remaining base stations can still meet the minimum quality of service requirements after shutting down a base station; whether the shutdown duration exceeds thermal safety or lifetime constraints; and whether such shutdown operations are permitted during the current time period. The system also checks whether the shutdown plan will cause conflicts with network resources (such as bandwidth and processing capacity) or power resources. A risk assessment is performed on the validated shutdown plan, quantifying its potential risks (such as the probability of impact on business and possible losses). A detailed validation report is output, indicating which rules were violated and the reasons for the violations.Finally, a risk assessment is performed on the verified shutdown plans to calculate their risk level and trigger a manual approval process for high-risk plans. The scheduling layer designs scheduling strategies and optimizes the execution order of verified shutdown plans based on risk assessment results, base station importance, and geographical topology. For example, it may proceed from edge base stations to core base stations or shut down low-risk base stations first. Time windows are managed to ensure shutdown operations are executed within predetermined time windows, avoiding impact on peak service periods. Complex shutdown plans are decomposed into a series of atomic, independently executable control commands, such as "command base station A to enter energy-saving mode" or "adjust the downtilt angle of base station B sector." Based on the preset scheduling strategy and risk level, task scheduling is executed, and the scheduling results are converted into control commands that can be recognized and executed by the lower-level equipment of the base station. These commands are then distributed to the remote control units of each base station via southbound interfaces (such as TR-069, Netconf / Yang, or proprietary protocols). The feedback layer first receives the execution results of control commands from the base station or lower-level management system, as well as related equipment response data. All command issuance, execution status, device response, and any anomalies are recorded, including detailed timestamps, target devices, command content, and results. These logs are transmitted to the audit and display module. Based on the command execution results, the real-time base station operating status is updated in the central database or status center. Rule base updates and self-learning are performed, analyzing the actual effects of command execution. Based on new situations or unforeseen problems that arise during execution, the rule base of the rule engine layer is automatically or semi-automatically updated. Simultaneously, the execution results are fed back to the game decision module to optimize its internal reward function or model parameters. Formal definition formula for condition-action rules: , : No. Rule 1 :rule The number of conditions included. : No. A conditional function, defined on the input fact variables. (e.g., temperature, load, time period, lifespan, etc.) Input factual variables (e.g., current base station load, equipment room temperature, shutdown duration, remaining lifetime, etc.). Actions triggered when conditions are met (e.g., shutdown, delayed execution, manual approval, etc.). Logical AND operator. Rete algorithm matching efficiency optimization function: , Rule matching score, used to measure the activation level of the matching network. Total number of rules :rule The number of conditions, : Boolean value, indicating if the fact satisfies the condition Then take Otherwise , : Condition node weights, reflecting their importance or priority. Plan consistency verification function: , Shutdown plan consistency score (range of values) (The closer to 1, the more consistent they are). : A set of instruction pairs that may conflict (such as devices that are backups for each other or have dependencies). : Instruction conflict indicator function, if instruction and Conflict ,otherwise Rule validation consistency score: , Rule consistency score (range of values) ), Number of rules :rule The violation indicator variable, if the rule is violated... ,otherwise Multidimensional risk assessment model: , Total risk value (dimensionless or normalized to): ), Number of risk dimensions (such as business risk, equipment risk, thermal safety risk, power risk, etc.) : No. Weights of each risk dimension ( ), Probability of risk occurrence (value) ), Risk impact loss value (units can be cost, energy loss, default penalties, etc.). Shutdown task scheduling optimization objective function: , Shutdown execution order (scheduling sequence). Number of base stations to be scheduled Base station In the scheduling sequence Risk indicators below The time delay cost of the base station shutdown execution (such as waiting time or window mismatch). : The goal of optimizing the scheduling order (such as geographical proximity or importance penalty). Weighted coefficients for risk, delay, and optimization objectives (satisfying) Time window constraint model: , Base station The shutdown start time. Base station Recovery time : The earliest allowed execution time (set according to off-peak business hours). : The latest allowed end time (to avoid peak business periods). Atomized control instruction generation function: , Base station The Atomization control instructions, Base stations to be shut down High-level actions (shutdown / resumption). Command type parameters (such as energy saving mode adjustment, power limiting, downtilt angle adjustment, etc.). Base station specific parameters (equipment model, interface protocol, response delay, etc.). Control command mapping functions translate high-level decisions into a command format recognizable by the device. Feedback-based self-learning update rules (rule weight adjustment): , :rule At any moment Importance weights Learning rate The actual effect score after the rule is executed (such as execution success rate, false alarm rate, etc.). : Rule expected effect score. Execution feedback closed-loop feedback model: , System state vector (including the operating status and rule status of each base station). : Execution result error vector (the deviation between the expected instruction result and the actual feedback). Device response data (such as execution time, alarm status, temperature changes). : State update function, used to correct the system state based on execution feedback.

[0025] The anomaly verification module comprises a sampling layer, an analysis layer, an identification layer, and a recovery layer. The sampling layer collects operational parameters in real time, the analysis layer assesses the stability of the operational status, the identification layer detects the type and scope of anomalies, and the recovery layer generates and executes adaptive recovery commands. The anomaly verification module serves as the system's real-time monitoring and adaptive recovery layer. Its main function is to continuously monitor the system's operational status during the shutdown plan execution. This module can identify anomalies in parameters such as temperature, energy consumption, and service quality in real time, and execute automatic recovery operations based on the level and type of the anomaly, ensuring service continuity and equipment security. The sampling layer is responsible for collecting various operational parameters of the base station in real time and at high frequency. The sampling layer defines monitoring data points, clearly defining key operational parameters that need to be monitored in real time. This includes environmental data such as indoor temperature and humidity, and outdoor ambient temperature; energy consumption data such as total base station power and real-time power consumption of key equipment like BBU / RRU; KPI business data such as user activity, service traffic, RRC connections, drop rate, latency, throughput, and handover success rate; and equipment status data such as the operating status and fault alarm information of various devices (BBU, RRU, transmission equipment, power supply, air conditioning). Monitoring agents or interfaces are deployed in various parts of the system. Based on the importance and rate of change of the parameters, reasonable collection cycles are set to acquire data from base station equipment and data center environmental monitors. Data is pulled or subscribed to using standard interfaces provided by the base station; for example, temperature and service traffic can be collected every minute or every 10 seconds to ensure timely response to anomalies. The collected real-time data is sent to the analysis layer in real time through high-performance message queues (such as Kafka and Pulsar) to ensure low latency and high throughput data transmission. The collected data undergoes timestamp calibration and preliminary format conversion and data type validation to ensure data quality. The analysis layer uses stream processing technologies (such as Apache Flink and Apache Spark Streaming) to build a real-time data stream processing pipeline, continuously analyzing the incoming real-time data to determine the stability of the operational status. This includes aggregating high-frequency data within a sliding time window (e.g., the past 1 minute, the past 5 minutes), calculating statistics (mean, maximum, minimum, standard deviation, etc.), and assessing its fluctuation amplitude; abnormally drastic fluctuations are usually warning signals. An operational status assessment model is built, using time series analysis methods (such as moving averages and exponential smoothing) to evaluate parameter trends and determine if there are any abnormal upward / downward trends. A historical baseline model is established for each monitored parameter under normal operating conditions. Real-time data is compared with historical baseline data before shutdown, or with a trained normal operating mode, calculating the deviation between real-time data and the baseline model. Methods such as moving averages and exponential smoothing are used to monitor short-term and long-term trends of parameters and analyze the real-time correlation between multiple parameters.The identification layer constructs anomaly detection algorithms, employing statistical anomaly detection methods (such as statistical thresholds and distance-based anomaly detection), machine learning anomaly detection methods (such as IsolationForest, One-ClassSVM, LocalOutlierFactor, or autoencoders), and a rule engine, combined with preset business rules to automatically identify anomalies. It implements anomaly type and scope identification, categorizing anomalies into different types, such as overheating anomalies, service quality degradation anomalies, unexpected energy consumption anomalies, and equipment failure anomalies. If the failure mode is known, a classifier (such as a decision tree or random forest) can be trained to identify the anomaly type, and the scope of the anomaly's impact can be determined based on the source of the anomaly parameters and the site cluster association information. Anomalies are classified into different levels (e.g., "general alarm," "critical alarm," "emergency alarm") based on their severity and potential impact, with each level corresponding to different processing priorities and recovery strategies, allowing the recovery layer to take different levels of response measures. Where possible, the root cause of the anomaly is analyzed, such as whether it is due to improper shutdown strategies, sudden changes in the external environment, or equipment malfunction. The recovery layer constructs a recovery strategy library, pre-configuring multiple recovery strategies. For each anomaly type and level, one or more executable recovery strategies are pre-defined. For example, for a warning-level temperature increase, some cooling equipment can be activated first; for a severe-level service quality degradation, it may be necessary to immediately restore the shut-down core equipment. Strategies can be simple action sequences or complex decision trees. Adaptive recovery logic is implemented. When the identification layer detects an anomaly and determines its level, the corresponding recovery strategy is automatically triggered. The generation of recovery commands is not fixed but dynamically adjusted based on the parameters of the current specific anomaly state and network resource conditions. For example, if a base station experiences an abnormal temperature, its fans are activated first; if this is ineffective, its air conditioning is activated. During the recovery process, if necessary, it collaborates with the base station profiling module and the game theory decision module to dynamically adjust service routing, transferring services from affected base stations to other healthy base stations. Recovery commands are sent to the corresponding base stations or equipment through the control channel, and relevant parameters are continuously monitored to verify whether the recovery operation achieves the expected results. If the effect is unsatisfactory, the next stage of the recovery plan can be tried. Finally, for operations that may cause greater impact, a rollback mechanism should be designed to quickly undo the operation if a problem is detected. If this is still ineffective or the business suffers significant impact, a hard shutdown recovery operation should be executed immediately, and a multi-step recovery mechanism should be designed to address persistent problems. Through the anomaly verification module, the system can achieve real-time, fine-grained monitoring of the remote hard shutdown process and has strong self-healing capabilities, providing core security for energy-saving control. Sliding time window statistical aggregation formula: , , : within the time window The mean of the parameters within, The standard deviation within the same window reflects volatility. : No. Time-sampling value, : Length of the sliding time window (unit: seconds or minutes). Current time index. Exponential smoothing trend estimation model: , : The predicted smoothed value for the next time step. : The measured value at the current moment, The smoothed estimate from the previous time step. Smoothing factor ( Historical baseline deviation calculation: , : Current parameter deviation Real-time monitoring values The historical average of this parameter under normal conditions. : The historical standard deviation of this parameter. Multi-parameter correlation anomaly detection function: , In time parameters and Pearson correlation coefficient, Covariance of two parameters , : Standard deviation of the two parameters , Real-time values ​​of two monitoring metrics. Statistical threshold anomaly detection rules: , : Abnormal flag, 1 indicates abnormal, 0 indicates normal. Current monitoring value, Historical baseline mean Historical baseline standard deviation Threshold coefficient (usually 2 or 3). Distance-based outlier metric (LOF local outlier factor): ,in, , :sample Local outlier :sample of A nearest neighbor set Local reachability density. :sample and Distance metrics between them (such as Euclidean distance). : Neighbor count exceeds parameter. IsolationForest anomaly score: , :sample Abnormal scores ( (The closer it is to 1, the more abnormal it is). :sample The average path length (average tree depth from the root node to the leaf node). Normalization factor ,in For harmonic numbers, Sample size. Anomaly rating function: , Abnormal comprehensive score, Baseline deviation Multi-parameter average correlation shift (optional) ), Modeled anomaly scores Weighting coefficients for each indicator (satisfying) Recovery strategy selection function: , The final recovery strategy chosen. A set of feasible recovery strategies (such as "restart the fan", "enable the air conditioner", "rollback to shutdown"). In strategy The probability that the following anomaly will be eliminated. Benefits of strategy implementation (e.g., energy efficiency improvements or business recovery). The cost of strategy execution (increased energy consumption, risk, latency, etc.). : Cost penalty coefficient. Resume execution of closed-loop verification model: , The recovery error assessment value at the next moment. Current recovery error assessment value. Current monitoring parameters (such as temperature or disconnection rate). The target value or normal threshold after recovery. Forgetting factor ( Rollback trigger criteria: , Rollback trigger flag, : Recovery error, : Maximum allowable error threshold : Rate of change of recovery error. Adaptive feedback learning rule update: , The set of parameters (such as weights, thresholds, etc.) of the current recovery strategy model. Learning rate : Real feedback ratings on recovery effectiveness (such as success rate or recovery time metrics). The model's predicted recovery outcome. Multi-step recovery sequence optimization: , : Optimal recovery strategy sequence : A set of candidate recovery action sequences : Restore step count, Future earnings discount factor : No. The reward function for the step recovery action. The execution cost of this step. Anomaly impact propagation model: , Base station The amount of change due to abnormal influences Base stations in the site cluster association matrix and The connection weights, Base station The abnormal intensity (by or (Given) : with base station The set of directly adjacent base stations. Comprehensive abnormal health score: , System health score (range) (The higher the height, the healthier) Number of monitoring parameters :parameter Abnormal scores, : Parameter weights, : Normalization coefficient, making Scope .

[0026] The audit display module comprises a log recording layer, an evidence storage layer, a reputation assessment layer, and a display layer. The log recording layer stores operational logs; the evidence storage layer encrypts and prevents tampering of data; the reputation assessment layer calculates operational reputation scores; and the display layer generates audit reports and provides a graphical user interface. The audit display module serves as the system's result backtracking and visualization layer. Its main functions are recording system operational logs and energy efficiency data, performing encrypted evidence storage and reputation calculations, and ultimately generating detailed audit reports and intuitive operational visualizations. This module aims to provide transparent, traceable, and reliable energy-saving control process management. The log recording layer defines log categories and structures, including event logs, performance logs, and alarm logs. Event logs record all user operations (such as policy adjustments), interactions between modules within the system (such as policy output from the game decision-making module), control command issuance, equipment responses, abnormal alarms, recovery operations, and other information, including timestamps, operation subjects, event descriptions, and related data. Performance logs periodically or during critical events capture snapshots of base station business KPIs (such as call drop rate and latency), energy consumption data, equipment room temperature, and equipment lifespan estimates. These snapshots permanently store the original energy consumption data for each base station, energy consumption comparisons before and after shutdown, cumulative energy-saving data, and deviations between predicted and actual energy-saving benefits. Alarm logs specifically record alarm information issued by the anomaly verification module. A distributed log collection system is built, deploying log collection clients in various services and modules to monitor their respective log files. Collected logs are sent to a centralized log storage system (such as an Elasticsearch cluster, Splunk, or data warehouse) via message queues for unified log management and querying. The collected log data is cleaned, structured, and indexed for efficient subsequent searching and analysis. The evidence storage layer ensures the integrity, immutability, and trustworthiness of recorded data, implements data encryption, identifies sensitive information in the log data, and uses industry-standard symmetric encryption algorithms (such as AES-256) for encrypted storage and transmission of sensitive logs and energy efficiency data. By employing distributed ledger technology (such as blockchain) or trusted timestamp services, and defining data storage interfaces through smart contracts, key data is hashed and digitally signed on the blockchain. Batch-processed log data undergoes hash calculation, and the calculated hash value is submitted to a smart contract on the blockchain for storage, ensuring that data cannot be tampered with once recorded, providing strong evidence. Each log entry or batch of data can generate a cryptographic hash value, stored on the distributed ledger, enabling hierarchical storage and archiving of long-term data, ensuring historical data traceability, and optimizing storage resources, thereby achieving non-repudiation and traceability. Alternatively, third-party trusted timestamp services can be integrated to generate globally recognized timestamp certificates for key log entries, while using asymmetric encryption to digitally sign the key log data.Develop a data storage strategy, hierarchically archive long-term historical logs and energy efficiency data, and set data retention periods to ensure compliance and storage efficiency. The reputation assessment layer is used to calculate the system's operational reputation score, evaluate the overall performance of the energy-saving control system, and establish a comprehensive evaluation index system, including energy-saving benefits (actual energy saving rate, deviation between predicted and actual energy saving rate), service quality (SLA satisfaction, number of failures), equipment health (equipment failure rate, MTTR, lifetime prediction accuracy), system security (anomaly detection accuracy, recovery success rate, misoperation rate), and operation and maintenance efficiency (number of manual interventions, fault handling time). Establish a reputation calculation model, assign weights to each index, and calculate the overall system reputation score based on the weights and performance of each index using methods such as weighted summation, fuzzy comprehensive evaluation, analytic hierarchy process (AHP), or TOPSIS. A deduction / addition mechanism can be used to dynamically adjust the reputation score. In addition to the overall system reputation, reputation assessments can also be performed on individual modules (such as the strategy generation quality of the game decision-making module) or individual base stations (performance after shutdown) to identify potential bottlenecks or optimization points. Based on the latest operational data, logs, and evaluation results, the credit score is calculated and updated at a set frequency. The trend of credit score changes is monitored, and the credit evaluation results are fed back to the game decision-making module and the linkage orchestration module to optimize the strategy generation parameters of the game decision-making module or the rule base of the linkage orchestration module. The presentation layer provides an intuitive and user-friendly interface, displaying audit reports and operational visualizations. The user interface (UI) / user experience (UX) is designed to build an intuitive and interactive web interface or desktop application. The front-end can use React, Vue.js, or Angular, and the back-end can use Python (Django / Flask) or Java (SpringBoot). Multi-dimensional visualization is implemented, displaying real-time data and trend charts of core KPIs through dashboards. Geographic Information System (GIS) views are used to display the location, operational status, energy consumption heatmap, and service traffic distribution of all base stations on a map. Charts (such as ECharts, Highcharts, and D3.js) are used to display energy consumption curves, temperature change curves, service traffic prediction vs. actual comparison charts, equipment lifespan trend charts, and abnormal alarm distribution charts. The system displays the real-time operational status, energy consumption distribution, traffic flow, temperature curves, shutdown / recovery plan execution progress, abnormal alarm distribution, and historical trends of the base station cluster through charts, curves, heat maps, and Geographic Information System (GIS) interfaces. It provides an audit report generator with various customizable report templates and supports exporting to common formats such as PDF and Excel. Based on user needs, it automatically generates customized audit reports, including energy-saving benefit analysis, abnormal event lists, SLA compliance status, equipment health reports, and credit score trends, supporting multiple export formats such as PDF and Excel.It offers flexible query functions, allowing users to filter and query logs and historical data based on various conditions such as time range, base station ID, region, event type, and alarm level, and supports drill-down functionality. Finally, it implements real-time alarm notifications, sending real-time alerts to relevant operations and maintenance personnel via email, SMS, WeChat, and app push notifications when the credit score declines or high-level anomalies occur. Through the audit display module, operators can comprehensively understand the energy-saving system's operational efficiency, security, and reliability, providing data support for decision-makers and promoting continuous system optimization. Log entry normalization and fingerprint hashing: , : No. The fingerprint hash value of each log entry. Cryptographic hash functions (such as...) ), Field normalization and serialization functions ensure consistency across systems. : Timestamp (UTC, accurate to milliseconds). : Operation entity identifier (account or service identifier) Event type identifier (event log, performance log, alarm log). : Event description string, Structured representation of related data (such as KPI snapshots, parameter sets). Field-level join operations. Merkle root calculation for batch logs: , The Merkle root of this batch of logs. The algorithm for calculating the root of a binary Merkle tree from bottom to top. : All log fingerprint sequences within the batch Number of log entries within a batch. Blockchain evidence mapping and on-chain index: , On-chain notarized transaction identifier (transaction hash). The process of submitting hashes and metadata to a smart contract. Batch metadata (time range, source, signer ID), On-chain retrieval index, An indexing function based on contract address and block height. Smart contract address Block height. Merkle root. Digital signatures and verification (ECDSA abstraction): , , Digital signature Using the private key The signature operator, Using public key The signature verification operator, Message digest functions (such as...) ), Key data pending signature (such as...) (Report summary, report version number) , Public-private key pair. Trusted timestamp token (TST) generation and verification: , Trusted timestamp token The signature operator of the Timestamp Service Authority (TSA). Summary function : Time-stamped data (such as batch Merkle roots). : A timestamp (UTC) issued by the TSA. AES-256-GCM encryption and integrity label: , : Ciphertext, Integrity authentication label (128 bits). Use key AES-GCM encryption operator, Initialization vector (96-bit random value). Plain text (sensitive logs or energy efficiency data). Additional authentication data (such as log headers, version numbers). Symmetric key (256 bits). Cost optimization of hierarchical archiving and retention strategies: , , , Total storage cost Number of storage tiers (hot storage, warm storage, cold storage, offline storage). Data in the first Layer allocation ratio, : No. Unit capacity cost per floor : Assigned to the The amount of data in the layer The read / write reliability metrics for this layer. The average retrieval latency for this layer is... Minimum reliability constraint threshold Maximum allowed retrieval latency threshold. Actual energy saving rate versus prediction deviation: , , Actual energy saving rate Prediction bias normalization index Energy consumption before shutdown (baseline period, kWh). Energy consumption after shutdown (during execution, kWh). Energy savings (kWh) predicted by the model before execution. SLA compliance rate: , Service quality compliance rate : The number of time slices within the assessment period : No. Moment-time KPI vector (disconnection rate, latency, throughput, handover success rate). The set of KPIs that meet the SLA threshold. Indicator function, takes the value if the condition is met. Otherwise take Equipment health metrics (MTTR and failure rate): , , Mean repair time : No. The time required to repair this fault Number of failures during the statistical period Failure rate per unit time Number of failures during the statistical period Cumulative equipment uptime. Safety and maintenance efficiency indicators: , , Overall accuracy of anomaly detection The four-item count of a binary confusion matrix. Operations automation efficiency rating Number of manual interventions Total number of processing times (automatic + manual). Credit score (weighted summation model): , Overall system reputation score (normalized to) ), Number of indicator dimensions (energy saving efficiency, service quality, equipment health, system security, operation and maintenance efficiency). : No. The weights of each dimension (satisfying) ), : No. Normalized performance scores for each dimension. Fuzzy comprehensive evaluation (single-layer weighted type): , Fuzzy comprehensive scoring : Number of indicators The number of rating levels in the evaluation (e.g., Excellent, Good, Satisfactory, Unsatisfactory). : No. Each indicator weight, : No. Indicator for the first Membership degree of the level, : No. Quantitative scores for each level. AHP consistency test: , , Consistency indicators : Determine the largest eigenvalue of a matrix. Determine the order of a matrix. Consistency ratio Consistency index constant of the same order of randomness, judgment criteria: Consistency is considered acceptable. TOPSIS proximity calculation: , , , :plan The closer the similarity (the greater the better). The weighted Euclidean distance to the positive ideal solution. : Weighted Euclidean distance to the negative ideal solution :plan In terms of indicators Normalized value on, , :index Positive and negative ideal values, :index The weight, Number of indicators. Dynamic updating and deduction / addition mechanism for credit scores: , , :time and Reputation score, Historical score forgetting factor ( ), :time The Indicator performance score, Indicator weights : No. Intensity measures of violations or abnormal events (such as the number of high-level alarms). : No. Deduction coefficient for event type Number of types of incidents that resulted in point deductions : Number of indicators. Report consistency and traceability verification values: , Audit report consistency indicator (take) pass, (Not approved) Standardized content of the full audit report. Report signature Public key signature operator, Merkle root evidence stored on the blockchain. : Local batch logs The root of reconstruction Indicator function.

[0027] EWMA trend estimation of visual indicators: , : Smooth trend values ​​displayed on the dashboard (such as energy consumption curves, temperature curves). Original observations Smoothing coefficient ( ), Time index. High-level alarm trigger threshold rules: , High-level alarm trigger flag, Current credit score : Reputation threshold : The severity index of serious events in the current period (e.g., emergency alarm count). Severe event intensity threshold. Map heatmap weight normalization: , , Base stations on GIS heat maps Visual weights, Base station Comprehensive strength measurement, Energy intensity measurement (e.g., electricity consumption per unit time). Business impact metrics (such as traffic or user exposure). Alarm intensity metric (weighted by level). : Three linear composite weights (non-negative and normalizable) : The number of base stations displayed on the map.

[0028] The system features a periodic update mechanism. When a new base station is put into operation or an existing base station is decommissioned, the system automatically updates the base station group relationship matrix and base station group profile data to ensure that the model is synchronized with the actual operating status. The system runs in a cloud or local server environment. During system operation, the seven modules coordinate sequentially according to data flow and control flow. The base station group profile module provides structural feature input, the load thermal assessment module outputs thermal safety boundaries, the life assessment module provides reliability constraints, the game theory decision-making module generates energy-saving strategies, the linkage orchestration module executes strategy verification and distribution, the anomaly verification module realizes operation monitoring and adaptive recovery, and the audit display module completes data storage and result display. The modules exchange data and provide status feedback through standard interfaces, forming a closed loop of intelligent energy-saving control that enables continuous self-learning and parameter self-calibration.

[0029] Each module operates in a closed loop through data and control channels, realizing an energy-saving control process based on data perception, predictive evaluation, optimization decision-making, secure execution, anomaly verification, and result auditing. The data perception phase is undertaken by the station group profiling module. This module continuously collects raw operating status, energy consumption, service traffic, and environmental parameters from each base station through the data access layer. After cleaning, filling, and normalization by the data processing layer, the relationship analysis layer conducts in-depth analysis of service relationships, energy consumption complementarity, and geographical proximity characteristics between stations. Finally, the structure generation layer establishes a refined station group relationship matrix, and the output layer generates station group profiling data describing the current and historical status of the base station group. This profiling data serves as the foundational information for system operation and is efficiently distributed to subsequent modules via an internal data bus (such as Apache Kafka). The predictive evaluation phase is collaboratively completed by the load assessment module and the lifetime assessment module. The load assessment module receives service data provided by the station group profiling and external real-time / predictive meteorological data, and its feature fusion layer integrates this information. The prediction layer predicts future load and energy consumption trends based on the fused features. The thermal analysis layer uses an established data center thermal model to simulate temperature changes in the data center under different potential shutdown scenarios. The security assessment layer, based on temperature simulation results and preset thermal safety thresholds, accurately determines the thermal safety level and safe shutdown duration for each base station. The lifetime assessment module obtains key parameters such as device runtime, thermal cycle count, and power fluctuations from the data access layer, and the feature extraction layer extracts features strongly correlated with device lifetime. The lifetime calculation layer uses a pre-established lifetime prediction model (such as a Weibull distribution or machine learning model) to calculate the device's remaining lifetime estimate and reliability range, and performs continuous calibration through the historical correction layer and update layer, while providing a suggested minimum recovery interval. The outputs of these two assessment modules (such as thermal safety boundaries, shutdown duration, remaining lifetime estimates, and suggested recovery intervals) are transmitted to the game decision module as hard constraints and important considerations through an internal data channel. The optimization decision phase is handled by the game decision module. The game decision module receives base station group collaboration information from the base station group profiling module, the thermal safety boundaries from the load thermal assessment module, and the device reliability constraints provided by the lifetime assessment module. The state modeling layer constructs a complete representation of the station cluster's state and objective functions, covering multiple dimensions such as energy efficiency, service quality, and equipment health. The optimization calculation layer employs advanced optimization algorithms (such as the multi-objective genetic algorithm NSGA-II, mixed-integer linear programming, or reinforcement learning) to find the optimal base station combination shutdown and recovery plan within a multi-dimensional constraint space, aiming to maximize overall energy efficiency while strictly ensuring service SLA and equipment lifespan. The policy generation layer transforms the optimization results into specific shutdown and recovery plans, clearly specifying execution time, target base stations / equipment, and expected effects, and sends them to the linkage orchestration module via the control channel. The safe execution phase is implemented by the linkage orchestration module.The orchestration module receives the shutdown plan output by the game decision module. The verification layer first performs rigorous logical consistency verification and security compliance checks on the plan based on the rule base preset by the rule engine layer, ensuring that the plan complies with all business SLAs, thermal safety, and lifetime constraints. For plans that pass verification, the scheduling layer decomposes them into a series of atomic control tasks according to risk level and instruction optimization order. These tasks undergo protocol conversion via the southbound interface (e.g., conversion to SNMP's `SET` operation or Netconf / Yang's `edit-config` operation) and are distributed to the remote control units or devices of each base station through the control channel. The feedback layer receives the instruction execution results in real time and updates the system status, while providing rule base optimization suggestions based on the execution effect. Throughout the execution process, the anomaly verification module runs continuously, undertaking the responsibility of real-time monitoring and adaptive recovery. The sampling layer collects the operating parameters of each base station (including temperature, energy consumption, business KPIs, etc.) at high frequency in real time. The analysis layer performs real-time stream processing and status assessment on this data, calculating the deviation from the preset baseline. The identification layer employs advanced anomaly detection algorithms and preset rules to detect and classify anomaly types (such as excessively high temperature or degraded service quality) in real time, and assigns anomaly levels. Once a high-level anomaly is detected, the recovery layer immediately triggers a preset adaptive recovery strategy, generating and issuing recovery instructions (such as immediately restoring base station shutdown, adjusting air conditioning settings, or activating backup equipment), which are transmitted to the base station for execution via the control channel to ensure uninterrupted service or minimize impact. The system audits and visualizes the results through an audit display module. The log recording layer centrally collects and structures all critical logs (events, performance, alarms, recovery actions) and energy efficiency data during system operation. The evidence storage layer encrypts key data and uses blockchain or trusted timestamp services for tamper-proof evidence storage, ensuring data authenticity and non-repudiation. The reputation assessment layer calculates the reputation score of the system and each module based on multi-dimensional indicators (such as energy-saving effect, SLA, equipment health, system security, etc.) and analyzes its changing trends. The presentation layer visualizes the system's operational status, energy efficiency, energy consumption distribution, anomaly alarms, SLA compliance, and reputation assessment results through intuitive dashboards, Geographic Information System (GIS) views, various charts, and customized audit reports. It also provides flexible query and audit report export functions, offering quantitative and transparent decision support for operations and maintenance personnel and management. The entire system forms an intelligent closed loop of iterative optimization and continuous improvement. Data flow begins in the station group profiling module, passes through load assessment and lifespan evaluation, and is aggregated in the game theory decision-making module. The game theory decision-making module generates a control command flow, which is then executed after security verification and scheduling by the linkage orchestration module. The anomaly verification module monitors the execution results in real time and triggers recovery when anomalies occur. This anomaly information is also fed back to other modules for decision adjustments.The auditing and display module records, evaluates, and presents the entire process. Its evaluation results serve as guidance for optimizing the strategy parameters of the game-theoretic decision-making module and the rule base of the linkage orchestration module, thereby achieving continuous self-learning and parameter self-calibration of the system, continuously improving energy-saving performance and system stability. When a new base station is put into operation or an existing base station is decommissioned, the system automatically triggers the station group profiling module to re-execute the above process through a periodic update mechanism, updating the station group relationship matrix and station group profiling data to ensure that the model is synchronized with the actual operating status, providing an accurate and real-time data foundation for the entire closed loop. The system can run in a cloud or local server environment. During system operation, the seven modules coordinate sequentially according to data flow and control flow. The modules exchange data and provide status feedback through standard interfaces, forming a smart energy-saving control closed loop with continuous self-learning and parameter self-calibration.

[0030] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An energy-saving control system for remote intelligent hard shutdown in base station equipment rooms, characterized in that: include: The site group profiling module is used to collect and process the operating status, energy consumption data, service traffic and environmental parameters of multiple base stations, analyze the service association, energy consumption complementarity and geographical proximity characteristics between sites, establish a site group association matrix and generate a site group profile for subsequent analysis and control. The load assessment module is used to perform load prediction and thermal safety analysis based on the site group profile and external environment data. It determines the temperature change trend, thermal safety level and shutdown duration of the data center by integrating business data and meteorological parameters. The life assessment module is used to calculate the equipment life distribution and reliability range based on the equipment's operating time, number of thermal cycles, power fluctuations and maintenance records, and to generate an estimate of remaining life and recovery interval. The game decision-making module is used to establish an optimization decision-making model among multiple base stations based on the data provided by the load assessment module and the lifetime assessment module, and to generate shutdown and recovery plans that meet thermal safety and lifetime constraints. The linkage orchestration module is used to perform semantic verification and secure orchestration on the shutdown plan output by the game decision module, perform consistency verification and task scheduling based on the rule base, and generate securely executable control instructions. The anomaly verification module is used to monitor the system's operating status during the shutdown plan execution, identify anomalies in parameters such as temperature, energy consumption, and service quality, and perform automatic recovery based on the anomaly level. The audit display module is used to record system operation logs and energy efficiency data, perform encrypted evidence storage and reputation calculation, and generate audit reports and operational visualizations. Each module operates in a closed loop through data and control channels, realizing an energy-saving control process based on data perception, prediction and evaluation, optimization decision-making, safe execution, anomaly verification, and result auditing.

2. The energy-saving control system for remote intelligent hard shutdown of base station equipment room according to claim 1, characterized in that: The site group profiling module includes a data access layer, a data processing layer, a relationship analysis layer, a structure generation layer, and an output layer. The data access layer is used to collect base station power, service, and environmental data through the communication interface and perform time synchronization. The data processing layer is used to clean outliers and fill in missing data. The relationship analysis layer is used to calculate the similarity and correlation of each site. The structure generation layer is used to establish a site group relationship model. The output layer is used to generate a site group profiling result file.

3. The energy-saving control system for remote intelligent hard shutdown of base station equipment room according to claim 1, characterized in that: The load thermal assessment module includes a feature fusion layer, a prediction layer, a thermal analysis layer, and a security assessment layer. The feature fusion layer is used to integrate business data and external environmental parameters, the prediction layer is used to calculate future load trends, the thermal analysis layer is used to simulate temperature change processes, and the security assessment layer is used to determine the thermal safety level and shutdown time.

4. The energy-saving control system for remote intelligent hard shutdown of base station equipment room according to claim 1, characterized in that: The life assessment module includes a feature extraction layer, a life calculation layer, a history correction layer, and an update layer. The feature extraction layer is used to extract life-related features from power fluctuations, thermal cycling, and maintenance information. The life calculation layer is used to extrapolate the life distribution based on multidimensional data. The history correction layer is used to correct estimation errors. The update layer is used to adjust the life prediction results when the operating data changes.

5. The energy-saving control system for remote intelligent hard shutdown of base station equipment room according to claim 1, characterized in that: The game decision-making module includes a state modeling layer, an optimization calculation layer, and a strategy generation layer. The state modeling layer is used to describe the operating state of the base station group, the optimization calculation layer is used to calculate the coordinated shutdown scheme of each site, and the strategy generation layer is used to output energy-saving strategies that meet security and reliability constraints.

6. The energy-saving control system for remote intelligent hard shutdown of base station equipment room according to claim 1, characterized in that: The linkage orchestration module includes a rule engine layer, a verification layer, a scheduling layer, and a feedback layer. The rule engine layer is used to store security constraints, the verification layer is used to detect the logical consistency between control commands, the scheduling layer is used to adjust the execution order according to the risk level, and the feedback layer is used to record the execution results and update the rule base.

7. The energy-saving control system for remote intelligent hard shutdown of base station equipment room according to claim 1, characterized in that: The anomaly verification module includes a sampling layer, an analysis layer, an identification layer, and a recovery layer. The sampling layer is used to collect operating parameters in real time, the analysis layer is used to evaluate the stability of the operating state, the identification layer is used to detect the anomaly type and range, and the recovery layer is used to generate and execute adaptive recovery instructions.

8. The energy-saving control system for remote intelligent hard shutdown of base station equipment room according to claim 1, characterized in that: The audit display module includes a log recording layer, an evidence storage layer, a reputation assessment layer, and a display layer. The log recording layer is used to store operation logs, the evidence storage layer is used for data encryption and tamper-proof storage, the reputation assessment layer is used to calculate the operation reputation score, and the display layer is used to generate audit reports and graphical display interfaces.

9. An energy-saving control system for remote intelligent hard shutdown of a base station equipment room according to claim 1, characterized in that: The system has a periodic update mechanism. When a new base station is put into operation or an existing base station is decommissioned, the system automatically updates the base station group relationship matrix and base station group profile data to ensure that the model is synchronized with the actual operating status.

10. An energy-saving control system for remote intelligent hard shutdown of a base station equipment room according to claim 1, characterized in that: The system runs in a cloud or local server environment. The seven modules work together sequentially according to the data flow and control flow during system operation. The station group profiling module provides structural feature input, the load thermal assessment module outputs thermal safety boundaries, the life assessment module provides reliability constraints, the game decision module generates energy-saving strategies, the linkage orchestration module executes strategy verification and distribution, the anomaly verification module realizes operation monitoring and adaptive recovery, and the audit display module completes data storage and result display. The modules exchange data and provide status feedback through standard interfaces, forming a closed loop of intelligent energy-saving control that enables sustainable self-learning and parameter self-calibration.

Images