UPS power supply redundancy configuration and life optimization method based on dynamic load prediction

Abstract

The present application relates to the UPS power control technical field, specifically to the UPS power redundancy configuration and life optimization method based on dynamic load prediction, including the following steps: S1, multi-source state monitoring and data preprocessing;S2, dynamic load prediction and fluctuation feature extraction;S3, redundant capacity dynamic distribution and weight calculation;S4, life-aware load balancing adjustment;S5, adaptive redundancy triggering and switching control;S6, closed-loop verification and model iteration update.The present application can dynamically adjust load distribution according to the state of each module through the dynamic redundancy capacity distribution model fused with temperature and fault history, avoids local overload, and through the exponential temperature compensation function, effectively compensates the module performance degradation in high temperature environment, improves the temperature distribution uniformity among modules, and avoids the device aging acceleration caused by thermal stress concentration.

Description

Technical Field

[0001] This invention relates to the field of UPS power control technology, specifically to a method for UPS power redundancy configuration and lifespan optimization based on dynamic load prediction. Background Technology

[0002] Uninterruptible power supplies (UPS) are core devices for providing uninterrupted power to critical loads and are widely used in critical infrastructure fields such as data centers, industrial control, finance, and healthcare. To ensure power supply reliability, UPS systems typically adopt a parallel redundancy architecture, which achieves redundancy backup by configuring multiple UPS modules in parallel to avoid power outages caused by single points of failure.

[0003] However, traditional UPS redundancy configuration schemes have the following problems: First, traditional solutions often adopt a static N+X fixed redundancy architecture, and the load distribution adopts a static strategy with a fixed ratio. This is difficult to adapt to the dynamic load fluctuation characteristics in modern scenarios, resulting in some modules being overloaded for a long time and aging faster, while some modules are underloaded for a long time, causing resource waste and failing to fully utilize the resource potential of the redundant system. Secondly, traditional redundancy triggering mechanisms rely on manually preset fixed thresholds, lacking the ability to anticipate load fluctuations in advance. This can easily lead to untimely redundancy switching causing power outages, or premature switching causing unnecessary resource consumption. Furthermore, existing technologies lack the ability to perceive and balance the remaining lifespan of each UPS module. The significant differences in aging rates between modules limit the overall system lifespan to the module that ages first, failing to fully realize the lifespan potential of the redundant system. Simultaneously, the performance degradation of modules under high-temperature environments is not effectively compensated, further exacerbating the performance differences between modules. Therefore, we propose a UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction. Summary of the Invention

[0004] The purpose of this invention is to provide a method for UPS power supply redundancy configuration and lifespan optimization based on dynamic load prediction, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The UPS power supply redundancy configuration and lifetime optimization method based on dynamic load prediction includes the following steps: S1. Multi-source status monitoring and data preprocessing: Real-time acquisition of operating status data of each parallel UPS module, completion of data cleaning and normalization processing, and provision of standardized input for subsequent analysis; S2. Dynamic load forecasting and fluctuation feature extraction: Based on historical load time series data, a forecasting model is built to output the load forecast value within a preset time window in the future, and the first-order rate of change and second-order acceleration features of load fluctuation are extracted. S3. Dynamic allocation and weight calculation of redundant capacity: Based on the load prediction results, combined with the temperature and historical failure rate parameters of each UPS module, the real-time load allocation weight of each module is calculated to achieve dynamic adaptation of redundant capacity. S4. Lifetime-aware load balancing adjustment: Based on the remaining lifetime assessment results of each UPS module, the load allocation weight is adjusted to balance the aging rate of each module and achieve overall system lifetime optimization. S5. Adaptive Redundancy Triggering and Switching Control: Based on load fluctuation characteristics, the redundancy triggering threshold is dynamically adjusted. When the triggering conditions are met, the redundancy module with the highest health is activated first to complete the smooth switching of the load. S6. Closed-loop verification and model iterative update: After the redundancy switch is completed, closed-loop state verification is performed, and the parameters of the load prediction model are updated based on the feedback of actual operating data to achieve self-iterative optimization of the model.

[0006] Preferably, in step S1, multi-source status monitoring and data preprocessing, the process of real-time acquisition of the operating status data of each parallel UPS module and completion of data cleaning and normalization is as follows: The voltage and current sampling points of each UPS module are connected to the AD conversion module of the monitoring unit via double-shielded coaxial cables to collect the current load rate of each module in real time. Internal temperature Historical number of failures and cumulative running time ; Outlier removal was performed on the collected raw data. The 3σ criterion was used to remove abrupt sample values ​​caused by electromagnetic interference, and the moving average algorithm was used to eliminate high-frequency noise interference. The processed state data is normalized and mapped to the [0, 1] interval to eliminate the influence of different dimensional parameters on subsequent model calculations, thus obtaining standardized state input data.

[0007] Preferably, in step S2, dynamic load prediction and fluctuation feature extraction, the process of constructing a prediction model based on historical load time-series data, outputting the predicted load value within a preset future time window, and extracting the first-order rate of change and second-order acceleration features of load fluctuation is as follows: From history Load timing data within the time window Using this as input, a load forecasting model is constructed that integrates time-series trends and fluctuation characteristics to calculate future load forecasts. Load forecast within the time window ,satisfy: in, This represents the actual load value at the current moment. The first-order rate of load change, For the second-order acceleration due to load change, , These are the weighting coefficients used to adjust the contribution of first-order and second-order features to the predicted value, respectively. Calculate the statistical variance of load fluctuations based on load data within the sliding time window. This is used to characterize the degree of load fluctuation and to provide a basis for subsequent redundancy threshold adjustments.

[0008] Preferably, in step S3, the process of calculating the real-time load allocation weight of each UPS module based on the load prediction results and the temperature and historical failure rate parameters of each UPS module is as follows: Based on load forecast values By combining the status parameters of each UPS module, a dynamic redundancy capacity allocation model is constructed, and the initial allocation weight of each module is calculated. ,satisfy: in, The adjustment coefficient is dynamically adjusted based on the load changes and fault history of each module. This is a temperature compensation function used to compensate for the impact of temperature on module performance; The time decay factor, For duration parameter, For module Historical failure rate; The temperature compensation function satisfy: in, This is the temperature compensation gain coefficient. The temperature decay coefficient is This refers to the critical value of the rated operating temperature of the UPS module. For module The current operating temperature.

[0009] Preferably, in step S4, the lifespan-aware load balancing adjustment, the process of adjusting the load allocation weight based on the remaining lifespan assessment results of each UPS module to balance the aging rate of each module is as follows: Based on the Arrhenius aging model, and combining the current operating temperature and initial weights of each module, the remaining lifespan of each UPS module is evaluated. ,satisfy: in, This refers to the initial remaining lifespan of the module under rated operating conditions. For activation energy, The gas constant is The rated reference temperature, For module The current absolute temperature, To initially assign weights, This is an index of aging rate; The optimization objective is to minimize the variance of the remaining lifetime of each module, and the initial weights are assigned accordingly. Make corrections to obtain the corrected weights. This ensures that the aging rate of each module is consistent, preventing premature aging of some modules and optimizing the overall lifespan of the system.

[0010] Preferably, in step S5, the adaptive redundancy triggering and switching control, the process of dynamically adjusting the redundancy triggering threshold based on load fluctuation characteristics and prioritizing the activation of the redundancy module with the highest health status is as follows: Statistical variance based on load fluctuations Dynamically adjust the redundancy trigger threshold based on predicted load changes. ,satisfy: when At that time, the redundancy trigger threshold will be adjusted to: in, The threshold for judging abnormal fluctuations. The default trigger threshold is... To adjust the step size coefficient, This is a sign function used to characterize the direction of load change; When the trigger condition is met When this occurs, initiate the redundancy switchover procedure, prioritizing the activation of health indicators. The highest redundancy module, the health index satisfy: in, For module The percentage of remaining runtime. , , These are the weighting coefficients, corresponding to the weights of remaining lifespan, fault history, and temperature status, respectively.

[0011] Preferably, in step S6, closed-loop verification and model iterative update, the process of performing closed-loop state verification after redundancy switching and updating the parameters of the load prediction model based on actual operating data feedback is as follows: After the redundancy switch is completed, a test current pulse with an amplitude of 10%-15% of the rated current and a width of 2-5 power cycles is injected into the newly activated redundant module to verify the output stability and response speed of the module. The actual load data is compared with the previous forecast data to calculate the forecast error, and the weight coefficients of the load forecast model are updated based on the backpropagation of the error. , This enables the model to perform self-iterative optimization, thereby improving the accuracy of subsequent predictions.

[0012] Compared with the prior art, the beneficial effects of the present invention are: This invention, through dynamic load prediction and fluctuation feature extraction, can predict load change trends in advance, providing accurate input for redundant configuration. It solves the problem that traditional static configuration cannot adapt to dynamic loads and improves the system's adaptability to load fluctuations.

[0013] This invention utilizes a dynamic redundancy capacity allocation model that integrates temperature and fault history to dynamically adjust load distribution based on the status of each module, avoiding local overload. Simultaneously, through an exponential temperature compensation function, it effectively compensates for module performance degradation under high-temperature environments, improves the uniformity of temperature distribution among modules, and avoids accelerated device aging caused by thermal stress concentration.

[0014] This invention uses lifetime-aware load balancing adjustment to assess the remaining lifetime of each module based on the Arrhenius model and adjusts the load distribution with the goal of minimizing the remaining lifetime variance. This effectively balances the aging rate of each module, avoids the system lifetime being limited by the module that ages first, significantly extends the overall lifespan of the system, and fully utilizes the lifetime potential of redundant systems.

[0015] This invention, through adaptive redundancy trigger threshold adjustment, can dynamically adjust the triggering conditions according to the statistical characteristics of load fluctuations. It can activate redundant modules in advance during the rapid load increase phase, avoiding power outages caused by untimely switching, and at the same time, avoid unnecessary switching during the stable load phase, thereby improving the reliability and resource utilization of the system.

[0016] This invention, through closed-loop verification and model iteration, can verify the status of redundant modules after switching, and continuously optimize the parameters of the prediction model based on actual operating data, thereby achieving self-learning and self-optimization of the system and improving the long-term adaptability and prediction accuracy of the system.

[0017] This invention employs a dual-shielded coaxial cable and fiber optic channel to separately transmit monitoring data and control signals, effectively suppressing interference in complex electromagnetic environments, improving the accuracy of status monitoring, and ensuring the stable operation of the system. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0019] 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.

[0020] like Figure 1 As shown, the UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction includes the following steps: S1. Multi-source status monitoring and data preprocessing: The process of real-time acquisition of operating status data from each parallel UPS module, data cleaning and normalization, and providing standardized input for subsequent analysis is as follows: The voltage and current sampling points of each UPS module are connected to the AD conversion module of the monitoring unit via double-shielded coaxial cables to collect the current load rate of each module in real time. Internal temperature Historical number of failures and cumulative running time ; Outlier removal was performed on the collected raw data. The 3σ criterion was used to remove abrupt sample values ​​caused by electromagnetic interference, and the moving average algorithm was used to eliminate high-frequency noise interference. The processed state data is normalized and mapped to the [0, 1] interval to eliminate the influence of different dimensional parameters on subsequent model calculations, thus obtaining standardized state input data.

[0021] S2. Dynamic Load Forecasting and Fluctuation Feature Extraction: The process of constructing a forecasting model based on historical load time-series data, outputting the predicted load value within a preset future time window, and extracting the first-order rate of change and second-order acceleration features of load fluctuations is as follows: From history Load timing data within the time window Using this as input, a load forecasting model is constructed that integrates time-series trends and fluctuation characteristics to calculate future load forecasts. Load forecast within the time window ,satisfy: in, This represents the actual load value at the current moment. The first-order rate of load change, For the second-order acceleration due to load change, , These are the weighting coefficients used to adjust the contribution of first-order and second-order features to the predicted value, respectively. Calculate the statistical variance of load fluctuations based on load data within the sliding time window. This is used to characterize the degree of load fluctuation and to provide a basis for subsequent redundancy threshold adjustments.

[0022] S3. Dynamic Allocation and Weight Calculation of Redundant Capacity: Based on load forecasting results and combined with the temperature and historical failure rate parameters of each UPS module, the real-time load allocation weight of each module is calculated to achieve dynamic adaptation of redundant capacity. Based on load forecast values By combining the status parameters of each UPS module, a dynamic redundancy capacity allocation model is constructed, and the initial allocation weight of each module is calculated. ,satisfy: in, The adjustment coefficient is dynamically adjusted based on the load changes and fault history of each module. This is a temperature compensation function used to compensate for the impact of temperature on module performance; The time decay factor, For duration parameter, For module Historical failure rate; The temperature compensation function satisfy: in, This is the temperature compensation gain coefficient. The temperature decay coefficient is This refers to the critical value of the rated operating temperature of the UPS module. For module The current operating temperature.

[0023] S4. Lifetime-Aware Load Balancing Adjustment: Based on the remaining lifetime assessment results of each UPS module, the load allocation weight is adjusted to balance the aging rate of each module, achieving overall system lifetime optimization. Based on the Arrhenius aging model, and combining the current operating temperature and initial weights of each module, the remaining lifespan of each UPS module is evaluated. ,satisfy: in, This refers to the initial remaining lifespan of the module under rated operating conditions. For activation energy, The gas constant is The rated reference temperature, For module The current absolute temperature, To initially assign weights, This is an index of aging rate; The optimization objective is to minimize the variance of the remaining lifetime of each module, and the initial weights are assigned accordingly. Make corrections to obtain the corrected weights. This ensures that the aging rate of each module is consistent, preventing premature aging of some modules and optimizing the overall lifespan of the system.

[0024] S5. Adaptive Redundancy Triggering and Switching Control: Based on load fluctuation characteristics, the redundancy triggering threshold is dynamically adjusted. When the triggering conditions are met, the redundant module with the highest health is activated first, completing the smooth load switching process as follows: Statistical variance based on load fluctuations Dynamically adjust the redundancy trigger threshold based on predicted load changes. ,satisfy: when At that time, the redundancy trigger threshold will be adjusted to: in, The threshold for judging abnormal fluctuations. The default trigger threshold is... To adjust the step size coefficient, This is a sign function used to characterize the direction of load change; When the trigger condition is met When this occurs, initiate the redundancy switchover procedure, prioritizing the activation of health indicators. The highest redundancy module, the health index satisfy: in, For module The percentage of remaining runtime. , , These are the weighting coefficients, corresponding to the weights of remaining lifespan, fault history, and temperature status, respectively.

[0025] S6. Closed-loop verification and model iterative update: After the redundancy switch is completed, closed-loop state verification is performed, and the parameters of the load prediction model are updated based on the feedback of actual operating data. The process of realizing the self-iterative optimization of the model is as follows: After the redundancy switch is completed, a test current pulse with an amplitude of 10%-15% of the rated current and a width of 2-5 power cycles is injected into the newly activated redundant module to verify the output stability and response speed of the module. The actual load data is compared with the previous forecast data to calculate the forecast error, and the weight coefficients of the load forecast model are updated based on the backpropagation of the error. , This enables the model to perform self-iterative optimization, thereby improving the accuracy of subsequent predictions. Specific Implementation Example 1: This invention relates to a UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction. It is applied to a parallel redundant system comprising six 20kW modular UPS units, which provides uninterrupted power to the core load of a data center. The specific execution process is as follows: First, multi-source status monitoring and data preprocessing steps are performed: The system connects the sampling points of each UPS module via double-shielded coaxial cables to collect real-time data on the load rate, internal temperature, historical failure count, and cumulative operating time of the six modules. For example, the system collects data showing that at the current moment, the load rate of a certain module is 72%, the temperature is 32℃, the historical failure count is 1, and the cumulative operating time is 12,000 hours. The raw data is then cleaned, using the 3σ criterion to remove abrupt changes caused by electromagnetic interference, such as an outlier where the load rate of a sampling point suddenly jumps to 150%. High-frequency noise is eliminated using a 5-point moving average algorithm. Finally, all data is normalized to the [0, 1] interval to obtain standardized input data.

[0027] The next step is to perform dynamic load forecasting and fluctuation feature extraction: The system uses historical load time-series data from the past hour as input to calculate the first-order rate of change and second-order acceleration of the load, and substitutes these values ​​into the forecast model to calculate the load forecast for the next 5 minutes. For example, if the current load is 85kW, the first-order rate of change is 2kW / min, and the second-order acceleration is 0.5kW / min... 2 Substituting these values ​​into the formula, the predicted load for the next 5 minutes is 96kW. Simultaneously, the variance of the load fluctuation over the past 10 minutes is calculated, yielding... =0.82, indicating that the current load is in a steady upward phase.

[0028] The following steps involve dynamic allocation and weight calculation of redundant capacity: Based on the predicted load value, combined with the temperature and fault history of each module, the initial load allocation weight is calculated. For example, for a module with a temperature of 38℃, substituting the temperature compensation function yields ω=1.08, indicating that the module's temperature is slightly high and its allocation weight needs to be appropriately reduced. For a module that has experienced one fault, substituting the fault attenuation term, its weight is appropriately reduced to avoid the faulty module bearing too much load. Finally, the initial allocation weights for each module are obtained, ensuring that the total load is reasonably distributed among the modules.

[0029] The system then performs a lifetime-aware load balancing adjustment step: based on the Arrhenius model, it calculates the remaining lifetime of each module. For example, a module that has been running for 12,000 hours and currently has a high load rate is calculated to have a remaining lifetime of 18,000 hours; while another module has been running for a shorter time and has a lower load rate, with a remaining lifetime of 32,000 hours. The system aims to minimize the variance of the remaining lifetime by adjusting the initial weights, appropriately reducing the weight of the high-load module and increasing the weight of the low-load module, so that the remaining lifetimes of the two modules gradually become consistent, balancing the aging rate and preventing the former from aging prematurely.

[0030] Next, the adaptive redundancy triggering and switching control steps are executed: The system dynamically adjusts the redundancy triggering threshold based on the variance of load fluctuations. Since the current load fluctuation is relatively small and the predicted load is in an upward phase, the triggering threshold is adjusted from the default 0.8 to 0.75, triggering redundancy activation ahead of schedule. When the total load reaches the threshold, the system calculates the health of each redundant module, prioritizing the activation of modules with longer remaining lifespans, fewer faults, and lower temperatures to complete a smooth load switch and ensure continuous power supply.

[0031] Finally, closed-loop verification and model iterative update steps are performed: After the switch is completed, the system injects a test current pulse into the newly activated module to verify whether its output is stable and confirm that the module is working normally. Then, the actual load value is compared with the previous predicted value, the prediction error is calculated, and the prediction model is updated via backpropagation. and Parameters are used to optimize the model's prediction accuracy, providing a more accurate model for the next prediction.

[0032] Thus, the entire method completes the entire process from condition monitoring to load prediction, redundancy configuration, lifespan optimization, switching control and model iteration, realizing dynamic adaptive adjustment of UPS redundancy configuration, effectively extending the overall lifespan of the system and improving the reliability of power supply.

[0033] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for UPS power supply redundancy configuration and lifespan optimization based on dynamic load prediction, characterized in that, Includes the following steps: S1. Multi-source status monitoring and data preprocessing: Real-time acquisition of operating status data of each parallel UPS module, completion of data cleaning and normalization processing, and provision of standardized input for subsequent analysis; S2. Dynamic load forecasting and fluctuation feature extraction: Based on historical load time series data, a forecasting model is built to output the load forecast value within a preset time window in the future, and the first-order rate of change and second-order acceleration features of load fluctuation are extracted. S3. Dynamic allocation and weight calculation of redundant capacity: Based on the load prediction results, combined with the temperature and historical failure rate parameters of each UPS module, the real-time load allocation weight of each module is calculated to achieve dynamic adaptation of redundant capacity. S4. Lifetime-aware load balancing adjustment: Based on the remaining lifetime assessment results of each UPS module, the load allocation weight is adjusted to balance the aging rate of each module and achieve overall system lifetime optimization. S5. Adaptive Redundancy Triggering and Switching Control: Based on load fluctuation characteristics, the redundancy triggering threshold is dynamically adjusted. When the triggering conditions are met, the redundancy module with the highest health is activated first to complete the smooth switching of the load. S6. Closed-loop verification and model iterative update: After the redundancy switch is completed, closed-loop state verification is performed, and the parameters of the load prediction model are updated based on the feedback of actual operating data to achieve self-iterative optimization of the model.

2. The UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction according to claim 1, characterized in that, In step S1, multi-source status monitoring and data preprocessing, the process of real-time acquisition of the operating status data of each parallel UPS module and completion of data cleaning and normalization is as follows: The voltage and current sampling points of each UPS module are connected to the AD conversion module of the monitoring unit via double-shielded coaxial cables to collect the current load rate of each module in real time. Internal temperature Historical number of failures and cumulative operating time ; Outlier removal was performed on the collected raw data. The 3σ criterion was used to remove abrupt sample values ​​caused by electromagnetic interference, and the moving average algorithm was used to eliminate high-frequency noise interference. The processed state data is normalized and mapped to the [0, 1] interval to eliminate the influence of different dimensional parameters on subsequent model calculations, thus obtaining standardized state input data.

3. The UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction according to claim 1, characterized in that, In step S2, dynamic load prediction and fluctuation feature extraction, the process of constructing a prediction model based on historical load time series data, outputting the predicted load value within a preset future time window, and extracting the first-order rate of change and second-order acceleration features of load fluctuation is as follows: From history Load timing data within the time window Using this as input, a load forecasting model is constructed that integrates time-series trends and fluctuation characteristics to calculate future load forecasts. Load forecast within the time window ,satisfy: ,in, This represents the actual load value at the current moment. The first-order rate of load change, For the second-order acceleration due to load change, , The weighting coefficients are used to adjust the contributions of first-order and second-order features to the predicted values, respectively; the statistical variance of load fluctuations is calculated based on the load data within the sliding time window. This is used to characterize the degree of load fluctuation and to provide a basis for subsequent redundancy threshold adjustments.

4. The UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction according to claim 1, characterized in that, In step S3, dynamic allocation and weight calculation of redundant capacity, the process of calculating the real-time load allocation weight of each UPS module based on the load prediction results and the temperature and historical failure rate parameters of each UPS module is as follows: Based on load forecast values By combining the status parameters of each UPS module, a dynamic redundancy capacity allocation model is constructed, and the initial allocation weight of each module is calculated. ,satisfy: ,in, The adjustment coefficient is dynamically adjusted based on the load changes and fault history of each module. This is a temperature compensation function used to compensate for the impact of temperature on module performance; The time decay factor, For duration parameter, The module's historical failure rate; The temperature compensation function satisfy: ,in, This is the temperature compensation gain coefficient. The temperature decay coefficient is This refers to the critical value of the rated operating temperature of the UPS module. This is the current operating temperature of the module.

5. The UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction according to claim 1, characterized in that, In step S4, the lifespan-aware load balancing adjustment, the process of adjusting the load allocation weight based on the remaining lifespan assessment results of each UPS module to balance the aging rate of each module is as follows: Based on the Arrhenius aging model, combined with the current operating temperature of each module and the initial allocation weight, the remaining lifespan of each UPS module is assessed. ,satisfy: ,in, This refers to the initial remaining lifespan of the module under rated operating conditions. For activation energy, The gas constant is The rated reference temperature, This represents the current absolute temperature of the module. To initially assign weights, This is an index of aging rate; The optimization objective is to minimize the variance of the remaining lifetime of each module, and the initial weights are assigned accordingly. Make corrections to obtain the corrected weights. This ensures that the aging rate of each module is consistent, preventing premature aging of some modules and optimizing the overall lifespan of the system.

6. The UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction according to claim 1, characterized in that, In step S5, adaptive redundancy triggering and switching control, the process of dynamically adjusting the redundancy triggering threshold based on load fluctuation characteristics and prioritizing the activation of the redundancy module with the highest health is as follows: based on the statistical variance of load fluctuations... Dynamically adjust the redundancy trigger threshold based on predicted load changes. , satisfying: when When the redundancy trigger threshold is adjusted to ,in, The threshold for judging abnormal fluctuations. The default trigger threshold, To adjust the step size coefficient, This is a symbolic function used to characterize the direction of load changes; when the triggering condition is met, the redundancy switching procedure is initiated, prioritizing the activation of health indicators. The highest redundancy module, the health index satisfy: ,in, This represents the percentage of the module's remaining runtime. , , These are the weighting coefficients, corresponding to the weights of remaining lifespan, fault history, and temperature status, respectively.

7. The UPS power supply redundancy configuration and lifespan optimization method based on dynamic load prediction according to claim 1, characterized in that, In step S6, closed-loop verification and model iterative update, the process of performing closed-loop state verification after redundancy switching and updating the parameters of the load prediction model based on actual operating data feedback is as follows: After the redundancy switch is completed, a test current pulse with an amplitude of 10%-15% of the rated current and a width of 2-5 power cycles is injected into the newly activated redundant module to verify the output stability and response speed of the module. The actual load data is compared with the previous forecast data to calculate the forecast error, and the weight coefficients of the load forecast model are updated based on the backpropagation of the error. , This enables the model to perform self-iterative optimization, thereby improving the accuracy of subsequent predictions.

Images