Intelligent communication power supply management method and system based on reliability
By using a multi-task LSTM model with attention enhancement mechanism and generative adversarial network, combined with reinforcement learning algorithm, a fault prediction model is constructed to generate load balancing and charging/discharging strategies. This solves the reliability and efficiency problems of intelligent communication power supplies in complex environments, and improves power supply reliability and communication system stability.
Patent Information
- Application Number
- CN202511258766.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-04
- Publication Date
- 2025-12-16
AI Technical Summary
Existing power management technologies are insufficient to meet the reliability and operational efficiency requirements of communication networks for intelligent communication power supplies, especially in complex operating environments where they struggle to capture multi-source, highly time-sequential time dependencies and sudden anomalies.
A multi-task LSTM model structure with attention enhancement mechanism is adopted, combined with generative adversarial networks and reinforcement learning algorithms, to construct a fault prediction model, generate load balancing and charging/discharging strategies, and optimize the management of intelligent communication power supply.
It improves the power supply reliability and communication system stability of distributed intelligent power sources in complex environments, reduces the false alarm rate of fault prediction, and avoids communication interruptions caused by unplanned shutdowns.
Smart Images
Figure CN121144809A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power equipment management technology, and in particular to a reliability-based intelligent communication power management method and system. Background Technology
[0002] Intelligent communication power supplies are the energy core of communication systems. Their stable operation directly determines the continuity and reliability of communication networks. Once the power supply fails, it may lead to major losses such as communication interruption and data loss. Therefore, the industry has put forward extremely high requirements for its reliability, fault early warning capability and energy efficiency management.
[0003] With the expansion of communication services, the operating environment of intelligent communication power supplies is becoming more complex, and the operational data they generate exhibits characteristics of being multi-sourced and highly time-series-dependent. However, existing power management technologies have significant shortcomings; traditional fault prediction relies on fixed threshold judgments, which is insufficient to meet the requirements of communication networks for power reliability and operational efficiency. Summary of the Invention
[0004] The present invention aims to provide a reliability-based intelligent communication power management method and system, which can capture the complex timing dependencies in power supply operation and improve the power supply reliability and communication system stability of distributed intelligent power supply in complex operating environments.
[0005] In a first aspect, embodiments of the present invention provide a reliability-based intelligent communication power management method, comprising:
[0006] Acquire historical operating data of the intelligent communication power supply; the historical operating data includes historical fault operating data and historical normal operating data;
[0007] Based on a multi-task LSTM model structure with attention enhancement mechanism, the power supply capability of intelligent communication power supply is evaluated, and a fault prediction model is constructed.
[0008] The fault prediction model is trained using the historical operating data.
[0009] The real-time operating data of the intelligent communication power supply is input into the trained fault prediction model to obtain the fault prediction result;
[0010] Based on the fault prediction results and the load data of the intelligent communication power supply, an optimized management strategy for the intelligent communication power supply is generated; the optimized management strategy includes a load balancing strategy and a charging / discharging strategy.
[0011] As an improvement to the above solution, the acquisition of historical operating data of the intelligent communication power supply includes:
[0012] Acquire historical operating data of the intelligent communication power supply under preset dimensions; the preset dimensions include the power supply's internal electrical parameters, environmental parameters, and battery performance parameters;
[0013] The historical operating data is labeled according to its operating status; the operating status includes normal status and fault status.
[0014] As an improvement to the above scheme, the evaluation of the power supply capability of the intelligent communication power supply based on the multi-task LSTM model structure with attention enhancement mechanism, and the construction of a fault prediction model based on the power supply capability, includes:
[0015] Based on the initial preset number of layers, an LSTM backbone network including a shared hidden layer is constructed to extract general temporal features from the input data;
[0016] Extract current fluctuation features from the input data, and input the current fluctuation features and the general time series features together into the attention mechanism layer to calculate the attention enhancement features at each time step;
[0017] Based on the dimensions of the input data, a parallel multi-task branch architecture is constructed. In each task branch, based on its data dimensions and the attention enhancement features, regression analysis is used to output the predicted power supply capability of the intelligent communication power supply.
[0018] A fault prediction model is constructed by comparing the predicted power supply capacity with the actual power supply capacity.
[0019] As an improvement to the above solution, the step of training the fault prediction model using the historical operating data includes:
[0020] Generative adversarial networks are used to enhance the historical operation data to obtain enhanced historical operation data;
[0021] The enhanced historical operational data is divided into a training set and a test set;
[0022] The fault prediction model is trained using the training set through a minimum loss function optimization algorithm.
[0023] The training results of the fault prediction model are verified using the test set. Based on the verification results, the parameters of the fault prediction model are adjusted and retrained until the prediction accuracy requirements are met.
[0024] As an improvement to the above scheme, the step of training the fault prediction model using the training set through a minimum loss function optimization algorithm includes:
[0025] pass The loss function is obtained; where y iLabeling for actual faults. The fault predicted by the fault prediction model;
[0026] The fault prediction model is trained using the training set according to the loss function.
[0027] As an improvement to the above solution, the step of generating an optimized management strategy for the intelligent communication power supply based on the fault prediction results and the load data of the intelligent communication power supply includes:
[0028] Based on the load data of the intelligent communication power supply, a reinforcement learning algorithm is used to dynamically adjust the load distribution at each time step to obtain a load balancing strategy.
[0029] Based on the load balancing strategy and the fault prediction results, the charging, discharging and hibernation simulations of the intelligent communication power supply are performed to obtain the charging and discharging strategy.
[0030] Based on the load balancing strategy and the charging / discharging strategy, an optimized management strategy for the intelligent power supply is obtained.
[0031] As an improvement to the above scheme, the step of dynamically adjusting the load distribution at each time step based on the load data of the intelligent communication power supply using a reinforcement learning algorithm to obtain a load balancing strategy includes:
[0032] Based on the load data of the intelligent communication power supply, the degree of deviation between the load of each component of the intelligent communication power supply and the average load is calculated to obtain the load balance.
[0033] Based on the fault prediction results, the system reliability of the intelligent communication power supply is calculated;
[0034] Obtain the input and output power of the intelligent communication power supply, and calculate the energy efficiency based on the input and output power;
[0035] The reward function of the reinforcement learning algorithm is obtained based on the load balancing degree, the system reliability, and the energy efficiency.
[0036] Based on the reward function, the load distribution at each time step is dynamically adjusted to obtain the load balancing strategy.
[0037] As an improvement to the above scheme, the step of simulating charging, discharging, and hibernation of the intelligent communication power supply based on the load balancing strategy and the fault prediction results to obtain a charging and discharging strategy includes:
[0038] Based on the load balancing strategy and the fault prediction results, the predicted internal electrical parameters and predicted battery performance parameters of the intelligent communication power supply are obtained.
[0039] Based on the predicted power supply's internal electrical parameters and predicted battery performance parameters, constant voltage and constant current charging simulation, discharging simulation, and low-load sleep simulation are performed on the components in the intelligent communication power supply. A decision tree based on information gain is used to obtain the charging and discharging strategy.
[0040] As an improvement to the above solution, after generating the optimized management strategy for the intelligent communication power supply based on the fault prediction results and the load data of the intelligent communication power supply, the reliability-based intelligent communication power supply management method further includes:
[0041] The reliability of the intelligent communication power supply is evaluated within a preset period.
[0042] The algorithm for generating the optimization management strategy is optimized based on the reliability.
[0043] Secondly, embodiments of the present invention also provide a reliability-based intelligent communication power management system, comprising:
[0044] The data acquisition module is used to acquire historical operating data of the intelligent communication power supply; the historical operating data includes historical fault operating data and historical normal operating data.
[0045] The model building module is used to evaluate the power supply capability of intelligent communication power supplies and build a fault prediction model based on a multi-task LSTM model structure with attention enhancement mechanism.
[0046] The model training module is used to train the fault prediction model using the historical operating data;
[0047] The model prediction module is used to input the real-time operating data of the intelligent communication power supply into the trained fault prediction model to obtain the fault prediction result;
[0048] The power management module is used to generate an optimized management strategy for the intelligent communication power supply based on the fault prediction results and the load data of the intelligent communication power supply; the optimized management strategy includes a load balancing strategy and a charging / discharging strategy.
[0049] Compared with existing technologies, this invention discloses a reliability-based intelligent communication power management method and system. This method acquires historical operating data of the intelligent communication power supply, including historical fault operation data and historical normal operation data. Based on a multi-task LSTM model structure employing an attention enhancement mechanism, the power supply capability of the intelligent communication power supply is evaluated, and a fault prediction model is constructed. The fault prediction model is trained using the historical operating data. Real-time operating data of the intelligent communication power supply is input into the trained fault prediction model to obtain fault prediction results. Based on the fault prediction results and the load data of the intelligent communication power supply, an optimized management strategy for the intelligent communication power supply is generated. The optimized management strategy includes a load balancing strategy and a charging / discharging strategy. Using this invention, complex temporal dependencies in power supply operation can be captured, improving the power supply reliability of distributed intelligent power supplies and the stability of communication systems in complex operating environments. Attached Figure Description
[0050] Figure 1 This is a flowchart illustrating the steps of a reliability-based intelligent communication power management method provided in an embodiment of the present invention.
[0051] Figure 2 This is a schematic diagram of a reliability-based intelligent communication power management system provided in an embodiment of the present invention. Detailed Implementation
[0052] 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.
[0053] In the description and claims, it should be understood that the terms "first," "second," etc., used in the description and claims are only for the purpose of distinguishing the description of the same technical features, and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated, nor necessarily the order of description or chronological order. The terms are interchangeable where appropriate. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature.
[0054] This invention provides a reliability-based intelligent communication power management method. Please refer to [link / reference]. Figure 1 In this embodiment, the reliability-based intelligent communication power management method is specifically executed through steps S1 to S5:
[0055] S1. Obtain historical operating data of the intelligent communication power supply; the historical operating data includes historical fault operating data and historical normal operating data.
[0056] It should be noted that, in this embodiment of the invention, the historical operation data includes its time attribute, which facilitates subsequent analysis of the historical operation data based on time series.
[0057] S2. Based on the multi-task LSTM model structure with attention enhancement mechanism, the power supply capability of intelligent communication power supply is evaluated, and a fault prediction model is constructed.
[0058] Unlike traditional LSTM structures, this invention employs a multi-task LSTM network structure. By predicting multiple key power supply indicators, it fully assesses power supply capabilities and further generates a fault prediction model. This addresses the shortcomings of traditional methods in capturing complex temporal dependencies and sudden anomalies, and improves the ability to model multivariate coupling relationships.
[0059] S3. The fault prediction model is trained using the historical operating data.
[0060] S4. Input the real-time operating data of the intelligent communication power supply into the trained fault prediction model to obtain the fault prediction result.
[0061] S5. Based on the fault prediction results and the load data of the intelligent communication power supply, generate an optimized management strategy for the intelligent communication power supply; the optimized management strategy includes a load balancing strategy and a charging / discharging strategy.
[0062] In this embodiment of the invention, a multi-source, highly time-series intelligent communication power supply is analyzed using a multi-task LSTM model structure. At the same time, an attention mechanism is used to focus on key time-series characteristics, making the power supply capacity assessment closer to the actual operating characteristics and further reducing the false alarm rate of fault prediction. In addition, an optimization management strategy, including load balancing and charging / discharging strategies, is used to pre-optimize and adjust the intelligent communication power supply dynamically, avoiding communication losses caused by unplanned downtime and ensuring the continuity of the communication system.
[0063] As a preferred implementation, step S1, acquiring historical operating data of the intelligent communication power supply, includes:
[0064] Acquire historical operating data of the intelligent communication power supply under preset dimensions; the preset dimensions include the power supply's internal electrical parameters, environmental parameters, and battery performance parameters;
[0065] The historical operating data is labeled according to its operating status; the operating status includes normal status and fault status.
[0066] In some preferred embodiments, the internal electrical parameters of the power supply include voltage, current, and power, the environmental parameters include temperature and humidity, and the battery performance parameters include SOC and remaining capacity.
[0067] As a preferred implementation, step S2, based on a multi-task LSTM model structure employing an attention enhancement mechanism, evaluates the power supply capability of the intelligent communication power supply and constructs a fault prediction model based on the power supply capability, including:
[0068] Based on the initial preset number of layers, an LSTM backbone network including a shared hidden layer is constructed to extract general temporal features from the input data;
[0069] Extract current fluctuation features from the input data, and input the current fluctuation features and the general time series features together into the attention mechanism layer to calculate the attention enhancement features at each time step;
[0070] Based on the dimensions of the input data, a parallel multi-task branch architecture is constructed. In each task branch, based on its data dimensions and the attention enhancement features, regression analysis is used to output the predicted power supply capability of the intelligent communication power supply.
[0071] A fault prediction model is constructed by comparing the predicted power supply capacity with the actual power supply capacity.
[0072] To enhance the response to power fluctuations and sudden load events, in this embodiment of the invention, the model introduces an attention mechanism to dynamically focus on input features at key moments, thereby improving prediction accuracy.
[0073] In some preferred embodiments, the functional representation of the LSTM backbone network is as follows:
[0074]
[0075] in, The predicted value of the general temporal features at the next time step; f θ Let x be the nonlinear mapping function parameterized by LSTM, and θ be the parameter set of the LSTM backbone network; t This refers to the general timing characteristics of the input collected over the past n time points. For example, these general timing characteristics include current, voltage, temperature, output load current, and battery status.
[0076] It should be noted that, more preferably, in the above embodiment, the current fluctuation features are first extracted from the input data, and then the current fluctuation features are coupled with the general time series features obtained by the LSTM backbone network for attention focusing, thereby enhancing the model's ability to respond to sudden current fluctuations.
[0077] It should also be noted that, taking the above-mentioned internal electrical parameters of the power supply as an example, there is a correlation between the voltage and current parameters. When constructing a multi-task branch architecture, the internal electrical parameters of the power supply can be predicted as a whole as a task branch, or a task branch can be constructed and predicted separately for each specific parameter.
[0078] In some preferred embodiments, a loss function as described below is constructed to compare the predicted power supply capacity with the actual power supply capacity, in order to build a fault prediction model:
[0079]
[0080] in, P represents the predicted power supply; max R is the maximum power that the system can currently provide; bat λ1 represents the percentage of remaining battery capacity; λ2 and λ1 are preset weighting coefficients.
[0081] Furthermore, preferably, by means of an adaptive time window mechanism, the length of the input sequence of the LSTM backbone network is automatically adjusted based on the load change rate, so as to achieve the effect of balancing the responsiveness and stability of the LSTM model.
[0082] As a preferred implementation, step S3, training the fault prediction model using the historical operating data, includes:
[0083] Generative adversarial networks are used to enhance the historical operation data to obtain enhanced historical operation data;
[0084] The enhanced historical operational data is divided into a training set and a test set;
[0085] The fault prediction model is trained using the training set through a minimum loss function optimization algorithm.
[0086] The training results of the fault prediction model are verified using the test set. Based on the verification results, the parameters of the fault prediction model are adjusted and retrained until the prediction accuracy requirements are met.
[0087] In the above scheme, to address the problem of scarce samples under abnormal operating conditions, a data augmentation method based on generative adversarial networks is adopted to expand the model training samples and improve the model's generalization ability.
[0088] Further, preferably, the step of training the fault prediction model using the training set through a minimum loss function optimization algorithm includes:
[0089] pass The loss function is obtained; where y i Labeling for actual faults. The fault predicted by the fault prediction model;
[0090] The fault prediction model is trained using the training set according to the loss function.
[0091] The aforementioned loss function enables the fault prediction model to automatically learn the complex nonlinear relationship between faults and parameters, rather than relying on simple rules summarized manually, thus improving prediction accuracy.
[0092] In a preferred implementation, step S5 involves generating an optimized management strategy for the intelligent communication power supply based on the fault prediction results and the load data of the intelligent communication power supply, and executing this strategy through steps S51-S53.
[0093] S51. Based on the load data of the intelligent communication power supply, a reinforcement learning algorithm is used to dynamically adjust the load distribution at each time step to obtain a load balancing strategy.
[0094] S52. Based on the load balancing strategy and the fault prediction results, the intelligent communication power supply is subjected to charging, discharging and hibernation simulations to obtain a charging and discharging strategy.
[0095] S53. Based on the load balancing strategy and the charging / discharging strategy, an optimized management strategy for the intelligent power supply is obtained.
[0096] In the above scheme, load balancing strategy and charging / discharging strategy are generated based on fault prediction results, and a synchronous adaptation scheme for load balancing strategy and charging / discharging strategy is given, realizing the active optimization management of intelligent power supply.
[0097] Further, preferably, step S51 involves using a reinforcement learning algorithm to dynamically adjust the load distribution at each time step based on the load data of the intelligent communication power supply, thereby obtaining a load balancing strategy, including:
[0098] Based on the load data of the intelligent communication power supply, the degree of deviation between the load of each component of the intelligent communication power supply and the average load is calculated to obtain the load balance.
[0099] Based on the fault prediction results, the system reliability of the intelligent communication power supply is calculated;
[0100] Obtain the input and output power of the intelligent communication power supply, and calculate the energy efficiency based on the input and output power;
[0101] The reward function of the reinforcement learning algorithm is obtained based on the load balancing degree, the system reliability, and the energy efficiency.
[0102] Based on the reward function, the load distribution at each time step is dynamically adjusted to obtain the load balancing strategy.
[0103] In some preferred embodiments, the reward function of the reinforcement learning algorithm is obtained by weighting the load balance, the system reliability, and the energy efficiency according to preset weight coefficients.
[0104] The agent in the reinforcement learning algorithm selects actions based on the state of the intelligent communication power supply and receives corresponding rewards, learning the optimal strategy through continuous trial and error. It should be noted that in this embodiment of the invention, the state set of the reinforcement learning algorithm is obtained through prediction results and actual load data, and the action set represents the scheduling actions of each component of the intelligent communication power supply, such as disconnecting / connecting a certain output or changing the output voltage of a certain DC / DC module. The load balancing strategy can be represented as π(a|s), where a is the action to be executed and s is the current state.
[0105] Based on the above preferred embodiments, the load balancing strategy is iterated through π. * =argmax E s～S,a～π [r(s,a)] is executed, where S is the state space, π is the action space, and r(s,a) is the reward function.
[0106] Preferably, step S52, which involves simulating charging, discharging, and hibernation of the intelligent communication power supply based on the load balancing strategy and the fault prediction result to obtain a charging and discharging strategy, includes:
[0107] Based on the load balancing strategy and the fault prediction results, the predicted internal electrical parameters and predicted battery performance parameters of the intelligent communication power supply are obtained.
[0108] Based on the predicted power supply's internal electrical parameters and predicted battery performance parameters, constant voltage and constant current charging simulation, discharging simulation, and low-load sleep simulation are performed on the components in the intelligent communication power supply. A decision tree based on information gain is used to obtain the charging and discharging strategy.
[0109] It should be noted that, in the embodiments of the present invention, the charging and discharging strategy includes a charging strategy, a discharging strategy, and a hibernation strategy.
[0110] Preferably, the charging current and voltage are dynamically adjusted according to the battery's state of charge and health, and a constant current and constant voltage charging method is adopted to obtain the charging strategy.
[0111] In some preferred embodiments, the formulas for calculating the charging current and voltage are as follows:
[0112]
[0113] Among them, P bat Battery power; V batΔV is the battery voltage; k1 and k2 are preset adjustment coefficients.
[0114] Preferably, the discharge strategy is obtained by reasonably controlling the depth of discharge and the discharge rate according to the load requirements and the remaining battery power.
[0115] In some preferred embodiments, the formula for calculating the discharge cutoff voltage is:
[0116] V discharge_end =V bat_min +k3·SOC;
[0117] Among them, V bat_min is the minimum allowable voltage of the battery; SOC is the state of charge of the battery; k3 is the preset correction coefficient.
[0118] Preferably, the principle of the hibernation strategy is to automatically switch some power modules to hibernation mode under low load or no load conditions to reduce energy consumption.
[0119] In some preferred embodiments, the formula for setting the sleep threshold is:
[0120] P sleep_threshold =P max ·θ;
[0121] Among them, P max θ represents the maximum output power of the intelligent communication power supply; θ is the sleep threshold proportional coefficient.
[0122] In the preferred embodiment described above, the decision criterion for the decision tree uses information gain, expressed as:
[0123]
[0124] Where T is the dataset; a is the attribute; T v Let T be a subset of dataset T where the value of attribute a is v; Entropy is the information entropy.
[0125] In some other preferred embodiments, the splitting criterion of the decision tree is the Gini index, expressed as:
[0126]
[0127]
[0128] Where T is the dataset; a is the attribute; T v Let T be a subset of the dataset where the value of attribute a is v; Entropy is the information entropy; C is the number of categories, p j Let be the proportion of category j in dataset T.
[0129] In a preferred embodiment, after performing step S5, the reliable intelligent communication power management method further includes:
[0130] The reliability of the intelligent communication power supply is evaluated within a preset period.
[0131] The algorithm for generating the optimization management strategy is optimized based on the reliability.
[0132] Preferably, the formula for calculating the reliability of the intelligent communication power supply is:
[0133] R 综合 =ω1·MTBF + ω2·(1-failure rate) + ω3·maintainability;
[0134] Wherein, MTBF is the mean time between failures (MTBF) obtained from the fault prediction model; ω1, ω2 and ω3 are preset weighting coefficients, set based on the actual application scenario.
[0135] In some preferred embodiments, in addition to optimizing the generation algorithm of the optimized management strategy based on the reliability, the parameters of the fault prediction model can also be further optimized, such as optimizing and adjusting the number of layers of the LSTM backbone network in the fault prediction model.
[0136] The present invention provides a reliability-based intelligent communication power management method that analyzes multi-source, highly time-series intelligent communication power supplies using a multi-task LSTM model structure. It also utilizes an attention mechanism to focus on key time-series characteristics, making the power supply capacity assessment closer to actual operating characteristics and further reducing the false alarm rate of fault prediction. Furthermore, it employs optimization management strategies, including load balancing and charging / discharging strategies, to pre-optimize and adjust the intelligent communication power supply dynamically, avoiding communication losses caused by unplanned downtime and ensuring the continuity of the communication system.
[0137] This invention provides a reliability-based intelligent communication power management system. Please refer to [link / reference]. Figure 2 The reliability-based intelligent communication power management system includes a data acquisition module 11, a model building module 12, a model training module 13, a model prediction module 14, and a power management module 15, wherein:
[0138] Data acquisition module 11 is used to acquire historical operating data of the intelligent communication power supply; the historical operating data includes historical fault operating data and historical normal operating data;
[0139] Model building module 12 is used to evaluate the power supply capability of intelligent communication power supply and build a fault prediction model based on a multi-task LSTM model structure with attention enhancement mechanism.
[0140] Model training module 13 is used to train the fault prediction model using the historical operating data;
[0141] The model prediction module 14 is used to input the real-time operating data of the intelligent communication power supply into the trained fault prediction model to obtain the fault prediction result;
[0142] The power management module 15 is used to generate an optimized management strategy for the intelligent communication power supply based on the fault prediction results and the load data of the intelligent communication power supply; the optimized management strategy includes a load balancing strategy and a charging and discharging strategy.
[0143] In a preferred embodiment, the data acquisition module 11 is specifically used for:
[0144] Acquire historical operating data of the intelligent communication power supply under preset dimensions; the preset dimensions include the power supply's internal electrical parameters, environmental parameters, and battery performance parameters;
[0145] The historical operating data is labeled according to its operating status; the operating status includes normal status and fault status.
[0146] In a preferred embodiment, the model building module 12 is specifically used for:
[0147] Based on the initial preset number of layers, an LSTM backbone network including a shared hidden layer is constructed to extract general temporal features from the input data;
[0148] Extract current fluctuation features from the input data, and input the current fluctuation features and the general time series features together into the attention mechanism layer to calculate the attention enhancement features at each time step;
[0149] Based on the dimensions of the input data, a parallel multi-task branch architecture is constructed. In each task branch, based on its data dimensions and the attention enhancement features, regression analysis is used to output the predicted power supply capability of the intelligent communication power supply.
[0150] A fault prediction model is constructed by comparing the predicted power supply capacity with the actual power supply capacity.
[0151] In a preferred embodiment, the model training module 13 includes:
[0152] The data augmentation unit is used to augment the historical operation data using a generative adversarial network to obtain augmented historical operation data;
[0153] A data partitioning unit is used to divide the enhanced historical operation data into a training set and a test set;
[0154] An iterative training unit is used to train the fault prediction model using the training set through a minimum loss function optimization algorithm.
[0155] The model validation unit is used to validate the training results of the fault prediction model using the test set, adjust the parameters of the fault prediction model according to the validation results, and retrain until the prediction accuracy requirements are met.
[0156] Furthermore, the iterative training unit is specifically used for:
[0157] pass The loss function is obtained; where y i Labeling for actual faults. The fault predicted by the fault prediction model;
[0158] The fault prediction model is trained using the training set according to the loss function.
[0159] In a preferred embodiment, the power management module 15 includes:
[0160] The load balancing strategy generation unit is used to dynamically adjust the load distribution at each time step based on the load data of the intelligent communication power supply and to obtain the load balancing strategy.
[0161] The charging and discharging strategy generation unit is used to simulate the charging, discharging, and hibernation of the intelligent communication power supply based on the load balancing strategy and the fault prediction result, so as to obtain the charging and discharging strategy.
[0162] An optimized management strategy generation unit is used to obtain an optimized management strategy for the intelligent optimized power supply based on the load balancing strategy and the charging and discharging strategy.
[0163] Further, preferably, the load balancing strategy generation unit is specifically used for:
[0164] Based on the load data of the intelligent communication power supply, the degree of deviation between the load of each component of the intelligent communication power supply and the average load is calculated to obtain the load balance.
[0165] Based on the fault prediction results, the system reliability of the intelligent communication power supply is calculated;
[0166] Obtain the input and output power of the intelligent communication power supply, and calculate the energy efficiency based on the input and output power;
[0167] The reward function of the reinforcement learning algorithm is obtained based on the load balancing degree, the system reliability, and the energy efficiency.
[0168] Based on the reward function, the load distribution at each time step is dynamically adjusted to obtain the load balancing strategy.
[0169] Preferably, the charge / discharge strategy generation unit is specifically used for:
[0170] Based on the load balancing strategy and the fault prediction results, the predicted internal electrical parameters and predicted battery performance parameters of the intelligent communication power supply are obtained.
[0171] Based on the predicted power supply's internal electrical parameters and predicted battery performance parameters, constant voltage and constant current charging simulation, discharging simulation, and low-load sleep simulation are performed on the components in the intelligent communication power supply. A decision tree based on information gain is used to obtain the charging and discharging strategy.
[0172] In a preferred embodiment, the reliability-based intelligent communication power management system further includes a policy update module, used for:
[0173] The reliability of the intelligent communication power supply is evaluated within a preset period.
[0174] The algorithm for generating the optimization management strategy is optimized based on the reliability.
[0175] The present invention provides a reliability-based intelligent communication power management system. This system analyzes multi-source, highly time-series intelligent communication power supplies using a multi-task LSTM model structure. Simultaneously, it leverages an attention mechanism to focus on key time-series characteristics, making the power supply capacity assessment closer to actual operating characteristics and further reducing the false alarm rate of fault prediction. Furthermore, it employs optimization management strategies, including load balancing and charging / discharging strategies, to pre-optimize and dynamically adjust the intelligent communication power supply, avoiding communication losses caused by unplanned downtime and ensuring the continuity of the communication system.
[0176] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.
[0177] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A reliability-based intelligent communication power management method, characterized in that, include Acquire historical operating data of the intelligent communication power supply; the historical operating data includes historical fault operating data and historical normal operating data; Based on a multi-task LSTM model structure with attention enhancement mechanism, the power supply capability of intelligent communication power supply is evaluated, and a fault prediction model is constructed. The fault prediction model is trained using the historical operating data. The real-time operating data of the intelligent communication power supply is input into the trained fault prediction model to obtain the fault prediction result; Based on the fault prediction results and the load data of the intelligent communication power supply, an optimized management strategy for the intelligent communication power supply is generated; the optimized management strategy includes a load balancing strategy and a charging / discharging strategy.
2. The reliability-based intelligent communication power management method as described in claim 1, characterized in that, The acquisition of historical operating data of the intelligent communication power supply includes: Acquire historical operating data of the intelligent communication power supply under preset dimensions; the preset dimensions include the power supply's internal electrical parameters, environmental parameters, and battery performance parameters; The historical operating data is labeled according to its operating status; the operating status includes normal status and fault status.
3. The reliability-based intelligent communication power management method as described in claim 1, characterized in that, The multi-task LSTM model structure based on the attention enhancement mechanism is used to evaluate the power supply capability of the intelligent communication power supply, and a fault prediction model is constructed based on the power supply capability, including: Based on the initial preset number of layers, an LSTM backbone network including a shared hidden layer is constructed to extract general temporal features from the input data; Extract current fluctuation features from the input data, and input the current fluctuation features and the general time series features together into the attention mechanism layer to calculate the attention enhancement features at each time step; Based on the dimensions of the input data, a parallel multi-task branch architecture is constructed. In each task branch, based on its data dimensions and the attention enhancement features, regression analysis is used to output the predicted power supply capability of the intelligent communication power supply. A fault prediction model is constructed by comparing the predicted power supply capacity with the actual power supply capacity.
4. The reliability-based intelligent communication power management method as described in claim 1, characterized in that, The step of training the fault prediction model using the historical operational data includes: Generative adversarial networks are used to enhance the historical operation data to obtain enhanced historical operation data; The enhanced historical operational data is divided into a training set and a test set; The fault prediction model is trained using the training set through a minimum loss function optimization algorithm. The training results of the fault prediction model are verified using the test set. Based on the verification results, the parameters of the fault prediction model are adjusted and retrained until the prediction accuracy requirements are met.
5. The reliability-based intelligent communication power management method as described in claim 4, characterized in that, The step of training the fault prediction model using the training set through the minimum loss function optimization algorithm includes: pass The loss function is obtained; where y i Labeling for actual faults. The fault predicted by the fault prediction model; The fault prediction model is trained using the training set according to the loss function.
6. The reliability-based intelligent communication power management method as described in claim 1, characterized in that, The step of generating an optimized management strategy for the intelligent communication power supply based on the fault prediction results and the load data of the intelligent communication power supply includes: Based on the load data of the intelligent communication power supply, a reinforcement learning algorithm is used to dynamically adjust the load distribution at each time step to obtain a load balancing strategy. Based on the load balancing strategy and the fault prediction results, the charging, discharging and hibernation simulations of the intelligent communication power supply are performed to obtain the charging and discharging strategy. Based on the load balancing strategy and the charging / discharging strategy, an optimized management strategy for the intelligent power supply is obtained.
7. The reliability-based intelligent communication power management method as described in claim 6, characterized in that, The load balancing strategy, obtained by dynamically adjusting the load distribution at each time step based on the load data of the intelligent communication power supply using a reinforcement learning algorithm, includes: Based on the load data of the intelligent communication power supply, the degree of deviation between the load of each component of the intelligent communication power supply and the average load is calculated to obtain the load balance. Based on the fault prediction results, the system reliability of the intelligent communication power supply is calculated; Obtain the input and output power of the intelligent communication power supply, and calculate the energy efficiency based on the input and output power; The reward function of the reinforcement learning algorithm is obtained based on the load balancing degree, the system reliability, and the energy efficiency. Based on the reward function, the load distribution at each time step is dynamically adjusted to obtain the load balancing strategy.
8. The reliability-based intelligent communication power management method as described in claim 6, characterized in that, The step of simulating charging, discharging, and hibernation of the intelligent communication power supply based on the load balancing strategy and the fault prediction results to obtain a charging and discharging strategy includes: Based on the load balancing strategy and the fault prediction results, the predicted internal electrical parameters and predicted battery performance parameters of the intelligent communication power supply are obtained. Based on the predicted power supply's internal electrical parameters and predicted battery performance parameters, constant voltage and constant current charging simulation, discharging simulation, and low-load sleep simulation are performed on the components in the intelligent communication power supply. A decision tree based on information gain is used to obtain the charging and discharging strategy.
9. The reliability-based intelligent communication power management method as described in claim 1, characterized in that, After generating the optimized management strategy for the intelligent communication power supply based on the fault prediction results and the load data of the intelligent communication power supply, the reliability-based intelligent communication power supply management method further includes: The reliability of the intelligent communication power supply is evaluated within a preset period. The algorithm for generating the optimization management strategy is optimized based on the reliability.
10. A reliability-based intelligent communication power management system, characterized in that, include: The data acquisition module is used to acquire historical operating data of the intelligent communication power supply; the historical operating data includes historical fault operating data and historical normal operating data. The model building module is used to evaluate the power supply capability of intelligent communication power supplies and build a fault prediction model based on a multi-task LSTM model structure with attention enhancement mechanism. The model training module is used to train the fault prediction model using the historical operating data; The model prediction module is used to input the real-time operating data of the intelligent communication power supply into the trained fault prediction model to obtain the fault prediction result; The power management module is used to generate an optimized management strategy for the intelligent communication power supply based on the fault prediction results and the load data of the intelligent communication power supply; the optimized management strategy includes a load balancing strategy and a charging / discharging strategy.