A smart power supply system for an integrated energy cabinet for communication base stations

Abstract

This invention discloses an intelligent power supply system for an integrated energy cabinet of a communication base station, including a power supply parameter acquisition module to acquire parameters such as input voltage fluctuation sequences and multi-channel load current data; a power supply feature extraction module to fuse and analyze the parameters and generate a set of power supply feature vectors; a decision space construction module to pre-set a power supply strategy knowledge base and feature vector set, and establish a power supply decision space index structure; a power supply decision module to match the optimal power supply strategy combination and output a first set of power supply control commands; and a power supply execution module to drive the power distribution unit to perform load power supply switching and battery charging and discharging control. The system includes an energy pool module that integrates photovoltaic, mains power and other equipment to achieve access and dynamic allocation; an anti-reverse current control unit to ensure the safety of energy flow; a zero-loss switch array to regulate the power path and improve efficiency; and a power supply execution module to drive multi-power supply switching, safety isolation and path switching, supporting mains power peak shaving, photovoltaic superposition and off-peak power consumption, and improving the reliability and energy efficiency of base station power supply.

Description

Technical Field

[0001] This invention relates to the field of power supply technology for communication base stations, specifically to an intelligent power supply system for an integrated energy cabinet for communication base stations. Background Technology

[0002] In modern communication networks, base stations, as key nodes for signal transmission, rely directly on a reliable power supply system for stable operation. Communication base stations are typically located in complex environments, potentially facing issues such as voltage fluctuations, frequent load changes, and large differences in temperature and humidity. These factors place extremely high demands on the adaptability and stability of the power supply system.

[0003] Traditional communication base station power supply systems have significant limitations in parameter acquisition. Most systems can only acquire single or a few power supply parameters, such as input voltage or total load current, making it difficult to comprehensively reflect the actual operating status of the power supply cabinet. For key parameters such as the current distribution of multiple loads, the charging and discharging details of the battery pack, and ambient temperature and humidity, there is often a lack of effective real-time monitoring methods, resulting in blind spots in the system's perception of the power supply status.

[0004] In power supply strategy decision-making, traditional systems mostly rely on preset fixed strategies and lack dynamic adjustment capabilities. These fixed strategies are usually based on experience and cannot be flexibly adapted to multi-dimensional parameters collected in real time. When the input voltage fluctuates frequently, the current distribution of multiple loads changes significantly, or the ambient temperature and humidity exceed the normal range, the fixed strategies are difficult to respond quickly and make optimal adjustments, which may lead to unstable power supply to some loads, or even overload or insufficient power supply.

[0005] As a crucial backup power source for base stations, the rationality of battery pack charging and discharging control directly affects its lifespan and emergency power supply capability. Traditional systems often have relatively simple charging and discharging control for battery packs, lacking refined management that combines real-time charging and discharging status, ambient temperature and humidity, and load demands. This can easily lead to problems such as overcharging and deep discharging, shortening the battery pack's service life and increasing maintenance costs and replacement frequency.

[0006] Traditional power supply systems have shortcomings in multi-parameter fusion analysis. They cannot effectively integrate multi-source information such as input voltage fluctuations, load current distribution, battery status, and environmental parameters, making it difficult to form an accurate judgment on the overall operating status of the power supply system. This results in insufficient decision-making basis and further affects the reliability and economy of the power supply system. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent power supply system for an integrated energy cabinet for communication base stations, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides an intelligent power supply system for an integrated energy cabinet for communication base stations, the system comprising:

[0009] The power supply parameter acquisition module is used to acquire the input voltage fluctuation sequence of the energy cabinet, multi-channel load current distribution data, battery pack charging and discharging state vector and ambient temperature and humidity parameters in real time.

[0010] The power supply feature extraction module performs power supply feature fusion analysis based on the input voltage fluctuation sequence, multi-channel load current distribution data, battery pack charging and discharging state vector, and ambient temperature and humidity parameters to generate a power supply feature vector set.

[0011] The decision space construction module establishes a dynamic power supply decision space index structure based on a preset power supply strategy knowledge base and the power supply feature vector set.

[0012] The real-time power supply decision module, based on the dynamic power supply decision space index structure and the current power supply feature vector set, matches the optimal power supply strategy combination and outputs the first power supply control command set.

[0013] The power supply execution module, based on the first set of power supply control commands, drives the power distribution unit of the energy cabinet to perform multi-load power supply switching and battery charging and discharging control.

[0014] Preferably, the system further includes:

[0015] The load demand forecasting module performs load demand fluctuation forecasting based on historical multi-channel load current distribution data and ambient temperature and humidity parameters, and generates a predicted load demand distribution curve.

[0016] The decision compensation module performs power supply strategy offset calibration based on the predicted load demand distribution curve and the first power supply control command set, and generates an optimized power supply control command set.

[0017] The abnormal behavior identification module calculates the load abnormality deviation based on real-time multi-channel load current distribution data and the predicted load demand distribution curve, and marks the abnormal load port identifier.

[0018] The correlation graph generation module constructs a load anomaly propagation path graph based on historical abnormal load port identifiers and power supply feature vector sets.

[0019] The strategy optimization module reconstructs the dynamic power supply decision space index structure based on the load anomaly propagation path map and the optimized power supply control command set.

[0020] Preferably, the power supply feature extraction module includes:

[0021] The fluctuation feature analysis submodule is used to perform frequency domain transformation on the input voltage fluctuation sequence and extract the proportion of voltage sag characteristic frequency bands and the fluctuation duration coefficient.

[0022] The load balancing calculation submodule is used to analyze the phase difference and peak overlap between multiple load current distribution data and generate a load balancing deviation index.

[0023] The state feature mapping submodule is used to map the charge and discharge state vector of the battery pack to a preset capacity decay model and output the health decay rate value.

[0024] The environmental coupling analysis submodule is used to calculate the time-domain correlation between environmental temperature and humidity parameters and load current distribution data, and to generate environmental coupling interference factors.

[0025] The feature aggregation submodule is used to integrate the voltage sag characteristic frequency band ratio, fluctuation duration coefficient, load balance deviation index, health decay rate value and environmental coupling interference factor to form a power supply feature vector set.

[0026] Preferably, the decision space construction module includes:

[0027] The strategy index generation submodule is used to traverse the strategy rule entries in the power supply strategy knowledge base and extract the power supply scenario constraints corresponding to each entry.

[0028] The feature matching submodule is used to calculate the similarity between the power supply feature vector set and the power supply scenario constraints, and generate a strategy matching confidence matrix.

[0029] The spatial topology construction submodule establishes decision path connection relationships between policy rule entries based on the policy matching confidence matrix.

[0030] The dynamic update submodule is used to update the strategy matching confidence matrix based on the real-time power supply feature vector set and output the dynamic power supply decision space index structure.

[0031] Preferably, the real-time power supply decision module includes:

[0032] The strategy retrieval submodule is used to retrieve neighboring strategy nodes in the dynamic power supply decision space index structure based on the current power supply feature vector set.

[0033] The strategy evaluation submodule is used to calculate the power supply efficiency improvement rate and equipment loss reduction coefficient corresponding to the neighboring strategy nodes;

[0034] The combined optimization submodule is used to integrate the control parameters of the top K strategy nodes with the highest power supply efficiency improvement rate and equipment loss reduction coefficient to generate the first power supply control command set.

[0035] Preferably, the load demand prediction module includes:

[0036] The time-series decomposition submodule is used to perform seasonal trend decomposition on historical multi-path load current distribution data and extract the fundamental component and harmonic component fluctuation period.

[0037] The environmental correlation modeling submodule is used to establish a regression mapping relationship between environmental temperature and humidity parameters and the fluctuation period of harmonic components;

[0038] The predictive synthesis submodule outputs a predicted load demand distribution curve based on the fundamental component variation trend and regression mapping relationship.

[0039] Preferably, the decision compensation module includes:

[0040] The offset detection submodule is used to compare the predicted load demand distribution curve with the load allocation parameters in the first power supply control command set and calculate the strategy offset.

[0041] The compensation strategy generation submodule generates a compensation instruction vector based on the compensation rules of the strategy offset and the power supply strategy knowledge base.

[0042] The instruction reconfiguration submodule is used to superimpose the compensation instruction vector onto the first power supply control instruction set and output the optimized power supply control instruction set.

[0043] Preferably, the abnormal behavior recognition module includes:

[0044] The residual calculation submodule is used to collect the difference between the multi-channel load current distribution data and the predicted load demand distribution curve in real time, and generate a load residual sequence.

[0045] The abnormal threshold determination submodule sets a dynamic abnormal threshold based on the load balance deviation index in the power supply feature vector set.

[0046] The anomaly tagging submodule is used to filter load residual sequence ports that exceed the dynamic anomaly threshold and output anomaly load port identifiers.

[0047] Preferably, the association map generation module includes:

[0048] The node construction submodule is used to map historical abnormal load port identifiers to graph nodes;

[0049] The edge relationship extraction submodule calculates the anomaly propagation intensity between nodes based on the spatiotemporal distribution of the proportion of voltage sag characteristic frequency bands in the power supply feature vector set.

[0050] The path generation submodule constructs directed connection edges between nodes based on the anomaly propagation strength and outputs a load anomaly propagation path graph.

[0051] Preferably, the strategy optimization module includes:

[0052] The path parsing submodule is used to extract the sequence of high-frequency propagation path nodes in the load anomaly propagation path graph.

[0053] The strategy constraint update submodule modifies the strategy matching weights of the dynamic power supply decision space index structure based on the high-frequency propagation path node sequence.

[0054] The spatial reconstruction submodule reconstructs the decision path connection relationship based on the updated strategy and matching weights, and outputs the reconstructed dynamic power supply decision space index structure.

[0055] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0056] 1. The intelligent power supply system of this integrated energy cabinet for communication base stations optimizes base station power supply in multiple ways through the collaborative work of multiple modules. The power supply parameter acquisition module can acquire input voltage fluctuation sequences, multi-channel load current distribution data, battery pack charging and discharging state vectors, and environmental temperature and humidity parameters in real time, realizing a comprehensive perception of the energy cabinet's operating status. This multi-dimensional parameter acquisition method breaks through the limitations of traditional system parameter monitoring, allowing the system to capture subtle changes in the power supply process and providing rich basic information for subsequent analysis and decision-making.

[0057] 2. Based on the comprehensively collected parameters, the power supply feature extraction module performs power supply feature fusion analysis and generates a set of power supply feature vectors. By fusing multi-source parameters, the inherent correlation between various parameters can be uncovered, more accurately reflecting the actual power supply status of the energy cabinet. Compared with the isolated parameter analysis in traditional systems, this fusion analysis method avoids the one-sidedness of single-parameter judgment, making the description of the power supply status more comprehensive and in-depth, and providing a more reliable basis for subsequent strategy decisions.

[0058] 3. The decision space construction module establishes a dynamic power supply decision space index structure based on a preset power supply strategy knowledge base and a set of power supply feature vectors, changing the traditional system's reliance on fixed strategies. The dynamic decision space index structure can be flexibly adjusted according to real-time power supply feature vectors, organically combining strategies in the preset knowledge base with the actual operating state to form a decision framework adaptable to different scenarios. This structure enables the system to cope with complex and changing power supply environments, quickly locating the appropriate strategy range when input voltage fluctuates, load changes, or environmental parameters change.

[0059] 4. The real-time power supply decision module matches the optimal power supply strategy combination and outputs control commands based on the dynamic power supply decision space index structure and the current power supply feature vector set. This real-time matching mechanism ensures that the power supply strategy can keep up with changes in the system's operating status, avoiding the lag of traditional fixed strategies under complex operating conditions. By accurately matching the optimal strategy combination, it can achieve reasonable allocation of power supply to multiple loads, preventing some loads from being affected by improper power supply.

[0060] 5. The power supply execution module drives the power distribution unit to perform switching and charging / discharging control based on control commands, translating decision-level optimization into actual power supply adjustment actions. This precise execution process ensures that the optimal strategy can be effectively implemented, achieving refined management of battery pack charging and discharging, avoiding overcharging and discharging issues, and extending the service life of the battery pack. At the same time, through flexible switching of power supply to multiple loads, it can adapt to the dynamic needs of different loads, improve the stability and adaptability of the entire energy cabinet power supply system, and better cope with the complex operating environment of communication base stations. Attached Figure Description

[0061] Figure 1 This is a timing diagram of the intelligent power supply system of the integrated energy cabinet for communication base stations described in this invention;

[0062] Figure 2 Workflow diagram for expanding the functionality of smart power supply systems;

[0063] Figure 3 A flowchart of the power supply feature extraction module;

[0064] Figure 4 This is a flowchart of the decision-making compensation module. Detailed Implementation

[0065] 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.

[0066] Please see Figure 1 This invention provides an intelligent power supply system for an integrated energy cabinet for communication base stations, the system comprising:

[0067] The power supply parameter acquisition module acquires various operating parameters of the communication base station's energy cabinet in real time: input voltage fluctuation sequence, multi-channel load current distribution data, battery pack charging and discharging state vector, and ambient temperature and humidity parameters. The power supply feature extraction module performs multi-dimensional fusion analysis based on these parameters: the input voltage fluctuation sequence is converted to the frequency domain to extract the proportion of voltage sag characteristic frequency bands and the fluctuation duration coefficient; the multi-channel load current distribution data is used to generate a load balance deviation index by calculating phase difference and peak overlap; the battery pack charging and discharging state vector is mapped to a preset capacity decay model to output a health decay rate value; and the ambient temperature and humidity parameters are correlated with the load current data to generate an environmental coupling interference factor. These features are integrated to form a power supply feature vector set. The decision space construction module establishes a dynamic power supply decision space index structure based on a preset power supply strategy knowledge base: the knowledge base stores strategy rule entries and scenario constraints; the feature vector set and constraints are used to calculate similarity to generate a strategy matching confidence matrix, which is then used to construct decision path connections and dynamically update the matrix. The real-time power supply decision module retrieves neighboring strategy nodes in the index structure based on the current power supply feature vector set, evaluates the power supply efficiency improvement rate and equipment loss reduction coefficient of the nodes, and merges the top K optimized nodes to generate the first power supply control command set. The power supply execution module parses the command to drive the energy cabinet distribution unit to perform load power supply switching and battery charging and discharging operations, completing the system closed-loop control.

[0068] Example 1: See Figure 2 and Figure 3 This implementation integrates load demand prediction, decision compensation, abnormal behavior identification, correlation graph construction, and strategy optimization mechanisms into the overall scheme of the intelligent power supply system for integrated energy cabinets in communication base stations. The power supply parameter acquisition module continuously acquires input voltage fluctuation sequences, multi-channel load current distribution data, battery pack charge / discharge state vectors, and environmental temperature and humidity parameters. The power supply feature extraction module performs frequency domain transformation when processing the input voltage fluctuation sequence, extracting the proportion of voltage sag characteristic frequency bands and fluctuation duration coefficients; it generates a load balance deviation index by analyzing the phase difference and peak overlap of multi-channel load current distribution data; it maps the battery charge / discharge state vector into a preset battery capacity decay model to a health decay rate value; and it calculates the time-domain correlation between environmental temperature and humidity parameters and load current data to obtain the environmental coupling interference factor. The feature aggregation operation integrates the above four features to output a set of power supply feature vectors.

[0069] The load demand prediction module receives historical multi-channel load current distribution data and environmental temperature and humidity parameters. Its internal time-series decomposition unit decomposes the historical load current data into fundamental and harmonic components and identifies their fluctuation period characteristics. The environmental correlation modeling unit constructs a numerical mapping model between environmental temperature and humidity parameters and the fluctuation period of the harmonic components using regression analysis. The prediction synthesis unit outputs a predicted load demand distribution curve for future periods based on the fundamental component's variation trend and the aforementioned mapping model. The decision compensation module receives this predicted curve and the first set of power supply control commands generated by the real-time power supply decision module. The offset detection unit compares the differences in load distribution parameters between the two to calculate the strategy offset, which is quantified using the absolute value of the residual or variance algorithm. The compensation strategy generation unit generates a compensation command vector based on a preset compensation rule library, which contains a mapping table of offset threshold intervals and corresponding proportional compensation parameters. The command reconstruction unit performs an additive synthesis operation on the compensation command vector and the original control commands, outputting an optimized power supply control command set.

[0070] The abnormal behavior identification module synchronously acquires real-time multi-channel load current distribution data and predicted load demand distribution curves. The residual calculation unit performs time-series alignment of the two data points to generate a load residual sequence, calculating the arithmetic difference between each port current data point and its corresponding point on the predicted curve. The abnormal threshold determination unit sets a dynamic abnormal threshold based on the load balance deviation index in the power supply feature vector set: it reads the current load balance deviation index value, multiplies it by a preset offset coefficient, and then superimposes it onto the baseline residual threshold to form the dynamic threshold boundary. The abnormal marking unit scans the residual sequences of all load ports, filters out port numbers that continuously exceed the dynamic threshold, and outputs abnormal load port identifiers.

[0071] The correlation graph generation module collects historical abnormal load port identifiers and power supply feature vector sets. The node construction unit encodes the historical abnormal port identifiers as unique node identifiers for the graph. The edge relationship extraction unit analyzes the spatiotemporal distribution characteristics of the voltage sag feature frequency band proportion in the power supply feature vector set: after segmenting the data by time window, it calculates the spatial correlation of voltage sag features of adjacent ports and quantifies the anomaly propagation intensity value using the transfer entropy algorithm. The path generation unit uses the propagation intensity value as weight to construct directed weighted edges between associated nodes to form a load anomaly propagation path graph.

[0072] The strategy optimization module processes the load anomaly propagation path graph and optimizes the power supply control command set. The path parsing unit uses a graph traversal algorithm to extract high-frequency propagation path node sequences: it statistically analyzes node occurrence frequency and path traversal frequency, filtering propagation paths with frequencies exceeding a preset threshold. The strategy constraint update unit adjusts the dynamic power supply decision space index structure based on the high-frequency node sequences: it retrieves strategy rule entries associated with high-frequency nodes in the index structure and increases their matching priority according to a preset weight adjustment ratio. The space reconstruction unit reconstructs the decision path connection relationships based on the updated strategy rule weights: it recalculates the connection strength between strategy nodes and reconstructs the topology network, outputting the reconstructed dynamic power supply decision space index structure. The system achieves self-optimization of power supply strategies under dynamic environments through continuous iteration of the above processes.

[0073] Example 2: This implementation focuses on two core modules in the overall scheme of the intelligent power supply system for integrated energy cabinets of communication base stations: decision space construction and real-time power supply decision-making. The decision space construction module establishes a dynamic power supply decision space index structure to provide a strategy matching basis for real-time power supply decisions. The real-time power supply decision-making module retrieves the optimal strategy combination based on this index structure and generates a set of power supply control commands.

[0074] The decision space construction module includes a strategy index generation submodule, a feature matching submodule, a spatial topology construction submodule, and a dynamic update submodule. The strategy index generation submodule traverses the strategy rule entries in the power supply strategy knowledge base, extracting the corresponding power supply scenario constraints for each entry. Power supply scenario constraints include parameters such as input voltage fluctuation range, load current distribution characteristics, battery health status threshold, and environmental temperature and humidity limits. The feature matching submodule calculates the similarity between the power supply feature vector set and the power supply scenario constraints, generating a strategy matching confidence matrix. The similarity calculation uses an improved cosine similarity algorithm, as shown in the following formula:

[0075]

[0076] in: Indicates the first The power supply feature vector and the first The matching similarity of each strategy rule entry; The first in the power supply feature vector set One eigenvalue; For the first The first of the strategy rule entries Standard values ​​for each constraint condition; For the first The weight coefficients for each feature are dynamically adjusted based on the feature's impact on the power supply strategy. The spatial topology construction submodule establishes decision path connections between policy rule entries based on the policy matching confidence matrix. These decision path connections are represented using a graph structure, with policy rule entries as nodes, and the edge weights between nodes determined by the matching confidence. The dynamic update submodule updates the policy matching confidence matrix based on the real-time power supply feature vector set, ensuring the index structure can dynamically respond to changes in the power supply environment.

[0077] The real-time power supply decision module comprises a strategy retrieval submodule, a strategy evaluation submodule, and a combined optimization submodule. The strategy retrieval submodule searches for neighboring strategy nodes in the dynamic power supply decision space index structure based on the current power supply feature vector set. The retrieval process uses the k-nearest neighbor algorithm, prioritizing the matching of strategy rule entries with the highest similarity. The strategy evaluation submodule calculates the power supply efficiency improvement rate and equipment loss reduction coefficient corresponding to neighboring strategy nodes. The power supply efficiency improvement rate reflects the degree of optimization of energy utilization by the strategy, calculated by comparing the change in the load balance deviation index before and after strategy execution. The equipment loss reduction coefficient quantifies the protection effect of the strategy on batteries and power distribution equipment, calculated based on the health decay rate value and the peak-valley difference of the load current. The combined optimization submodule integrates the control parameters of the top K strategy nodes with the highest power supply efficiency improvement rate and equipment loss reduction coefficient to generate the first power supply control command set. The control parameter fusion uses a weighted average method, with weights dynamically allocated based on the strategy evaluation results.

[0078] The similarity calculation in the decision space construction module optimizes the limitations of traditional cosine similarity by introducing feature weight coefficients to enhance the matching priority of key features. The policy matching confidence matrix employs sparse matrix storage technology to reduce computational resource consumption. The spatial topology construction submodule applies the minimum spanning tree algorithm to optimize decision path connectivity, reducing the complexity of policy retrieval. The dynamic update submodule implements an incremental update mechanism, recalculating similarity only for policy rule entries affected by changes in power supply characteristics, improving the efficiency of index structure updates.

[0079] The real-time power supply decision module employs a locality-sensitive hashing algorithm to accelerate neighbor node location and shorten decision response time during the strategy retrieval process. The strategy evaluation submodule uses a multi-objective optimization method to balance the trade-off between power supply efficiency and equipment losses, avoiding system imbalances caused by optimizing a single metric. The combinatorial optimization submodule uses Pareto front analysis to screen for the optimal strategy combination, ensuring optimal overall performance of the control command set.

[0080] This implementation combines a dynamic decision space index structure with a real-time policy matching mechanism to achieve intelligent optimization of the power supply strategy for communication base station energy cabinets. The similarity calculation and topology optimization of the decision space construction module ensure the accuracy of policy matching, while the multi-objective evaluation and combinatorial optimization of the real-time power supply decision module improve the adaptability of control commands. The system continuously updates the index structure and optimizes policy selection under dynamic power supply conditions, providing an efficient and stable power supply control scheme for communication base station energy cabinets.

[0081] Example 3: See Figure 4 This implementation method, within the overall scheme of the intelligent power supply system for integrated energy cabinets in communication base stations, focuses on two main functional modules: load demand prediction and decision compensation. The load demand prediction module analyzes historical load data and environmental parameters to generate a predicted load demand distribution curve for future periods. The decision compensation module dynamically calibrates real-time power supply control commands based on the prediction results, improving the system's adaptability to load fluctuations.

[0082] The load demand forecasting module comprises a time-series decomposition submodule, an environmental correlation modeling submodule, and a forecast synthesis submodule. The time-series decomposition submodule performs seasonal decomposition on historical multi-channel load current distribution data, separating it into fundamental, harmonic, and random noise components. The decomposition process employs an adaptive sliding window algorithm, with the window length dynamically adjusted according to the load fluctuation cycle. The fundamental component reflects the long-term trend of load current changes, the harmonic components exhibit periodic fluctuation characteristics, and the random noise component contains unpredictable instantaneous disturbances. The environmental correlation modeling submodule establishes a correlation model between environmental temperature and humidity parameters and the fluctuation cycle of harmonic components. The model is implemented through multiple regression analysis, considering the impact of the interaction between temperature and humidity parameters on load fluctuations. An environmental coupling factor is introduced into the regression analysis to quantify the variation patterns of harmonic components under different environmental conditions. The forecast synthesis submodule integrates the fundamental component variation trend with the output results of the environmental correlation model to generate a predicted load demand distribution curve. The synthesis process employs a weighted fusion method, with the weight of the fundamental component decreasing over time and the weight of the environmental correlation model increasing over time, ensuring that the forecast results take into account both long-term trends and short-term fluctuations.

[0083] The decision-making compensation module comprises an offset detection submodule, a compensation strategy generation submodule, and an instruction reconstruction submodule. The offset detection submodule compares the predicted load demand distribution curve with the load allocation parameters in the first power supply control instruction set to calculate the strategy offset. The offset calculation employs a dynamic time warping algorithm to eliminate minor time axis differences between the predicted curve and the actual instructions. The compensation strategy generation submodule generates compensation instruction vectors based on the strategy offset and compensation rules from the power supply strategy knowledge base. The compensation rule base uses a hierarchical structure; top-level rules define the mapping relationship between offset intervals and compensation types, while bottom-level rules refine specific compensation parameters. The instruction reconstruction submodule overlays the compensation instruction vectors onto the first power supply control instruction set, outputting an optimized power supply control instruction set. The overlay operation uses an element-wise weighted fusion method, with the compensation instruction weights dynamically adjusted based on the offset magnitude.

[0084] The time-series decomposition process of the load demand forecasting module incorporates a robust outlier handling mechanism to reduce the interference of random noise components on the prediction results. Environmental correlation modeling employs regularized regression to prevent overfitting, and the model parameter update cycle is synchronized with the temperature and humidity sampling frequency. The weight adjustment of the prediction synthesis submodule follows an exponential decay law, as shown in the following formula:

[0085]

[0086] in: Indicates the fundamental component at time [time]. The fusion weight; These are the initial weighting coefficients; This is the decay rate parameter; This is the prediction time step. This formula ensures that the prediction results gradually reflect the output of the environmental correlation model over time, adapting to the short-term fluctuation characteristics of the load.

[0087] The offset detection in the decision compensation module employs a sliding window comparison mechanism, with the window width matched to the load fluctuation cycle. The compensation strategy generation submodule dynamically prioritizes rules, automatically increasing the matching order of frequently used rules. The instruction reconstruction submodule introduces a smooth transition mechanism for weight adjustment to prevent system oscillations caused by abrupt changes in compensation instructions.

[0088] This implementation achieves dynamic optimization of the power supply strategy through the coordinated operation of load demand forecasting and decision compensation. The load demand forecasting module improves forecast accuracy through time-series decomposition and environmental correlation modeling, while the decision compensation module enhances control adaptability through offset detection and command reconstruction. The system continuously calibrates the power supply strategy under complex environmental conditions, providing accurate load tracking capabilities for the communication base station's energy cabinet. The closed-loop operation of the forecasting and compensation mechanisms enables the system to autonomously cope with load demand fluctuations, maintaining power supply stability and energy efficiency balance.

[0089] Example 4: This implementation method, in the intelligent power supply system of the integrated energy cabinet for communication base stations, mainly focuses on the detailed description of the identification and correlation analysis functions of abnormal load behavior. The system identifies abnormal ports and constructs an abnormal propagation path map by real-time monitoring of the deviation between the load current and the predicted curve, providing a basis for power supply strategy optimization.

[0090] The abnormal behavior identification module continuously receives the predicted load demand distribution curve from the load demand prediction module and real-time multi-channel load current data from the power supply parameter acquisition module. An internal residual calculation unit compares the measured current value with the predicted value for each load port every 5 minutes. Table 1 shows the abnormal detection data of a base station's power cabinet during the morning peak load period.

[0091] Table 1: Load port anomaly detection data record.

[0092]

[0093] The anomaly threshold determination unit dynamically adjusts the anomaly threshold for each port based on the load balance deviation index in the power supply characteristic vector. When the load balance deviation index increases, the system automatically tightens the anomaly determination criteria; when the index decreases, the determination criteria are appropriately relaxed. The anomaly marking unit marks ports that exceed the threshold for three consecutive detection cycles to avoid false alarms caused by transient interference. The information of the marked abnormal ports includes key parameters such as port number, anomaly start time, duration, and maximum residual value.

[0094] The correlation graph generation module receives historical abnormal port tagging data and constructs an abnormal behavior analysis window with a configurable time span. The node construction unit creates a unique node identifier for each abnormal port, and the node attributes include port type, rated power, and physical location information. The edge relationship extraction unit analyzes the spatiotemporal characteristics of voltage sag events. For example, in a voltage sag event, if the adjacent ports P03 and P05 successively experience abnormalities within 3 minutes after the P01 port experiences an anomaly, the system records this temporal sequence and spatial proximity.

[0095] The path generation unit uses a directed graph structure to represent anomaly propagation relationships. The node size reflects the frequency of anomaly occurrence, and the edge weight represents the propagation probability. For example, if there is a 70% probability that an anomaly at port P01 will trigger an anomaly at port P03, the weight of the corresponding edge is set to 0.7. The graph visualization function helps operations and maintenance personnel intuitively understand anomaly hotspots and typical propagation paths.

[0096] In actual operation, the system demonstrates the ability to identify the following typical abnormal scenarios: when the current at a certain load port is continuously high due to equipment failure, the abnormal behavior identification module can mark the abnormality within two detection cycles; when a voltage dip triggers a chain reaction, the correlation graph can clearly show the propagation path from the power input terminal to the end load; when a sudden change in ambient temperature causes multiple loads to deviate from the predicted value simultaneously, the system can identify this group abnormality and distinguish it from a single point of failure.

[0097] The anomaly analysis data storage adopts a circular buffer structure, retaining detailed records for the most recent 72 hours by default. For confirmed anomalies, the system automatically generates an analysis report containing the following elements: a list of anomaly ports, the time of first occurrence, the duration distribution, associated voltage sag events, and environmental parameter change curves. This data provides input for subsequent strategy optimization modules and also supports manual analysis needs by operations and maintenance personnel.

[0098] The system implementation took into account the actual constraints of industrial environments: the anomaly detection cycle can be configured from 1 to 10 minutes depending on the equipment's communication capabilities; the dynamic threshold adjustment setting limits the maximum variation range to avoid drastic parameter fluctuations; and the map construction algorithm uses an incremental update method to reduce computational resource consumption. These designs enable this implementation to operate stably in various base station environments, balancing detection sensitivity with system reliability.

[0099] Through continuous anomaly monitoring and correlation analysis, the system gradually establishes a database of anomaly patterns specific to each base station. When similar anomaly sequences recur, the system can issue early warnings for potentially affected ports. This predictive maintenance capability based on historical data significantly improves the operational reliability of the communication base station power system. Maintenance personnel can query anomaly patterns for specific time periods through a human-machine interface to help locate the root cause of systemic failures.

[0100] Example 5: This implementation focuses on the operation mechanism of the strategy optimization module in the intelligent power supply system architecture of the integrated energy cabinet for communication base stations. The strategy optimization module receives the load anomaly propagation path map from the correlation map generation module and the set of optimized power supply control commands from the decision compensation module, and achieves adaptive evolution of the strategy by adjusting the decision space index structure.

[0101] The path resolution submodule extracts high-frequency propagation node sequences from the load anomaly propagation path graph. Node sequence extraction is based on directed graph traversal analysis, calculating the in-degree and out-degree weights of each node. The node occurrence frequency is obtained by counting occurrences within a sliding time window. The system sets a frequency threshold; when the occurrence frequency of a node sequence continuously exceeds the threshold, it is marked as a high-frequency propagation path. These paths record typical anomaly propagation patterns, including the sequential relationship between the starting port, intermediate nodes, and ending nodes. Spatiotemporal correlation characteristics are preserved during path resolution, recording the timestamp of the first anomaly occurrence, the propagation time interval, and geographical location information.

[0102] The strategy constraint update submodule modifies the strategy matching weights of the dynamic power supply decision space index structure based on the high-frequency propagation path node sequence. The weight adjustment operation accesses the strategy rule entry library in the index structure to query strategy rules related to high-frequency nodes. Relevant rules are identified based on the mapping relationship between port numbers and strategy constraints. Matched strategy rule entries initiate the weight modification process: increasing the strategy weight related to abnormal source ports and decreasing the strategy matching priority of associated termination ports. Weight coefficients are adjusted hierarchically, with larger adjustments for primary path nodes than secondary nodes. The weight update uses an incremental algorithm to avoid abrupt changes in the original decision relationships.

[0103] The spatial reconstruction submodule re-establishes decision path connections based on the updated policy matching weights. This process includes two stages: connection strength recalculation and topology reconstruction. Connection strength recalculation iterates through the constraint feature vectors of all policy rule entries, recalculating the similarity correlation values ​​between policy entries using the updated weight coefficients. In the topology reconstruction stage, existing decision path connections are deleted, and new policy node groups are generated using a minimum distance clustering algorithm based on the recalculated similarity correlation values. Fully connected paths are established within each policy node group, while cross-group connections between core nodes are preserved. Finally, the reconstructed dynamic power supply decision space index structure is output, and the new structure is automatically synchronized to the real-time power supply decision module.

[0104] During system operation, the strategy optimization module periodically activates the execution process, with the execution cycle synchronized with the load monitoring interval. Each execution retains historical high-frequency path cache data, which is then merged and analyzed with newly added path data to form a long-term propagation pattern library. This library establishes a path pattern evolution record, supporting the retrospective analysis of anomaly propagation trend changes. The spatial reconstruction operation introduces a smooth transition mechanism, gradually transitioning through a weighted fusion of old and new index structures to avoid abrupt changes in power supply strategies. The reconstructed index structure generates a version identifier and establishes a correlation with the optimized power supply control command set.

[0105] This implementation exhibits the following characteristics in actual base station operation: when a specific port repeatedly experiences anomalies and triggers a propagation chain, the optimization mechanism increases the matching priority of preventative strategies; when the anomaly propagation pattern changes, the system autonomously updates the strategy decision path; and when the old strategy is incompatible with the new load characteristics, the reconstruction mechanism achieves seamless migration of the strategy space. An operation log is established during the execution process, recording each high-frequency path change, weight adjustment details, and reconstruction details. The log content includes the set of modified port numbers, the number of affected strategy rule entries, and the topology difference index before and after reconstruction.

[0106] The system employs a dynamic quota system for resource allocation. When the load is stable, optimization tasks run in the background with low priority; when high-frequency abnormal paths are continuously detected, the task priority is automatically increased. Selective reconfiguration is supported under resource constraints, with only partial updates performed on the sub-decision spaces associated with high-frequency paths. This design adapts to the varying computing capabilities of base station equipment, ensuring effective operation under different hardware configurations.

[0107] Through continuous self-optimization of the policy space, the system forms a comprehensive decision-making system that combines initial policy configuration with dynamic evolution. Each reconstruction operation retains historical versions of the index structure, supporting rollback to a stable version in abnormal situations. The entire process does not rely on manual configuration of policy rules; instead, it autonomously optimizes the power supply decision logic by analyzing actual operational data, forming a closed-loop intelligent optimization mechanism. Operational data shows that this mechanism enables the system to adapt to the long-term evolution of base station load characteristics, reduce the frequency of manual policy adjustments, and maintain the adaptability of power supply control in the energy cabinet.

[0108] The intelligent power supply system of the integrated energy cabinet for communication base stations described in this invention achieves refined management and control of base station energy through multi-module collaborative operation in actual deployment, demonstrating significant adaptability, especially in integrating diverse energy access and efficient power distribution. After system startup, the power supply parameter acquisition module immediately enters working mode, continuously capturing various key parameters during the operation of the energy cabinet: the input voltage fluctuation sequence records the instantaneous voltage changes when different power sources are connected via a high-frequency sampling device, including the 220V AC voltage fluctuation of the mains power, the DC output voltage fluctuation of the photovoltaic modules, and the voltage surge curve when the generator starts; multi-channel load current distribution data is monitored in real time for different load ports within the base station, such as the RF module, signal processing unit, and heat dissipation system, forming dynamic curves of current changes over time at each port; the battery pack charge / discharge state vector includes indicators such as current remaining capacity, charge / discharge current magnitude, and internal temperature, which are obtained in real time through the battery management system interface; environmental temperature and humidity parameters are synchronously collected by sensors inside and outside the cabinet, reflecting the temperature and humidity change trends of the environment where the base station is located. These parameters are summarized at millisecond-level frequencies, providing continuous and comprehensive raw data for subsequent processing.

[0109] After receiving the above parameters, the power supply feature extraction module initiates a multi-dimensional fusion analysis process. For the input voltage fluctuation sequence, the module separates the fluctuation components of different frequency bands through filtering and transformation, identifying the main influencing frequency bands of voltage sags and their duration characteristics. For example, mains voltage sags are mostly concentrated in specific low-frequency bands and have short durations, while voltage fluctuations during generator startup cover a wider frequency band and have longer durations. For multi-channel load current distribution data, the module calculates the load balance by comparing the phase differences and peak occurrence times of the current at each port. When the current peaks of several load ports occur simultaneously, they are marked as load concentration risk characteristics. The battery pack charge / discharge state vector is input into a preset capacity decay model. The model maps the decay trend of battery health status based on the current charge / discharge rate and historical data. For example, under high-temperature environments, continuous high-current discharge will show a significantly accelerated trend in health decay. Environmental coupling analysis compares temperature and humidity change curves with load current fluctuation curves to find the correlation between the two, such as the correlation characteristic that the load current of the cooling system increases linearly with increasing temperature under high-temperature environments. These processed features are integrated into a set of power supply feature vectors, which fully present the real-time operating status of the energy system.

[0110] The decision space construction module builds a dynamically adjustable decision framework based on a pre-defined power supply strategy knowledge base. This knowledge base stores power supply strategy rules for different scenarios, including load allocation schemes when photovoltaic power is the primary source, energy switching logic during peak grid power periods, and triggering conditions for generator emergency startup. Each rule corresponds to specific scenario constraints, such as photovoltaic irradiance thresholds, grid load rate limits, and battery capacity limits. The module compares the power supply feature vector set with these constraints, calculates the degree of matching, and generates a strategy matching confidence matrix. Each element in the matrix represents the degree of fit between a strategy rule and the current operating state. Based on this matrix, the module further constructs decision path connections between strategy rules. For example, when photovoltaic power is insufficient, it automatically links to a strategy path for grid power supplementation, or when battery capacity is low, it activates a strategy path for generator charging. As real-time parameters are continuously updated, the decision space index structure is dynamically adjusted to ensure it remains adapted to the current operating state.

[0111] The real-time power supply decision module, as the core decision-making unit of the system, performs optimal strategy matching based on the dynamic power supply decision space index structure and the current power supply feature vector set. The module first retrieves the strategy node in the index structure that is closest to the current feature vector. These nodes represent effective strategies adopted in historical scenarios similar to the current state. Subsequently, these neighboring strategy nodes are evaluated from multiple dimensions, analyzing their potential energy utilization efficiency and equipment operational stability under the current state. For example, one strategy node achieves smooth switching between photovoltaic and mains power under similar load distribution, while another node optimizes the battery charging and discharging rhythm under the same temperature and humidity conditions. After comprehensive comparison, the module selects the strategy combination with the highest adaptability and generates the first power supply control command set. These commands include specific parameters such as the power supply priority of each load port, the battery charging and discharging current threshold, and the access ratio of different power sources.

[0112] Supported by the energy pool module, the system achieves flexible integration and scheduling of diverse energy sources. This module integrates photovoltaic module interfaces, mains power access terminals, generator connection devices, and energy storage battery packs. Through an adaptive voltage conversion module, it is compatible with the output characteristics of different power sources: the DC output of the photovoltaic modules is directly connected to the low-voltage DC load circuit after voltage stabilization; the AC voltage of the mains power is connected to the AC load circuit through an isolation transformer; the emergency output of the generator is connected to the main power supply circuit through a fast-switching switch; and the energy storage battery pack achieves flexible switching between charging and discharging states through a bidirectional converter. When photovoltaic irradiance is sufficient, the energy pool module prioritizes the allocation of photovoltaic power to DC loads, storing any remaining power in the energy storage device. When the mains power supply is stable and during off-peak hours, the module controls the energy storage device to charge, reserving power to meet peak demand. In the event of a mains power outage, the generator automatically starts, providing continuous power to critical loads through the energy pool module, ensuring that the core functions of the base station are not affected.

[0113] The anti-reverse current control unit plays a crucial safety protection role in the collaborative power supply of multiple energy sources. This unit uses high-precision current sensors to monitor the current direction and magnitude at each energy access point in real time. When it detects photovoltaic power flowing back into the mains grid, it immediately triggers an isolation mechanism, disconnecting the photovoltaic power from the mains grid via a fast circuit breaker to prevent impact on the grid. When both generators and mains power are connected simultaneously, the control unit continuously compares their voltage phases. If the phase difference exceeds the safe range, it immediately disconnects one of the power sources to prevent circulating current. When energy storage devices supply power to the load, if reverse charging current is detected, the control unit quickly adjusts the switching state of the energy storage circuit to protect the battery pack from reverse current damage. This real-time detection and rapid response mechanism provides a solid safety guarantee for the hybrid power supply of multiple energy sources.

[0114] As a core component of power path regulation, the zero-loss switch array directly distributes power through high-performance power electronic switches, eliminating the need for boost, buck, or rectification stages in traditional power supply systems. The switching elements in the array utilize low-on-resistance semiconductor devices, resulting in virtually no energy loss in the on-state, and the switching response time is controlled in the microsecond range, ensuring continuous power supply to the load during power switching. When photovoltaic power needs to be distributed to AC loads, the switch array directly connects the inverter-processed photovoltaic power to the AC circuit through a preset path, eliminating the need for additional voltage conversion devices. During the switching between mains power and energy storage power, the switch array achieves a seamless transition through parallel switching logic, ensuring no voltage fluctuations are felt at the load end. For loads with different voltage levels, the voltage divider switch group in the array can directly adjust the output path to adapt to the load's voltage requirements. This direct regulation mode significantly reduces losses during energy conversion and improves the energy utilization efficiency of the entire power supply system.

[0115] The power supply execution module translates the first set of power supply control commands into specific operational actions, driving the coordinated operation of various hardware modules. Regarding multi-power supply switching, the module controls the switching action within the energy pool module based on the power priority parameters in the commands. For example, during peak electricity consumption periods on weekdays, if the command requires priority use of energy storage devices, the execution module drives the energy storage circuit switch to close, simultaneously disconnecting the main grid power supply circuit and retaining only auxiliary power. When the photovoltaic irradiance reaches a preset value, the command triggers the superposition mode, and the execution module controls the photovoltaic circuit and the grid power circuit to connect simultaneously. Through the path allocation of the zero-loss switch array, photovoltaic power is directed to non-critical loads, while the grid power ensures power supply to the core loads. Regarding reverse current prevention control, the execution module sets the detection sensitivity of the reverse current prevention control unit based on the safety threshold parameters in the commands. When the command requires enhanced photovoltaic reverse current prevention protection, the execution module increases the sampling frequency of the current sensor, shortening the response time of the isolation mechanism. In terms of power path regulation, the execution module controls the switching combination state of the zero-loss switch array according to the load distribution scheme in the instruction. For example, it switches the power supply path of the radio frequency module to the photovoltaic circuit and the power supply path of the heat dissipation system to the energy storage circuit, so as to achieve precise power distribution.

[0116] In actual operation scenarios, the system can automatically adjust its operating mode according to different environmental and load changes. During sunny days, when the photovoltaic module output is stable, the system starts the tandem photovoltaic mode, where photovoltaic and mains power jointly supply power to the load. Excess photovoltaic energy is stored in the energy storage device. At this time, the zero-loss switch array directly guides the photovoltaic energy to the DC load, reducing conversion links. During the evening peak electricity consumption period, when the mains load rate rises, the system automatically switches to the mains peak shaving mode. The energy storage device releases energy to bear part of the load, reducing the mains voltage pressure. The anti-reverse current control unit monitors the direction of energy storage supply to the load in real time to ensure that no current flows back into the mains. If there is cloudy or rainy weather and the photovoltaic output is insufficient, the system mainly supplies power from the mains, while keeping the generator in standby mode. Once the mains power is interrupted, the generator immediately starts and connects to the power supply circuit through the energy pool module to ensure uninterrupted operation of the base station.

[0117] Through the close collaboration of its various modules, the entire system achieves efficient integration of diverse energy sources, low-loss distribution of electrical energy, and safe and stable power supply. It can adapt to the power supply requirements of base stations under different environmental conditions and load demands, demonstrating strong environmental adaptability and operational reliability.

[0118] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0119] 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. An intelligent power supply system of a communication base station integrated energy cabinet, characterized in that, The system includes: The power supply parameter acquisition module is used to acquire the input voltage fluctuation sequence of the energy cabinet, multi-channel load current distribution data, battery pack charging and discharging state vector and ambient temperature and humidity parameters in real time. The power supply feature extraction module performs power supply feature fusion analysis based on the input voltage fluctuation sequence, multi-channel load current distribution data, battery pack charging and discharging state vector, and ambient temperature and humidity parameters to generate a power supply feature vector set. The decision space construction module establishes a dynamic power supply decision space index structure based on a preset power supply strategy knowledge base and the power supply feature vector set. The real-time power supply decision module, based on the dynamic power supply decision space index structure and the current power supply feature vector set, matches the optimal power supply strategy combination and outputs the first power supply control command set. The power supply execution module, based on the first set of power supply control instructions, drives the power distribution unit of the energy cabinet to perform multi-load power supply switching and battery charging and discharging control; The load demand forecasting module performs load demand fluctuation forecasting based on historical multi-channel load current distribution data and ambient temperature and humidity parameters, and generates a predicted load demand distribution curve. The decision compensation module performs power supply strategy offset calibration based on the predicted load demand distribution curve and the first power supply control command set, and generates an optimized power supply control command set. The abnormal behavior identification module calculates the load abnormality deviation based on real-time multi-channel load current distribution data and the predicted load demand distribution curve, and marks the abnormal load port identifier. The correlation graph generation module constructs a load anomaly propagation path graph based on historical abnormal load port identifiers and power supply feature vector sets. The strategy optimization module reconstructs the dynamic power supply decision space index structure based on the load anomaly propagation path map and the optimized power supply control command set. The power supply feature extraction module includes: The fluctuation feature analysis submodule is used to perform frequency domain transformation on the input voltage fluctuation sequence and extract the proportion of voltage sag characteristic frequency bands and the fluctuation duration coefficient. The load balancing calculation submodule is used to analyze the phase difference and peak overlap between multiple load current distribution data and generate a load balancing deviation index. The state feature mapping submodule is used to map the charge and discharge state vector of the battery pack to a preset capacity decay model and output the health decay rate value. The environmental coupling analysis submodule is used to calculate the time-domain correlation between environmental temperature and humidity parameters and load current distribution data, and to generate environmental coupling interference factors. The feature aggregation submodule is used to integrate the voltage sag characteristic frequency band ratio, fluctuation duration coefficient, load balance deviation index, health decay rate value and environmental coupling interference factor to form a power supply feature vector set. The decision space construction module includes: The strategy index generation submodule is used to traverse the strategy rule entries in the power supply strategy knowledge base and extract the power supply scenario constraints corresponding to each entry. The feature matching submodule is used to calculate the similarity between the power supply feature vector set and the power supply scenario constraints, and generate a strategy matching confidence matrix. The spatial topology construction submodule establishes decision path connection relationships between policy rule entries based on the policy matching confidence matrix. The dynamic update submodule is used to update the strategy matching confidence matrix based on the real-time power supply feature vector set and output the dynamic power supply decision space index structure.

2. The intelligent power supply system of claim 1, wherein, The real-time power supply decision module includes: The strategy retrieval submodule is used to retrieve neighboring strategy nodes in the dynamic power supply decision space index structure based on the current power supply feature vector set. The strategy evaluation submodule is used to calculate the power supply efficiency improvement rate and equipment loss reduction coefficient corresponding to the neighboring strategy nodes; The combined optimization submodule is used to integrate the control parameters of the top K strategy nodes with the highest power supply efficiency improvement rate and equipment loss reduction coefficient to generate the first power supply control command set.

3. The intelligent power supply system of claim 1, wherein, Also includes: The energy pool module is used to integrate photovoltaic, mains power, generator and energy storage equipment to realize flexible access and dynamic load distribution of power sources of different voltage levels; Anti-backflow control unit is used to ensure the safety of energy flow through intelligent detection and isolation mechanisms; Zero-loss switching arrays are used to directly control the power path using high-performance switching arrays, avoiding the two energy conversion stages of boost and buck, and improving power supply efficiency. The power supply execution module is also used to drive the energy pool module to perform multi-power supply access switching, control the anti-reverse current control unit to perform energy safety isolation operation, and regulate the power path switching of the zero-loss switch array based on the first power supply control command set, so as to support the functions of mains power peak shaving, superimposed photovoltaic and off-peak power consumption.

4. The intelligent power supply system according to claim 1, characterized in that, The load demand prediction module includes: The time-series decomposition submodule is used to perform seasonal trend decomposition on historical multi-path load current distribution data and extract the fundamental component and harmonic component fluctuation period. The environmental correlation modeling submodule is used to establish a regression mapping relationship between environmental temperature and humidity parameters and the fluctuation period of harmonic components; The predictive synthesis submodule outputs a predicted load demand distribution curve based on the fundamental component variation trend and regression mapping relationship.

5. The intelligent power supply system according to claim 1, characterized in that, The decision compensation module includes: The offset detection submodule is used to compare the predicted load demand distribution curve with the load allocation parameters in the first power supply control command set and calculate the strategy offset. The compensation strategy generation submodule generates a compensation instruction vector based on the compensation rules of the strategy offset and the power supply strategy knowledge base. The instruction reconfiguration submodule is used to superimpose the compensation instruction vector onto the first power supply control instruction set and output the optimized power supply control instruction set.

6. The intelligent power supply system according to claim 1, characterized in that, The abnormal behavior identification module includes: The residual calculation submodule is used to collect the difference between the multi-channel load current distribution data and the predicted load demand distribution curve in real time, and generate a load residual sequence. The abnormal threshold determination submodule sets a dynamic abnormal threshold based on the load balance deviation index in the power supply feature vector set. The anomaly tagging submodule is used to filter load residual sequence ports that exceed the dynamic anomaly threshold and output anomaly load port identifiers.

7. The intelligent power supply system according to claim 1, characterized in that, The association map generation module includes: The node construction submodule is used to map historical abnormal load port identifiers to graph nodes; The edge relationship extraction submodule calculates the anomaly propagation intensity between nodes based on the spatiotemporal distribution of the proportion of voltage sag characteristic frequency bands in the power supply feature vector set. The path generation submodule constructs directed connection edges between nodes based on the anomaly propagation strength and outputs a load anomaly propagation path graph.

8. The intelligent power supply system according to claim 1, characterized in that, The strategy optimization module includes: The path parsing submodule is used to extract the sequence of high-frequency propagation path nodes in the load anomaly propagation path graph. The strategy constraint update submodule modifies the strategy matching weights of the dynamic power supply decision space index structure based on the high-frequency propagation path node sequence. The spatial reconstruction submodule reconstructs the decision path connection relationship based on the updated strategy and matching weights, and outputs the reconstructed dynamic power supply decision space index structure.

Images