Intelligent power supply dynamic adjustment module software based on edge calculation

Abstract

The invention discloses intelligent power supply dynamic adjustment module software based on edge computing, and relates to the technical field of power supply adjustment, and the software comprises a terminal cloud system building module which completes the deployment and initialization of cloud and edge nodes; the edge data acquisition and extraction module acquires and preprocesses data and extracts working condition fingerprints; the edge heuristic matching module locates a working condition type and synchronizes data to the cloud; the cloud strategy generation module generates a resource allocation strategy and returns the resource allocation strategy; the end cloud closed-loop updating module realizes iterative optimization of end cloud data and an algorithm through bidirectional updating, and the software performance is improved; according to the invention, multi-modal fusion and self-adaptive acquisition are adopted, and strict quality control is combined to guarantee data reliability; efficiency and safety are both considered through dynamic bandwidth transmission and encryption processing; the edge end intelligently analyzes and accurately captures parameter changes, and misjudgment is avoided; and meanwhile, a time sequence coupling algorithm and a multi-dimensional early warning mechanism improve the risk identification timeliness, and resource optimization configuration is realized.

Description

Technical Field

[0001] This invention relates to the field of power regulation technology, specifically to intelligent power dynamic regulation module software based on edge computing. Background Technology

[0002] With the widespread adoption of electronic devices, industrial control systems, and IoT terminals, power modules, as core power supply units, directly impact the overall system performance through their operational stability and resource utilization efficiency. Currently, the operating environments of various devices are complex and ever-changing, exhibiting diverse and dynamic operating conditions. This places higher demands on the adjustment response speed and adaptation accuracy of power modules. Edge computing, with its advantages of distributed deployment and low-latency processing, has gradually become a key technology supporting real-time data processing. The edge-cloud collaborative architecture integrates the real-time processing capabilities of the edge with the global data resources of the cloud, providing a technological foundation for dynamic power supply adjustment under complex operating conditions. Furthermore, the operating data generated during power module operation encompasses various types, requiring efficient storage, retrieval, and analysis mechanisms. Multimodal data integration and intelligent strategy generation have become core requirements for improving power supply regulation performance, driving power supply regulation technology towards data-driven and dynamic optimization.

[0003] Traditional power supply regulation technologies have significant limitations in dealing with complex and ever-changing operating conditions. Some technologies employ centralized computing architectures, where all data processing and strategy generation rely on a single node, resulting in high data transmission latency. This makes it difficult to respond in real time to rapid changes in operating conditions and to adjust regulation strategies in a timely manner. At the data processing level, traditional methods often only focus on a few operating parameters, lacking comprehensive integration and standardized processing of multi-type operating condition data. This leads to insufficient completeness and accuracy in operating condition identification. The operating condition matching logic is relatively simple and fails to comprehensively consider multi-dimensional features, resulting in insufficient alignment between the matching results and actual operating conditions, thus affecting the adaptability of the strategy. In addition, traditional technologies lack a sound iterative update mechanism. After strategy generation, it is difficult to continuously optimize based on actual execution results. After long-term operation, problems such as decreased regulation accuracy and unreasonable resource allocation are likely to occur, making it unable to adapt to the dynamic evolution of operating conditions. This results in low operating efficiency of power modules and poor energy consumption control. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide intelligent power dynamic adjustment module software based on edge computing. It constructs an edge-cloud collaborative architecture, deploying a multimodal operating condition data index library, a strategy gene library, and a lightweight generation model in the cloud, while deploying an operating condition-strategy performance surface meta-model at the edge nodes. The edge module collects and preprocesses operational data, extracts operating condition fingerprints, and completes multi-dimensional matching. After parsing the edge data in the cloud, it calls the strategy gene to generate a computing power-power allocation strategy and sends it back for execution. Utilizing an edge-cloud closed-loop update mechanism, it continuously optimizes model and algorithm parameters, achieving iterative upgrades of data and strategies. The software integrates the advantages of real-time edge processing and global cloud resources, improving power adjustment accuracy and response efficiency, optimizing resource allocation, and ensuring stable and efficient operation of the power module.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Intelligent power dynamic adjustment module software based on edge computing, the software comprising: Edge-cloud system construction module: Deploy a multimodal working condition data index library, the network-wide optimal strategy gene library, and a lightweight strategy generative model in the cloud; deploy working condition type-strategy performance surface meta-models in each edge computing node; and initialize the cloud and edge computing nodes simultaneously. Edge data acquisition and extraction module: Edge computing nodes collect power module operation data, preprocess it to form a structured operating condition dataset, extract operating condition fingerprints from the structured operating condition dataset and perform validity verification, and synchronize the dataset to the operating condition type-strategy performance surface meta-model at the edge to provide standardized data for operating condition matching; Edge-inspired matching module: Edge computing nodes run a multi-dimensional matching algorithm for working condition fingerprints, calculate the comprehensive matching degree between the working condition fingerprint and the working condition type features in the working condition type-policy performance surface meta-model, locate the target working condition type and extract relevant data, encapsulate multiple types of working condition related information and synchronize them to the cloud; Cloud-based strategy generation module: The cloud receives and parses the encapsulated data from the edge, runs the cloud-edge collaborative strategy generation algorithm, retrieves the strategy gene based on the matching degree calculation result, generates a set of computing power-power resource allocation strategy parameters, and after verification, sends the parameter set back to the corresponding edge computing node. Edge-to-cloud closed-loop update module: Edge computing nodes parse parameter sets and issue adjustment commands to power modules, monitor operating status to form execution effect datasets, update edge terminal models and adjust parameters of multi-dimensional matching algorithm for operating condition fingerprints, synchronize all operating condition data to the cloud, update cloud databases and edge-to-cloud collaboration strategies to generate algorithm calculation benchmarks, and realize iterative optimization of edge-to-cloud data and algorithms. This software includes: Edge-cloud system construction module: Deploy a multimodal working condition data index library, the network-wide optimal strategy gene library, and a lightweight strategy generative model in the cloud; deploy working condition type-strategy performance surface meta-models in each edge computing node; and initialize the cloud and edge computing nodes simultaneously. Edge data acquisition and extraction module: Edge computing nodes collect power module operation data, preprocess it to form a structured operating condition dataset, extract operating condition fingerprints from the structured operating condition dataset and perform validity verification, and synchronize the dataset to the operating condition type-strategy performance surface meta-model at the edge to provide standardized data for operating condition matching; Edge-inspired matching module: Edge computing nodes run a multi-dimensional matching algorithm for working condition fingerprints, calculate the comprehensive matching degree between the working condition fingerprint and the working condition type features in the working condition type-policy performance surface meta-model, locate the target working condition type and extract relevant data, encapsulate multiple types of working condition related information and synchronize them to the cloud; Cloud-based strategy generation module: The cloud receives and parses the encapsulated data from the edge, runs the cloud-edge collaborative strategy generation algorithm, retrieves the strategy gene based on the matching degree calculation result, generates a set of computing power-power resource allocation strategy parameters, and after verification, sends the parameter set back to the corresponding edge computing node. Edge-cloud closed-loop update module: Edge computing nodes parse parameter sets and issue adjustment instructions to power modules, monitor operating status to form execution effect datasets, update edge terminal models and adjust parameters of multi-dimensional matching algorithm for operating condition fingerprints, synchronize all operating condition data to the cloud, update cloud database and edge-cloud collaboration strategy to generate algorithm calculation benchmarks, and realize iterative optimization of edge-cloud data and algorithms.

[0006] Furthermore, the specific contents of the multimodal operating condition data index library, the network-wide optimal strategy gene library, and the lightweight strategy generative model in the edge-cloud system construction module are as follows: The multimodal operating condition data index library is deployed using a distributed storage architecture. The library is divided into data storage partitions according to the operating scenarios of the power modules. Each partition contains four types of data: text-based operating condition descriptions, numerical operating parameters, waveform energy consumption curves, and fingerprint-based operating characteristics of the power modules. All data entering the library is processed by feature vectorization. A two-layer data retrieval index is established according to operating condition type and parameter dimension. The initialization phase completes the configuration of basic data classification rules, data vectorization processing standards, and retrieval index establishment rules. The optimal strategy gene library of the entire network and the multimodal operating condition data index library form a data interaction connection. During the initialization phase, the historical computing power-power scheduling strategy of the power module is subjected to feature extraction, feature screening and gene clustering. The processed strategy genes are classified and stored in the corresponding partitions of the library according to the operating condition type. The operation permissions for adding, modifying and retrieving strategy genes are configured for each partition, and the update cycle and calculation standard of gene clustering are set. The lightweight strategy generation model is deployed based on the power module computing power-power scheduling scenario. The model is equipped with a strategy gene retrieval interface, a parameter calculation module, and a strategy generation module. During the initialization phase, the model's inference calculation rate, data processing dimension, and strategy generation parameter output format are configured. The model is adapted to the multimodal operating condition data index library and the network-wide optimal strategy gene library through data interface adaptation. The model's start-up trigger conditions and data interaction protocol are configured.

[0007] Furthermore, in the edge-cloud system construction module, the working condition type-policy performance surface meta-models deployed on each edge computing node are all independently and lightweightly deployed. After deployment, they are initialized synchronously. Specifically, the meta-model stores the working condition type classification features, the policy performance distribution law corresponding to each working condition type, and the boundary range of working condition parameters. The working condition type classification features are divided into different dimensions according to the computing power load level, voltage fluctuation range, and energy consumption range. Each dimension feature is stored in a fixed data format. The policy performance distribution law is stored in the form of a numerical table, which contains the historical scheduling parameters and operating parameter records corresponding to each working condition type. The boundary range of working condition parameters is set separately for each working condition type, and the maximum, minimum, and standard values ​​of each operating parameter are clearly defined.

[0008] Furthermore, in the edge data acquisition module, the preprocessing of the power module operation data is performed sequentially in the steps of cleaning, deduplication, and standardization. First, noisy data and outlier data in the acquired data are removed. Then, duplicate data is deleted according to the data acquisition timestamp and data characteristics. Finally, the processed data is converted into a structured condition dataset according to a unified numerical range and a unified data format. The structured condition dataset is stored in an orderly manner according to the acquisition timestamp. The stored dataset is then synchronized with the edge terminal model.

[0009] Furthermore, in the edge data acquisition module, the specific process of extracting the operating condition fingerprint and verifying its validity from the structured operating condition dataset is as follows: extract the feature values ​​corresponding to computing power load, energy consumption, voltage and current, and equipment operating status from the structured operating condition dataset, and integrate them to form the operating condition fingerprint; the validity verification first checks whether each feature value is complete, and then compares each feature value with the preset parameter boundary range. If there are no missing features and all feature values ​​are within the preset range, the verification passes. If there are missing features or features that exceed the preset range, the missing items and out-of-range items are recorded and the verification is completed. The verification result is synchronously associated with and stored with the operating condition fingerprint.

[0010] Furthermore, in the edge-inspired matching module, the mathematical expression for the multi-dimensional matching algorithm for working condition fingerprints is: ,in β and γ are the weight coefficients of the algorithm and satisfy the following conditions: This is the operating condition fingerprint feature vector extracted from the structured operating condition dataset. The first pre-stored edge terminator model Feature vectors of different working conditions for and Feature similarity, The validity coefficient of the working condition fingerprint. For working condition fingerprint relative to the first Boundary deviation rate of working condition parameters for similar working conditions For the first The overall matching degree between the type of working condition and the fingerprint of the current working condition.

[0011] Furthermore, in the edge-heuristic matching module, the specific process of locating the target operating condition type and extracting relevant data is as follows: A comprehensive matching degree threshold of 0.8 is set. The comprehensive matching degree of each operating condition type calculated by the multi-dimensional matching algorithm of the operating condition fingerprint is compared with the threshold, and operating condition types with a comprehensive matching degree of not less than 0.8 are selected as candidate operating condition types. The operating condition type with the highest comprehensive matching degree is selected from the candidate operating condition types as the target operating condition type. If the comprehensive matching degree of all operating condition types is less than 0.8, the current matching result is recorded, and the target operating condition type is not located temporarily. After determining the target operating condition type, the historical scheduling parameters, running parameter records, and operating condition parameter boundary ranges corresponding to the operating condition type are extracted. The extracted data is organized in a fixed format and then incorporated into the subsequent encapsulation process along with the operating condition fingerprint and the comprehensive matching degree calculation results.

[0012] Furthermore, the edge-cloud collaborative strategy generation algorithm is used to calculate and generate a set of computing power-power resource allocation strategy parameters, the mathematical expression of which is: ,in The comprehensive matching degree is calculated by the multi-dimensional matching algorithm for working condition fingerprints. The first one retrieved from the optimal strategy gene pool of the entire network Strategy gene vectors for different work conditions This is the performance target weight vector corresponding to the user's preset performance target. For Hadama accumulation, Generate innovation coefficients for strategies. The policy innovation increment vector generated for the lightweight policy generative model. A vector of parameters for the generated computing power-power resource allocation strategy.

[0013] Furthermore, in the cloud-based strategy generation module, the computing power-power resource allocation strategy parameter set includes computing power load allocation ratio, power output voltage adjustment value, power output current adjustment value, power consumption threshold setting value, and computing power scheduling frequency parameter. The parameter set generated by the edge-cloud collaborative strategy generation algorithm is structurally integrated according to the above parameter categories. The verification of the parameter set is a parameter compliance check. First, the allowable range of various parameters for the operation of the power module hardware is preset. Then, each parameter in the parameter set is compared with the corresponding allowable range one by one. If all parameters are within the allowable range, the verification is passed. If any parameter exceeds the allowable range, the verification is deemed to have failed. The deviation value is marked for the failed parameters, and the parameter set is regenerated. The verified parameter set is classified according to the unique identifier of the edge computing node and then sent back to the corresponding edge computing node.

[0014] Furthermore, in the edge-cloud closed-loop update module, the update of the edge meta-model is a supplementary data entry, in which various data in the strategy execution effect dataset are entered into the meta-model according to the working condition type, and the weight coefficients and comprehensive matching degree judgment thresholds of the working condition fingerprint multi-dimensional matching algorithm are adjusted at the same time. The update of the cloud database includes the update of the multimodal working condition data index library and the network-wide optimal strategy gene library. First, the full amount of working condition data synchronized at the edge is entered into the multimodal working condition data index library, and then feature genes are extracted from the newly generated strategy parameter set to supplement the network-wide optimal strategy gene library. All strategy genes in the network-wide optimal strategy gene library are re-clustered according to the working condition type.

[0015] Compared with existing technologies, this edge computing-based intelligent power dynamic adjustment module software has the following advantages: I. This invention establishes an edge-cloud collaborative architecture, deploying multi-dimensional data storage and strategy generation resources in the cloud and lightweight meta-models on edge computing nodes. This enables layered support for data storage and computing tasks. Edge computing nodes preprocess power module operating data and extract operating condition fingerprints, quickly identifying operating condition types through multi-dimensional matching algorithms, filtering out suitable operating condition-related information, and synchronizing it to the cloud. This effectively reduces redundant content during data transmission and improves the efficiency and accuracy of operating condition identification. Based on standardized data synchronized from the edge, the cloud calls upon strategy genes accumulated across the entire network and combines them with the lightweight generation model to output a suitable computing power-power resource allocation strategy. This ensures that the strategy closely matches actual operating requirements. Simultaneously, layered deployment reduces the computational pressure on individual nodes, ensuring the overall smoothness of software operation and providing efficient support for the dynamic adjustment of power modules.

[0016] Second, this invention constructs a complete edge-cloud closed-loop update mechanism. After executing adjustment commands, the edge computing node uses the execution effect dataset generated by the running status to update the local meta-model and matching algorithm settings. At the same time, it synchronizes all operating condition data to the cloud, promoting the continuous optimization of the cloud database and the calculation benchmark of the strategy generation algorithm. This two-way update mode allows the software to adapt to the changes in operating conditions during the operation of the power module in real time, continuously optimizing the allocation logic of computing power and power consumption, so that the power adjustment always fits the actual operating state. The independent and lightweight operation at the edge ensures a fast response to data processing and command execution, avoiding the latency problems caused by centralized computing. The global resource integration in the cloud provides comprehensive data support for strategy generation, realizing the organic unity of local response speed and global adjustment effect, and improving the stability and resource utilization efficiency of the power module operation.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 A flowchart illustrating the software workflow of an edge computing-based intelligent power dynamic adjustment module; Figure 2 This is a diagram showing the overall software framework of an edge computing-based intelligent power dynamic adjustment module. Figure 3 This is a flowchart illustrating the software-cloud strategy interaction process for an edge-computing-based intelligent power dynamic adjustment module. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Example 1: Industrial intelligent manufacturing scenario (power regulation of production equipment in automotive parts processing plant).

[0022] Automotive parts processing plants have multiple stamping, welding, and assembly production lines. Each production line is equipped with multiple intelligent power modules to provide stable power support for the processing equipment control system. Different production processes have significantly different computing power and power consumption requirements. The equipment's operating status changes dynamically during production, necessitating dynamic adjustment of the intelligent power supply to achieve an optimal match between computing power and power consumption, ensuring production efficiency while reducing energy consumption. The software in this invention can precisely adapt to the dynamic adjustment requirements of the power modules in this scenario. Specific steps are as follows: Figure 1 As shown.

[0023] A multimodal operating condition data index library, the best-in-class strategy gene library, and a lightweight strategy generative model are deployed in the cloud. The multimodal operating condition data index library adopts a distributed storage architecture, dividing data storage into partitions according to production processes such as stamping, welding, and assembly. Each partition contains four types of data: textual operating condition descriptions, numerical operating parameters, waveform energy consumption curves, and fingerprint-based operating characteristics. All data is vectorized, and a two-layer data retrieval index is established according to operating condition type and parameter dimensions. During the initialization phase, basic data classification rules, data vectorization processing, and standard retrieval index establishment rules are configured, making the classification and storage of operating condition data for different processes clearer and subsequent data retrieval more efficient and accurate. The best-in-class strategy gene library and the multimodal operating condition data index library achieve data interaction and connection. During initialization, feature extraction, feature filtering, and gene clustering are performed on the historical computing power-power scheduling strategies of the power module. The processed strategy genes are classified and stored in corresponding partitions according to operating condition types such as stamping and welding. Addition, modification, and retrieval permissions for strategy genes are configured, and the update cycle and calculation standards for gene clustering are set to ensure standardized management of strategy genes and more convenient subsequent retrieval and use. The lightweight strategy-generative model is deployed based on the power supply module's computing power-power scheduling scenario in this factory. It includes a strategy gene retrieval interface, parameter calculation module, and strategy generation module. During initialization, the model's inference computing rate, data processing dimensions, and strategy generation parameter output format are configured. Data interface adaptation with two other libraries is completed, and the model's startup trigger conditions and data interaction protocol are configured to ensure the model can quickly respond to scheduling needs and smoothly complete data interaction and strategy generation. On the edge computing nodes corresponding to the power supply modules of each production line, an independent lightweight operating condition type-strategy performance surface meta-model is deployed and initialized. The meta-model stores operating condition type classification features divided by computing power load level, voltage fluctuation range, and energy consumption interval. Storing these features in a fixed data format makes feature retrieval more efficient. Historical scheduling parameters and operating parameter records corresponding to each operating condition type are stored in numerical tables for easy and quick reference. Operating condition parameter boundary ranges are set for each operating condition type, clearly defining the maximum, minimum, and standard values ​​of each operating parameter, providing a clear basis for subsequent data verification and operating condition matching. Independent lightweight deployment reduces the computing power consumption of edge nodes and improves operational response speed.

[0024] Edge computing nodes collect real-time operational data from corresponding power modules, including computing load data, energy consumption data, power output voltage and current data, and equipment operating status data during processing. The collected data undergoes preprocessing steps including cleaning, deduplication, and standardization. First, noisy and outlier data are removed. Then, duplicate data is deleted based on the data collection timestamp and characteristics. Finally, the processed data is converted into a structured work condition dataset according to a unified numerical range and data format, stored in order of collection timestamp, and synchronized with the edge's work condition type-strategy performance surface meta-model. This ensures a more standardized and unified dataset, providing high-quality, standardized data support for subsequent work condition matching. Feature values ​​corresponding to the operating status of computing power load, energy consumption, voltage, current, and equipment are extracted from the structured operating condition dataset and integrated to form an operating condition fingerprint. When performing validity verification, the completeness of each feature value is checked first, and then each feature value is compared with the preset parameter boundary range. If there are no missing features and all feature values ​​are within the preset range, the verification is passed. If there are missing features or features that exceed the preset range, the missing items and out-of-range items are recorded and the verification is completed. The verification results are synchronously associated with the operating condition fingerprint and stored to ensure that the operating condition fingerprint can truly reflect the operating status of the power module and avoid invalid data from participating in the subsequent matching process.

[0025] The edge computing node runs a multi-dimensional matching algorithm for operating condition fingerprints to calculate the comprehensive matching degree between the current operating condition fingerprint and the features of each operating condition type in the meta-model. The mathematical expression of the multi-dimensional matching algorithm for operating condition fingerprints is: ,in β and γ are the weight coefficients of the algorithm and satisfy the following conditions: This is the operating condition fingerprint feature vector extracted from the structured operating condition dataset. The first pre-stored edge terminator model Feature vectors of different working conditions for and Feature similarity, The validity coefficient of the working condition fingerprint. For working condition fingerprint relative to the first Boundary deviation rate of working condition parameters for similar working conditions For the first The comprehensive matching degree between the operating condition type and the current operating condition fingerprint is calculated through multi-dimensional considerations to ensure that the matching results are more closely aligned with actual operating conditions. The comprehensive matching degree of each operating condition type is compared with a preset 0.8 threshold. Operating condition types with a comprehensive matching degree of not less than 0.8 are selected as candidate operating condition types. The operating condition type with the highest comprehensive matching degree from the candidate types is selected as the target operating condition type. If the comprehensive matching degree of all operating condition types is less than 0.8, the current matching result is recorded, but the target operating condition type is not yet located, ensuring accurate target operating condition type location and avoiding mismatches that could affect the adjustment effect. After determining the target operating condition type, the historical scheduling parameters and running parameters corresponding to that operating condition type are extracted, and the boundary range of the operating condition parameters is recorded. This data is organized in a fixed format and then packaged together with the comprehensive matching degree calculation result of the operating condition fingerprint, and synchronized to the cloud. This allows the cloud to obtain complete and comprehensive operating condition-related information, providing sufficient data support for subsequent strategy generation.

[0026] The cloud receives and parses the data encapsulated at the edge, and runs the edge-cloud collaboration strategy generation algorithm. This algorithm calculates and generates a set of computing power-power resource allocation strategy parameters, the mathematical expression of which is: ,in The comprehensive matching degree is calculated by the multi-dimensional matching algorithm for working condition fingerprints. The first one retrieved from the optimal strategy gene pool of the entire network Strategy gene vectors for different work conditions This is the performance target weight vector corresponding to the user's preset performance target. For Hadama accumulation, Generate innovation coefficients for strategies. The policy innovation increment vector generated for the lightweight policy generative model. The generated computing power-power resource allocation strategy parameter set vector, combined with the comprehensive matching degree calculation results, retrieves the corresponding working condition type strategy gene from the optimal strategy gene library of the entire network. This generates a computing power-power resource allocation strategy parameter set containing computing power load allocation ratio, power output voltage adjustment value, power output current adjustment value, power consumption threshold setting value, and computing power scheduling frequency parameters, ensuring that the generated strategy parameters accurately adapt to the current working condition requirements. The generated parameter set undergoes parameter compliance verification. First, the allowable ranges for various parameters of the factory's power module hardware operation are preset. Then, each parameter in the parameter set is compared with its corresponding allowable range. If all parameters are within the allowable range, the verification passes. If any parameter exceeds the allowable range, the verification fails. The deviation value is marked for the failed parameters, and the parameter set is regenerated to ensure that the parameter set meets the hardware operating requirements, avoiding equipment damage or impact on operational stability due to abnormal parameters. After the verified parameter set is classified according to the unique identifier of the edge computing node, it is sent back to the corresponding edge computing node to ensure that each edge node can accurately receive its corresponding adjustment parameters. Specific steps are as follows... Figure 3 As shown.

[0027] Edge computing nodes parse the received policy parameter set and issue adjustment commands to the corresponding power modules. They adjust the power allocation of each processing unit according to the computing load distribution ratio and correct the voltage output according to the power output voltage adjustment value, allowing the power module's operating status to quickly adapt to the current production process requirements. Simultaneously, they monitor the power module's operating status in real time, collecting adjusted computing load data, energy consumption data, voltage and current data, etc., to form an execution effect dataset, providing a comprehensive understanding of the actual operating situation after adjustment. This execution effect dataset is used to supplement the edge-side working condition type-policy performance surface meta-model, enriching the meta-model's data reserves. At the same time, the weight coefficients and comprehensive matching degree judgment threshold of the working condition fingerprint multi-dimensional matching algorithm are adjusted to make subsequent working condition matching more accurate. All operating condition data is synchronized to the cloud, where it is entered into a multimodal operating condition data index library to enrich the data resources of the index library. Feature genes are then extracted from the newly generated strategy parameter set to supplement the optimal strategy gene library of the entire network. All strategy genes in the optimal strategy gene library of the entire network are re-clustered according to operating condition type. The strategy gene vector strategy innovation incremental calculation benchmark of the end-cloud collaborative strategy generation algorithm is updated synchronously, so that the cloud database and algorithm are continuously optimized. The subsequent generated strategies are more in line with the actual operating status of the equipment, realizing the iterative upgrade of end-cloud data and algorithms.

[0028] In summary, in the power supply regulation scenario of automotive parts processing plant production equipment, the software of this invention is systematically implemented through an edge-cloud collaborative architecture. Three core libraries and models are deployed in the cloud, while independent lightweight meta-models are deployed on edge nodes, laying the foundation for efficient regulation. After collecting power supply operation data, edge nodes preprocess and extract and verify operating condition fingerprints to form standardized data support. Using a multi-dimensional matching algorithm based on operating condition fingerprints, the target operating condition is accurately located. The cloud generates a compliant strategy parameter set through an edge-cloud collaborative strategy, and edge nodes execute the regulation and provide feedback on the results. The closed-loop edge-cloud update continuously optimizes data and algorithms, allowing power supply regulation to adapt to the needs of different production processes. This ensures stable equipment operation and production efficiency while achieving optimal matching of computing power and power consumption, effectively reducing energy consumption.

[0029] Example 2: Data center server cluster scenario (power regulation of internet e-commerce data centers).

[0030] Internet e-commerce companies typically deploy thousands of servers in their data centers, handling tasks such as product display, order processing, payment settlement, and backend data backup. The computing power, load, and power consumption requirements of these servers vary significantly across different business scenarios. To avoid insufficient computing power during peak hours and wasted energy during off-peak hours, intelligent power dynamic adjustment is needed to optimize resource allocation. The software in this invention can meet the dynamic adjustment requirements of the power supply modules in this data center server cluster. Specific steps are as follows: Figure 2As shown.

[0031] A multimodal operating condition data index library, the best-in-class strategy gene library, and a lightweight strategy generative model are deployed in the cloud. The multimodal operating condition data index library adopts a distributed storage architecture, dividing data storage into partitions according to business scenarios such as daily access, peak sales events, and data backup. Each partition contains four types of data: textual operating condition descriptions, numerical operating parameters, waveform energy consumption curves, and fingerprint-based operating features. All inbound data undergoes feature vectorization processing, and a two-layer data retrieval index is established based on operating condition type parameters. During the initialization phase, basic data classification rules, data vectorization processing, and standard retrieval index establishment rules are configured, making the classification of operating condition data for different business scenarios clearer and subsequent data retrieval more efficient. The best-in-class strategy gene library and the multimodal operating condition data index library achieve data interaction and connection. During initialization, feature extraction, feature filtering, and gene clustering are performed on the historical computing power-power scheduling strategies of the server power module. The processed strategy genes are classified and stored in corresponding partitions according to the operating condition types corresponding to different business scenarios. Addition, modification, and retrieval permissions for strategy genes are configured, and the update cycle and calculation standards for gene clustering are set to ensure orderly management of strategy genes and more convenient subsequent retrieval. The lightweight policy-generative model is deployed based on the computing power-power scheduling scenario of the data center servers. It includes a policy gene retrieval interface, parameter calculation module, and policy generation module. During initialization, the model's inference computing rate, data processing dimensions, and policy generation parameter output format are configured. Data interface adaptation with two other libraries is completed, and the model's startup trigger conditions and data interaction protocol are configured to ensure the model can quickly respond to the scheduling needs of different business scenarios and smoothly complete data interaction and policy generation. On the edge computing node corresponding to each server, an independent lightweight working condition type-policy performance surface meta-model is deployed and initialized. The meta-model stores working condition type classification features based on computing power load level, voltage fluctuation range, and energy consumption interval. Data is stored in a fixed format for easy retrieval. Historical scheduling parameters and operating parameter records for each working condition type are stored in numerical tables for easy reference. Working condition parameter boundary ranges are set for each working condition type, and the maximum, minimum, and standard values ​​of each operating parameter are clearly defined, providing an accurate basis for subsequent data verification and working condition matching. Independent lightweight deployment reduces edge node resource consumption and improves operating efficiency.

[0032] Edge computing nodes collect real-time operational data from the corresponding server power modules, including CPU computing power load data, memory usage and energy consumption data, power output voltage and current data, and server operating status data. The collected data undergoes preprocessing steps including cleaning, deduplication, and standardization. First, noisy and outlier data are removed. Then, duplicate data is deleted based on the data collection timestamp and data characteristics. Finally, the processed data is converted into a structured operating condition dataset according to a unified numerical range and data format, stored in order of collection timestamp, and synchronized with the edge's operating condition type-policy performance surface meta-model. This ensures a standardized and consistent dataset, providing a reliable data foundation for subsequent operating condition matching. Feature values ​​corresponding to the computing power load, energy consumption, voltage, current, and equipment operating status are extracted from the structured operating condition dataset and integrated to form an operating condition fingerprint. When performing validity verification, the completeness of each feature value is checked first, and then each feature value is compared with the preset parameter boundary range. If there are no missing features and all feature values ​​are within the preset range, the verification is passed. If there are missing features or features that exceed the preset range, the missing items and out-of-range items are recorded and the verification is completed. The verification results are synchronously associated with the operating condition fingerprint and stored to ensure that the operating condition fingerprint can truly reflect the operating status of the server power module and avoid invalid data interfering with the subsequent matching process.

[0033] The edge computing node runs a multi-dimensional matching algorithm for operating condition fingerprints to calculate the comprehensive matching degree between the current operating condition fingerprint and the features of each operating condition type in the meta-model. The mathematical expression of the multi-dimensional matching algorithm for operating condition fingerprints is: ,in β and γ are the weight coefficients of the algorithm and satisfy the following conditions: This is the operating condition fingerprint feature vector extracted from the structured operating condition dataset. The first pre-stored edge terminator model Feature vectors of different working conditions for and Feature similarity, The validity coefficient of the working condition fingerprint. For working condition fingerprint relative to the first Boundary deviation rate of working condition parameters for similar working conditions For the first The comprehensive matching degree between the type of work condition and the current work condition fingerprint is considered from multiple dimensions to make the matching results more accurate. The comprehensive matching degree of each work condition type is compared with a preset threshold of 0.8, and work condition types with a comprehensive matching degree of not less than 0.8 are selected as candidate work condition types. From the candidate work condition types, the work condition type with the highest comprehensive matching degree is selected as the target work condition type. If the comprehensive matching degree of all work condition types is less than 0.8, the target work condition type is not located after recording the current matching result, ensuring accurate location of the target work condition type and providing accurate direction for subsequent strategy generation. After the target work condition type is determined, the historical scheduling parameters and running parameters corresponding to the work condition type are extracted, the boundary range of the work condition parameters is recorded, and these data are organized in a fixed format. Then, they are packaged together with the comprehensive matching degree calculation result of the work condition fingerprint and synchronized to the cloud, so that the cloud can obtain comprehensive and complete work condition information and provide sufficient data support for strategy generation.

[0034] The cloud receives and parses the data encapsulated at the edge, and runs the edge-cloud collaboration strategy generation algorithm. This algorithm calculates and generates a set of computing power-power resource allocation strategy parameters, the mathematical expression of which is: ,in The comprehensive matching degree is calculated by the multi-dimensional matching algorithm for working condition fingerprints. The first one retrieved from the optimal strategy gene pool of the entire network Strategy gene vectors for different work conditions This is the performance target weight vector corresponding to the user's preset performance target. For Hadama accumulation, Generate innovation coefficients for strategies. The policy innovation increment vector generated for the lightweight policy generative model. The generated computing power-power resource allocation strategy parameter set vector, combined with the comprehensive matching degree calculation results, retrieves the corresponding working condition type strategy gene from the network-wide optimal strategy gene library. This generates a computing power-power resource allocation strategy parameter set containing parameters such as computing power load allocation ratio, power output voltage adjustment value, power output current adjustment value, power consumption threshold setting value, and computing power scheduling frequency parameters. This ensures that the strategy parameters accurately adapt to the server's current business scenario's operational needs. The generated parameter set undergoes parameter compliance verification. First, the allowable ranges for various parameters of the data center server power module hardware operation are preset. Then, each parameter in the parameter set is compared with its corresponding allowable range. If all parameters are within the allowable range, the verification passes. If any parameter exceeds the allowable range, the verification fails. The parameter set is regenerated after marking the deviation value for the failed parameter to prevent server hardware damage or operational abnormalities due to parameter exceeding limits. After the verified parameter set is categorized according to the unique identifier of the edge computing node, it is sent back to the corresponding edge computing node to ensure that each server can accurately receive the adjustment parameters adapted to its own needs.

[0035] Edge computing nodes parse the received policy parameter set and issue adjustment instructions to the corresponding server power modules. During peak sales periods, power supply to core business servers is increased according to the computing load allocation ratio, while energy consumption of non-core servers is limited according to power consumption thresholds, making server power allocation more reasonable and adapting to business scenario requirements. Simultaneously, the operating status of the power modules is monitored in real time, collecting adjusted computing load data, energy consumption data, voltage and current data, etc., to form an execution effect dataset, providing a comprehensive understanding of the adjusted operating conditions. This execution effect dataset is used to supplement the edge-side working condition type-policy performance surface meta-model, enriching the meta-model data reserve. At the same time, the weight coefficients and comprehensive matching degree judgment thresholds of the working condition fingerprint multi-dimensional matching algorithm are adjusted to improve the accuracy of subsequent working condition matching. All operating condition data is synchronized to the cloud, where it is entered into a multimodal operating condition data index library to enrich the index library resources. Feature genes are then extracted from the newly generated strategy parameter set to supplement the optimal strategy gene library across the entire network. All strategy genes in the optimal strategy gene library are re-clustered according to operating condition type, and the strategy gene vector strategy innovation incremental calculation benchmark of the end-cloud collaborative strategy generation algorithm is updated synchronously. This allows the cloud database and algorithm to be continuously optimized, and the subsequently generated strategies are more in line with the actual operating status of the server, realizing the iterative upgrade of end-cloud data and algorithms.

[0036] In summary, in the scenario of power regulation for internet e-commerce data center servers, the software of this invention plays a core role by relying on an edge-cloud collaborative model. The cloud stores multiple types of data and strategy genes according to business scenarios, and a lightweight model ensures rapid strategy generation. Edge nodes deploy dedicated meta-models to collect server power data in real time and complete standardized processing and operational condition fingerprint verification. A multi-dimensional matching algorithm for operational condition fingerprints identifies the target operational condition, and the cloud generates a set of strategy parameters adapted to business needs, which is then distributed after compliance verification. Edge nodes execute adjustments and synchronize all data; the edge and cloud sides update and optimize the model, algorithm, and database, enabling the server power supply to achieve reasonable resource allocation under different business scenarios, avoiding insufficient computing power during peak periods and energy waste during off-peak periods, and ensuring the stable and efficient operation of the data center.

[0037] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A smart power dynamic adjustment module software based on edge computing, characterized in that, The software includes: Edge-cloud system construction module: Deploy a multimodal working condition data index library, the network-wide optimal strategy gene library, and a lightweight strategy generative model in the cloud; deploy working condition type-strategy performance surface meta-models in each edge computing node; and initialize the cloud and edge computing nodes simultaneously. Edge data acquisition and extraction module: Edge computing nodes collect power module operation data, preprocess it to form a structured operating condition dataset, extract operating condition fingerprints from the structured operating condition dataset and perform validity verification, and synchronize the dataset to the operating condition type-strategy performance surface meta-model at the edge to provide standardized data for operating condition matching; Edge-inspired matching module: Edge computing nodes run a multi-dimensional matching algorithm for working condition fingerprints, calculate the comprehensive matching degree between the working condition fingerprint and the working condition type features in the working condition type-policy performance surface meta-model, locate the target working condition type and extract relevant data, encapsulate multiple types of working condition related information and synchronize them to the cloud; Cloud-based strategy generation module: The cloud receives and parses the encapsulated data from the edge, runs the cloud-edge collaborative strategy generation algorithm, retrieves the strategy gene based on the matching degree calculation result, generates a set of computing power-power resource allocation strategy parameters, and after verification, sends the parameter set back to the corresponding edge computing node. Edge-cloud closed-loop update module: Edge computing nodes parse parameter sets and issue adjustment instructions to power modules, monitor operating status to form execution effect datasets, update edge terminal models and adjust parameters of multi-dimensional matching algorithm for operating condition fingerprints, synchronize all operating condition data to the cloud, update cloud database and edge-cloud collaboration strategy to generate algorithm calculation benchmarks, and realize iterative optimization of edge-cloud data and algorithms.

2. The edge computing-based intelligent power dynamic adjustment module software according to claim 1, characterized in that, The specific contents of the multimodal operating condition data index library, the network-wide optimal strategy gene library, and the lightweight strategy generative model in the edge-cloud system construction module are as follows: The multimodal operating condition data index library is deployed using a distributed storage architecture. The library is divided into data storage partitions according to the operating scenarios of the power modules. Each partition contains four types of data: text-based operating condition descriptions, numerical operating parameters, waveform energy consumption curves, and fingerprint-based operating characteristics of the power modules. All data entering the library is processed by feature vectorization. A two-layer data retrieval index is established according to operating condition type and parameter dimension. The initialization phase completes the configuration of basic data classification rules, data vectorization processing standards, and retrieval index establishment rules. The optimal strategy gene library of the entire network and the multimodal operating condition data index library form a data interaction connection. During the initialization phase, the historical computing power-power scheduling strategy of the power module is subjected to feature extraction, feature screening and gene clustering. The processed strategy genes are classified and stored in the corresponding partitions of the library according to the operating condition type. The operation permissions for adding, modifying and retrieving strategy genes are configured for each partition, and the update cycle and calculation standard of gene clustering are set. The lightweight strategy generation model is deployed based on the power module computing power-power scheduling scenario. The model is equipped with a strategy gene retrieval interface, a parameter calculation module, and a strategy generation module. During the initialization phase, the model's inference calculation rate, data processing dimension, and strategy generation parameter output format are configured. The model is adapted to the multimodal operating condition data index library and the network-wide optimal strategy gene library through data interface adaptation. The model's start-up trigger conditions and data interaction protocol are configured.

3. The edge computing-based intelligent power dynamic adjustment module software according to claim 1, characterized in that, In the edge-cloud system construction module, the working condition type-policy performance surface meta-models deployed on each edge computing node are all independently and lightweightly deployed. After deployment, they are initialized synchronously. Specifically, the meta-model stores the working condition type classification features, the policy performance distribution law corresponding to each working condition type, and the boundary range of working condition parameters. The working condition type classification features are divided into different dimensions according to the computing power load level, voltage fluctuation range, and energy consumption range. Each dimension feature is stored in a fixed data format. The policy performance distribution law is stored in the form of a numerical table, which contains the historical scheduling parameters and running parameter records corresponding to each working condition type. The boundary ranges of operating parameters are set separately for each type of operating condition, and the maximum, minimum and standard values ​​of each operating parameter are clearly defined.

4. The edge computing-based intelligent power dynamic adjustment module software according to claim 1, characterized in that, In the edge data acquisition module, the preprocessing of the power module operation data is performed in sequence according to the steps of cleaning, deduplication, and standardization. First, noisy data and outlier data in the acquired data are removed. Then, duplicate data is deleted according to the data acquisition timestamp and data characteristics. Finally, the processed data is converted into a structured condition dataset according to a unified numerical range and a unified data format. The structured condition dataset is stored in an orderly manner according to the acquisition timestamp. The stored dataset is then connected to the edge terminal model for data synchronization.

5. The edge computing-based intelligent power dynamic adjustment module software according to claim 1, characterized in that, In the edge data acquisition module, the specific process of extracting operating condition fingerprints and verifying their validity from the structured operating condition dataset is as follows: extract the feature values ​​corresponding to computing power load, energy consumption, voltage and current, and equipment operating status from the structured operating condition dataset, and integrate them to form an operating condition fingerprint; The validity check first verifies whether each feature value is complete, and then compares each feature value with the preset parameter boundary range. If there are no missing features and all feature values ​​are within the preset range, the check passes. If there are missing features or features that exceed the preset range, the missing items or items that exceed the range are recorded and the check is completed. The check results are stored synchronously with the operating condition fingerprint.

6. The intelligent power dynamic adjustment module software based on edge computing according to claim 1, characterized in that, In the edge-inspired matching module, the mathematical expression for the multi-dimensional matching algorithm for working condition fingerprints is: ,in β and γ are the weight coefficients of the algorithm and satisfy the following conditions: This is the operating condition fingerprint feature vector extracted from the structured operating condition dataset. The first pre-stored edge terminator model Feature vectors of different working conditions for and Feature similarity, The validity coefficient of the working condition fingerprint. For working condition fingerprint relative to the first Boundary deviation rate of working condition parameters for similar working conditions For the first The overall matching degree between the type of working condition and the fingerprint of the current working condition.

7. The intelligent power dynamic adjustment module software based on edge computing according to claim 1, characterized in that, In the edge-heuristic matching module, the specific process of locating the target working condition type and extracting relevant data is as follows: set the comprehensive matching degree threshold to 0.8, compare the comprehensive matching degree of each working condition type calculated by the working condition fingerprint multi-dimensional matching algorithm with the threshold, and select the working condition types with a comprehensive matching degree of not less than 0.8 as candidate working condition types; select the working condition type with the highest comprehensive matching degree from the candidate working condition types as the target working condition type; if the comprehensive matching degree of all working condition types is less than 0.8, record the current matching result and temporarily do not locate the target working condition type. After determining the target operating condition type, extract the historical scheduling parameters, operation parameter records, and operating condition parameter boundary ranges corresponding to the operating condition type. After the extracted data is organized in a fixed format, it is included in the subsequent encapsulation process along with the operating condition fingerprint and the comprehensive matching degree calculation results.

8. The edge computing-based intelligent power dynamic adjustment module software according to claim 1, characterized in that, The edge-cloud collaboration strategy generation algorithm is used to calculate and generate a set of computing power-power resource allocation strategy parameters, and its mathematical expression is: ,in The comprehensive matching degree is calculated by the multi-dimensional matching algorithm for working condition fingerprints. The first one retrieved from the optimal strategy gene pool of the entire network Strategy gene vectors for different work conditions This is the performance target weight vector corresponding to the user's preset performance target. For Hadama accumulation, Generate innovation coefficients for strategies. The policy innovation increment vector generated for the lightweight policy generative model. A vector of parameters for the generated computing power-power resource allocation strategy.

9. The intelligent power dynamic adjustment module software based on edge computing according to claim 1, characterized in that, In the cloud-based strategy generation module, the computing power-power resource allocation strategy parameter set includes computing power load allocation ratio, power output voltage adjustment value, power output current adjustment value, power consumption threshold setting value, and computing power scheduling frequency parameter. The parameter set generated by the edge-cloud collaborative strategy generation algorithm is structurally integrated according to the above parameter categories. The parameter set verification is a parameter compliance check. First, the allowable range of various parameters for the operation of the power module hardware is preset. Then, each parameter in the parameter set is compared with the corresponding allowable range one by one. If all parameters are within the allowable range, the verification is passed. If any parameter exceeds the allowable range, the verification is deemed to have failed. The deviation value is marked for the failed parameters, and the parameter set is regenerated. The verified parameter set is classified according to the unique identifier of the edge computing node and then sent back to the corresponding edge computing node.

10. The intelligent power dynamic adjustment module software based on edge computing according to claim 1, characterized in that, In the edge-cloud closed-loop update module, the update of the edge meta-model is a supplementary data entry. The various data in the strategy execution effect dataset are entered into the meta-model according to the working condition type. At the same time, the weight coefficients and comprehensive matching degree judgment thresholds of the working condition fingerprint multi-dimensional matching algorithm are adjusted. The update of the cloud database includes the update of the multimodal working condition data index library and the network-wide optimal strategy gene library. First, the full amount of working condition data synchronized from the edge terminal is entered into the multimodal working condition data index library. Then, feature genes are extracted from the newly generated strategy parameter set to supplement the network-wide optimal strategy gene library. All strategy genes in the network-wide optimal strategy gene library are re-clustered according to working condition type.

Images