Self-adaptive intelligent edge calculation method and system

Abstract

The invention discloses a self-adaptive intelligent edge calculation method and system, and relates to the technical field of edge calculation. The method comprises the following steps: configuring an intelligent edge gateway for power transformation equipment in a new energy smart station; the intelligent edge gateway receives a multi-modal data stream returned by the multi-modal sensing unit through a heterogeneous industrial communication protocol, unifies a data format, executes equipment state anomaly detection and outputs a diagnosis result; the gateway performs abnormality diagnosis filtering to obtain a key abnormal event; and the cloud control center receives the abnormal event, executes and issues a dynamic control strategy, and generates a log file to record an operation event. The technical problems that integration is complex and real-time processing is difficult due to equipment protocol diversification and insufficient edge computing power in a new energy station are solved, and the technical effects that the system complexity is reduced through protocol conversion and local intelligent processing, the real-time response capability is improved, and the reliability and safety of the system are guaranteed through cloud edge cooperation are achieved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of edge computing, and in particular to an adaptive intelligent edge computing method and system. BACKGROUND

[0002] With the rapid development of new generation technologies such as the Internet of Things, artificial intelligence and edge computing, the traditional centralized computing architecture faces many challenges, especially under the background of limited bandwidth, heterogeneous devices, data transmission delay and security problems. In special environments such as new energy stations, power transformation devices and other intelligent devices often require efficient and real-time data processing capabilities to ensure the stability and security of the system. However, due to the different communication protocols and data formats between devices, the integration and compatibility of the system are affected. At the same time, the computing power and storage capacity of edge devices are limited, which makes it difficult to support the data processing needs of a large number of devices and effectively cope with various complex industrial application scenarios, so that the traditional computing architecture is difficult to provide sufficient performance support when facing dynamic changes and diversification of needs. SUMMARY

[0003] The application provides an adaptive intelligent edge computing method and system, which solves the technical problems of integration complexity and real-time processing difficulty caused by device protocol diversification and insufficient edge computing power in new energy stations.

[0004] The first aspect of the application provides an adaptive intelligent edge computing method, which comprises the following steps: P sets of intelligent edge gateways are configured for P sets of power transformation device entities in a new energy smart station, wherein the P sets of intelligent edge gateways are in communication connection with P sets of multi-modal sensing units, and the P sets of multi-modal sensing units are mapped and deployed in the P sets of power transformation device entities; after the P sets of intelligent edge gateways receive P sets of multi-modal data streams transmitted in real time by the P sets of multi-modal sensing units through heterogeneous industrial communication protocols, the P sets of intelligent edge gateways uniformly convert the data formats through a protocol conversion module, perform device state anomaly detection, and output P sets of device state diagnosis results; the P sets of intelligent edge gateways perform abnormal diagnosis filtering on the P sets of device state diagnosis results to obtain P key abnormal diagnosis events; after a cloud control center receives the P key abnormal diagnosis events sent by the P sets of intelligent edge gateways and performs dynamic device control strategy matching and delivery, the cloud control center generates a log file to record operation events, wherein the P sets of intelligent edge gateways are connected to the cloud control center through a standard Ethernet.

[0005] The second aspect of the application provides an adaptive intelligent edge computing system, which comprises the following steps: The gateway configuration component configures P intelligent edge gateways for P groups of power transformation equipment entities in a new energy smart station, wherein the P intelligent edge gateways are in communication connection with P groups of multi-modal sensing units, and the P groups of multi-modal sensing units are deployed in the P groups of power transformation equipment entities; the abnormality detection component, after the P intelligent edge gateways receive P groups of multi-modal data streams transmitted in real time by the P groups of multi-modal sensing units through heterogeneous industrial communication protocols, unifies the data formats through a protocol conversion module, performs equipment state abnormality detection, and outputs P groups of equipment state diagnosis results; the abnormality filtering component performs abnormality diagnosis filtering on the P groups of equipment state diagnosis results, and obtains P key abnormality diagnosis events; the event recording component, after the cloud control center receives the P key abnormality diagnosis events sent by the P intelligent edge gateways, performs dynamic equipment control strategy matching and delivery, generates a log file, and records operation events, wherein the P intelligent edge gateways are connected to the cloud control center through a standard Ethernet.

[0006] The one or more technical solutions provided in the present application have at least the following technical effects or advantages: First, in the new energy smart station, an intelligent edge gateway is configured for each group of power transformation equipment. These gateways communicate with multi-modal sensing units deployed on the power transformation equipment. Each sensing unit is responsible for collecting different types of data and transmitting it in real time to the intelligent edge gateway. Then, the gateway receives and processes these data streams uniformly through support for multiple industrial communication protocols, performs equipment state detection, and generates diagnosis results. The intelligent edge gateway further analyzes the diagnosis results and filters out key abnormal events. Then, the cloud control center receives these abnormal events, performs equipment management based on dynamic control strategies, and issues corresponding control instructions while generating log records of the operation process. All intelligent edge gateways are connected to the cloud control center through a standard Ethernet, ensuring data transmission and coordinated management of equipment. BRIEF DESCRIPTION OF DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0008] Figure 1 A self-adaptive intelligent edge computing method flowchart is provided for the embodiments of the present application.

[0009] Figure 2 A self-adaptive intelligent edge computing system structure diagram is provided for the embodiments of the present application.

[0010] Reference signs: gateway configuration component 11, anomaly detection component 12, anomaly filtering component 13, event recording component 14. DETAILED DESCRIPTION

[0011] To further illustrate the technical means and effects adopted by the present application to achieve the predetermined inventive objectives, and the specific embodiments, structures, features and effects thereof according to the present application, a detailed description is provided below in conjunction with the accompanying drawings and preferred embodiments.

[0012] In one embodiment, as shown in the accompanying drawings, the present application provides a self-adaptive intelligent edge computing method, which comprises the following steps: Figure 1 In one embodiment, as shown in the accompanying drawings, the present application provides a self-adaptive intelligent edge computing method, which comprises the following steps: In one embodiment, as shown in the accompanying drawings, the present application provides a self-adaptive intelligent edge computing method, which comprises the following steps:

[0013] In one embodiment, as shown in the accompanying drawings, the present application provides a self-adaptive intelligent edge computing method, which comprises the following steps:

[0014] In one embodiment, as shown in the accompanying drawings, the present application provides a self-adaptive intelligent edge computing method, which comprises the following steps: In one embodiment, as shown in the accompanying drawings, the present application provides a self-adaptive intelligent edge computing method, which comprises the following steps: In one embodiment, as shown in the accompanying drawings, the present application provides a self-adaptive intelligent edge computing method, which comprises the following steps:

[0015] Preferably, in the new energy smart station, in order to realize efficient monitoring and management of power transformation equipment, first of all, all power transformation equipment in the station is classified and summarized, based on the model, function, use environment and other factors of the equipment, the equipment is divided into P different types according to the type of the equipment, each type of power transformation equipment represents a kind of power transformation equipment entity, forming P groups of power transformation equipment entities, for example, transformer entity, switch device entity, circuit breaker entity, etc. Subsequently, according to the fault state attribute of each type of power transformation equipment, the corresponding fault monitoring demand matching is carried out. The fault state of different equipment types may be different, therefore, for each equipment type, according to its historical fault data and domain knowledge, the state parameters that change greatly when the equipment fails are identified, such as temperature, voltage, current, vibration, etc., so as to determine the key parameters that need to be monitored for each equipment, for example, for transformer, temperature, oil level, etc. may need to be monitored; while for switch device, switch state, operating current, etc. may need to be monitored. Then, according to the fault monitoring demand and key parameters of each power transformation equipment, the appropriate sensor type and configuration are determined, so as to formulate P multi-modal sensor configurations for P groups of power transformation equipment types, so that they can collect multiple types of data at the same time. Then, according to the determined P multi-modal sensor configurations, the corresponding sensor installation and deployment are carried out. Each group of power transformation equipment entities will install the corresponding sensor unit according to its corresponding sensor configuration, to ensure that the required monitoring parameters of each equipment type can be effectively collected. During the installation process, it is ensured that each sensor can accurately cover the required equipment part and is closely related to the running state of the equipment. Through the above steps, P groups of multi-modal sensor units will be finally obtained, each of which is configured and installed according to the type and fault monitoring demand of the specific power transformation equipment, thereby providing basic data support for subsequent equipment state monitoring and anomaly detection.

[0016] After the P intelligent edge gateways receive the P groups of multi-modal data streams returned by the P groups of multi-modal sensor units in real time through heterogeneous industrial communication protocols, the protocol conversion module unifies the data format and executes equipment state anomaly detection, and outputs P groups of equipment state diagnosis results. In one embodiment, in a new energy smart station, P intelligent edge gateways are connected in communication with P groups of multi-modal sensing units. Each multi-modal sensing unit is responsible for real-time monitoring of the operating state of the equipment, and collects multiple types of data such as temperature, humidity, voltage, current, etc. and transmits them back to the corresponding intelligent edge gateway through different industrial communication protocols, which may include Modbus, OPC UA, Ethernet / IP, etc. After each intelligent edge gateway receives the data stream from the multi-modal sensing unit, it performs unified format conversion on the data through the built-in protocol conversion module. Because different devices use different protocols and data formats, the protocol conversion module converts all the data transmitted back by the sensing units into a unified format to facilitate subsequent processing and analysis. Then, the intelligent edge gateway performs device state anomaly detection on the unified format data, and through the pre-set anomaly detection AI model, the intelligent edge gateway can analyze the working state of the device in real time to determine whether the device has abnormal conditions, such as device over-temperature, overload or other fault symptoms. If an anomaly is detected, the intelligent edge gateway will generate a corresponding diagnosis result, identify which devices have failed, and provide a preliminary judgment of the fault type and severity, otherwise an empty state diagnosis result will be generated. Through the above process, P sets of device state diagnosis results can be obtained, which can not only help the operation and maintenance personnel to understand the health status of the equipment in real time, but also provide feedback data for the system to optimize subsequent device control strategies.

[0017] Further, after the P intelligent edge gateways receive the P groups of multi-modal data streams transmitted back by the P groups of multi-modal sensing units in real time through heterogeneous industrial communication protocols, they perform device state anomaly detection and output P groups of device state diagnosis results, the method comprising: The first intelligent edge gateway performs device type filtering through heterogeneous industrial communication protocols to directionally receive the first group of multi-modal data streams transmitted back by the first group of multi-modal sensing units in real time; performs type-based data cleaning on the first group of multi-modal data streams to obtain K device multi-modal data streams of K first power transformation equipment entities; activates K first anomaly detection AI models and performs anomaly batch processing of the K device multi-modal data streams in parallel to output K device state diagnosis results, which constitute the first group of device state diagnosis results.

[0018] Optionally, the first intelligent edge gateway will preferentially filter the data from various types of sensing units in the station according to its internal preset device type filtering rules to identify and screen the data at the protocol level and type level. Since there are various types of power transformation equipment in the station, the sensing units corresponding to each device may report data through different heterogeneous industrial communication protocols, so the first edge gateway first identifies the protocol of all receivable data. After identification, the gateway only receives the real-time data stream returned by the multi-modal sensing unit belonging to the first group of power transformation equipment entities according to the device type filtering rules, thereby forming a first group of multi-modal data streams corresponding one-to-one to the first group of devices. After completing the protocol analysis and type filtering of the data, the first intelligent edge gateway performs type data cleaning processing on the first group of multi-modal data streams. In this process, the protocol conversion module will normalize the original data according to the signal characteristics and data format corresponding to different device types, including outlier rejection, sampling frequency unification, missing data filling, noise filtering, and cross-modal time alignment operations. Outlier rejection can be performed through Z-score outlier detection or IQR quartile range method; sampling frequency unification can be performed through resampling or linear interpolation to reconstruct the time series; missing data filling can be performed through moving average method or linear interpolation; noise filtering can be performed through wavelet denoising or Kalman filter; cross-modal time alignment can be performed through timestamp synchronization mechanism or dynamic time warping. Through the above cleaning process, the original data is optimized into K device multi-modal data streams that can be directly used for analysis, and each multi-modal data stream corresponds to one device in the first group of power transformation equipment entities. After completing data cleaning and generating standardized data streams, the K first anomaly detection AI models preloaded in the first intelligent edge gateway are activated in turn, and each anomaly detection AI model is trained and generated for the corresponding power transformation device, which can perform parallel analysis on data of different modalities. The edge gateway simultaneously inputs the K device multi-modal data streams into the corresponding K first anomaly detection AI models according to the batch processing mechanism to perform anomaly identification tasks, thereby outputting K device state diagnosis results. Finally, the K device state diagnosis results are aggregated in the intelligent edge gateway to form a first group of device state diagnosis results for the first group of power transformation equipment, which can be used for subsequent anomaly filtering, event generation, and cloud control strategy matching, thereby realizing intelligent and real-time monitoring of the running state of the power transformation equipment.

[0019] Further, the method further comprises: The first multi-modal sensing configuration is taken as a heterogeneous data type constraint to perform spatio-temporal correlation mining of the first type of power transformation equipment, to extract multi-modal time series fault data and time series fault event labels; the multi-modal time series fault data and time series fault event labels are taken as training data, a plurality of algorithm analysis models are locally called, model training is performed, and a plurality of verification set accuracies of a plurality of local diagnosis models are obtained; according to a preset accuracy threshold, the plurality of verification set accuracies are traversed, and N local diagnosis models are screened out from the plurality of local diagnosis models; the N local diagnosis models are connected in parallel, a confidence analysis layer is constructed based on N verification set accuracies at an output end, and construction of the first anomaly detection AI model is completed.

[0020] Optionally, when constructing the corresponding first anomaly detection AI model for the first type of power transformation equipment, first, according to the first multi-modal sensing configuration of the equipment type, it is taken as the heterogeneous data type constraint condition for subsequent data analysis. Based on this constraint, the intelligent edge gateway will perform spatio-temporal correlation mining from the collected equipment operation data. This mining process identifies modal parameters under the constraint of heterogeneous data types by analyzing time series data from multiple sensing dimensions, such as temperature sequences, current sequences, vibration spectrum sequences, etc., combined with the running environment of the equipment and the occurrence time of the event, thereby extracting multi-modal time series fault data that can be used for training. At the same time, by aligning and labeling the historical fault records, equipment operation logs and alarm events, the corresponding time series fault event labels are generated to realize event-level supervised learning. After obtaining the multi-modal time series fault data and the corresponding time series fault event labels, the intelligent edge gateway takes these data as the training data set and starts the local model training process. Since different algorithm models have different advantages on different equipment characteristics, multiple types of algorithm analysis models will be locally called, such as convolutional neural network (CNN), long short-term memory network (LSTM), time series Transformer, random forest, Bayesian classifier or other lightweight time series models. For each candidate algorithm model, the same data set is used for training, and the multiple local diagnosis models generated are verified one by one through the validation set to obtain the corresponding validation set accuracy index. Subsequently, according to the pre-set accuracy threshold, such as 90%, 95% or the accuracy requirement configured according to the actual scene, the validation accuracy of all models is traversed and filtered. Only the diagnosis models whose validation set accuracy reaches the threshold or above are selected to form N local diagnosis models with the best performance and the highest stability. After completing the model selection, the N local diagnosis models are integrated in a parallel structure, each model independently processes the input multi-modal time series data and generates an abnormality judgment result. At the output end, in order to improve the robustness and stability of the overall model, a confidence analysis layer is constructed based on the validation set accuracy of the N models. The confidence analysis layer will assign corresponding confidence weights to each model based on its performance in the validation phase, and generate the final comprehensive abnormality judgment result through weighted voting, probability fusion, etc. Through this parallel structure and confidence fusion mechanism, the system finally completes the construction of the first anomaly detection AI model for the first type of power transformation equipment, improving the reliability, accuracy and anti-noise ability of abnormality recognition.

[0021] The P intelligent edge gateways perform abnormality diagnosis filtering on the P sets of device state diagnosis results to obtain P key abnormality diagnosis events.

[0022] In one embodiment, after completing the device state diagnosis of each group of power transformation equipment, the P intelligent edge gateways perform abnormal diagnosis filtering operation on the P groups of device state diagnosis results output respectively. Specifically, when the edge gateway receives the state diagnosis result of the equipment in the group, first, all empty sets in the diagnosis result are removed, and then the abnormal scores are extracted from the remaining device state diagnosis results. Subsequently, the abnormal scores are compared with the preset safe operation threshold and alarm threshold, and combined with the abnormal change trend to determine whether it belongs to the negligible state, that is, whether it remains stable or decreases within the preset time window. Then, from the remaining device state diagnosis results, the abnormal scores within the safe operation threshold and the alarm threshold and the abnormal that cannot be ignored are selected and added to the P key abnormal diagnosis events to represent the core abnormality of the corresponding equipment group at the current time that needs to be handled most. Through the abnormal diagnosis filtering mechanism, the truly important abnormal events can be extracted from the massive real-time diagnosis results, the key equipment state can be efficiently identified and reported, thereby significantly improving the fault response speed, reducing the false positive rate, and ensuring the stable and reliable operation of the new energy smart station.

[0023] After the cloud control center receives the P key abnormal diagnosis events sent by the P intelligent edge gateways to perform dynamic device control strategy matching and delivery, a log file is generated for operation event recording, wherein the P intelligent edge gateways are connected to the cloud control center through a standard Ethernet.

[0024] In one embodiment, the cloud control center can continuously receive P key abnormal diagnosis events reported by P intelligent edge gateways through the standard Ethernet communication link established with each intelligent edge gateway. When a key abnormal diagnosis event is received, the device type, fault category, fault level, belonging substation entity and its influence range, etc. contained in the event are first analyzed and identified, and the gateway reference priority is set based on this. Then, the gateway reference priority is fused with the event urgency sequence determined by the P key abnormal diagnosis events to determine the processing order of the key abnormal diagnosis events, and the dynamic device control strategy is issued based on this, such as load limiting operation, switching standby device, segmentation isolation, starting emergency linkage or issuing safety alarm, etc. When issuing, the cloud control center issues the corresponding dynamic device control strategy to the target intelligent edge gateway that generates the abnormal event through standard Ethernet, and the edge side executes specific control actions. At the same time, the cloud control center automatically generates a corresponding log file to record the whole process of this event processing, and the log content includes the detailed information of the key abnormal diagnosis event (event number, occurrence time, device identification, abnormal type and level), the matched and issued control strategy content, the issue timestamp, the execution feedback result (success / failure and failure reason), and the necessary device state comparison data before and after. Through the above mechanism, the cloud control center not only realizes the dynamic and accurate control of key abnormalities, but also forms a complete traceable operation record, which provides data support for subsequent operation analysis, fault tracking, model optimization and compliance audit.

[0025] Further, the cloud control center receives the P key abnormal diagnosis events sent by the P intelligent edge gateways to execute dynamic device control strategy matching and issuing, and the method comprises: According to the fault influence range of the P groups of substation entities, the edge gateway safety level is classified to obtain the gateway reference priority; after constructing the fault influence evaluation matrix according to the P intelligent edge gateways, the P key abnormal diagnosis events are loaded into the fault influence evaluation matrix to quantify the fault urgency, and an event urgency sequence is output; the gateway reference priority and the event urgency sequence are fused to obtain a decision priority; according to the decision priority, the P key abnormal diagnosis events are executed to match and issue the dynamic device control strategy.

[0026] Optionally, after receiving the P key abnormal diagnosis events reported by the P intelligent edge gateways, the cloud control center first classifies the security level of the corresponding edge gateway according to the importance of the P groups of power transformation equipment entities in the operation of the station and the impact range caused by the fault, and determines a gateway reference priority for each edge gateway by normalizing and weighting the preset importance score of the power transformation equipment entity and the fault impact range in the key abnormal diagnosis event, so as to represent the weight of the gateway corresponding device type in the safety of the station operation. Subsequently, the cloud control center constructs a fault impact evaluation matrix according to the configuration of the P intelligent edge gateways and the relevance between devices, which is used to depict the possible chain reaction of different device faults on the overall operation of the station. When the P key abnormal diagnosis events are loaded into the matrix, the fault urgency quantification analysis is performed according to the abnormal level of the event, the fault duration, the potential impact of the event on other devices, the topological density of the fault location, etc., so as to output an event emergency degree sequence, which reflects the immediate processing necessity of each abnormal event at the current time. After generating the event emergency degree sequence, the event emergency degree sequence and the gateway reference priority are fused and calculated, and the fusion algorithm usually adopts a normalization weighting mechanism to ensure that the importance of the device and the current fault urgency are considered at the same time. Through weight integration, the cloud control center finally obtains a comprehensive decision priority for decision-making and scheduling. Then, based on the decision priority, the cloud control center executes device control strategy matching for the P key abnormal diagnosis events in the order of priority processing one by one, and issues the corresponding strategy to each target intelligent edge gateway. The events with higher priority will be processed first in the system, and the corresponding device control strategy will be issued faster, so as to ensure that the most important and urgent abnormalities can be responded in time, avoid fault diffusion or cause system-level chain risk, and improve the operation safety and processing efficiency of the overall station.

[0027] Further, the method further comprises: a preset response time window; if the first intelligent edge gateway does not receive the control strategy issued by the cloud control center within the response time window, a local security strategy mapping table is called to make a local control decision for the first key abnormal diagnosis event.

[0028] Optionally, to ensure continuous security response capability in the case of network fluctuations, communication anomalies or cloud unavailability, a response time window is preset in the first intelligent edge gateway during deployment. The response time window is set according to the actual operation requirements of the new energy station, for example, it can be 100 ms, 500 ms or longer, and is used to limit the cloud control center to respond to critical abnormal events and issue control strategies within the time limit. When the first intelligent edge gateway reports the first critical abnormal diagnosis event, the gateway immediately starts the local response time delay timer and continues to listen for policy issuance instructions from the cloud control center within the time window. If the cloud successfully issues a matching control strategy within the response time window, the gateway executes the corresponding device control operation according to the received strategy. However, in some cases, due to excessive cloud computing pressure, communication line interference or unavailability of the cloud control center, the edge gateway may not receive the policy instructions within the specified response time window. To avoid the device continuing to be in a risky state without human intervention, the first intelligent edge gateway will automatically trigger the local emergency control mechanism when it detects that the response time window has expired and still has not received the policy instructions. At this time, the gateway will call the pre-stored local security policy mapping table, which contains local emergency control strategies corresponding to different critical abnormal types, such as emergency disconnection, reduced capacity operation, enable protection mode, switch to backup device, lock operating parameters, etc. The intelligent edge gateway independently executes local control decisions according to the type, severity of the first critical abnormal event and its corresponding strategy in the mapping table. Through this localized control mechanism, even in the case of cloud unavailability or communication interruption, the intelligent edge gateway can ensure that the device is in a safe state, enabling autonomous protection and independent response, thereby avoiding further deterioration of the fault or triggering a chain of risks.

[0029] Further, the method further comprises: When the log file accumulates to a preset time window or reaches a preset abnormal event quantity threshold, the P abnormal detection AI models in the P intelligent edge gateways are iteratively trained based on the log file, obtaining P updated detection AI models; P incremental tuning packages of the P updated detection AI models and the P abnormal detection AI models are compared and output; the cloud control center issues the P incremental tuning packages to the P intelligent edge gateways, and executes model parameter updating through lightweight container hot updating technology.

[0030] Preferably, the cloud control center will continuously accumulate log files, when the log files reach a preset time window in the time dimension, for example, 24 hours, 72 hours, etc., or reach a preset abnormal event threshold in the event quantity dimension, for example, 100 key events are accumulated, the cloud control center will automatically trigger the iteration training process of the abnormal detection AI model. First, the cloud control center extracts the historical multi-modal data, abnormal diagnosis results and other contents corresponding to each edge gateway from the log files, and classifies and organizes these data according to the device types to which the gateways belong, forming P training sample sets. Each training sample corresponds to the abnormal detection AI model currently deployed on the gateway, which is used for model iteration training. The cloud control center then carries out cloud high-performance training on the P abnormal detection AI models. During the training process, the inference results of the model in the log data are compared with the true results of the events, and the model structure, parameter weight, etc. are optimized and adjusted through gradient descent, Adam, etc. After training, P updated detection AI models with better performance than the old models are obtained. After obtaining the updated models, the cloud control center compares each updated model with the corresponding old model, and generates P incremental parameter tuning packages through parameter difference. These incremental parameter tuning packages only contain the parameter increments required for model updating, such as weight difference, bias difference, threshold update amount, etc. Compared with reissuing complete models, they have the advantages of smaller data volume, faster update speed, and high transmission reliability. Then, the P incremental parameter tuning packages are respectively issued to the corresponding intelligent edge gateways. After receiving the parameter tuning packages, each edge gateway updates the locally running abnormal detection AI model through the built-in lightweight container hot update technology, such as Docker hot update, micro-container weight replacement technology or model inference engine non-stop parameter update mechanism. Since the hot update method is used, the edge gateway does not need to stop the inference task or restart the service during the model update process, thereby ensuring the continuity and real-time performance of the station monitoring capability. Through the above process, the P abnormal detection AI models are continuously iteratively optimized, so that the model can continuously adapt to changes in the station operating environment, improve the accuracy and robustness of abnormal detection, and realize the intelligent model closed-loop update mechanism under the cloud-edge collaboration.

[0031] In summary, the embodiments of the present application have at least the following technical effects: First, P intelligent edge gateways are configured for P groups of substation equipment entities in the new energy smart power station. These P intelligent edge gateways are communicatively connected to P groups of multimodal sensing units, which are mapped and deployed on the P groups of substation equipment entities. Next, the P intelligent edge gateways receive the P groups of multimodal data streams transmitted in real time from the P groups of multimodal sensing units via a heterogeneous industrial communication protocol. They then unify the data format through a protocol conversion module, perform equipment status anomaly detection, and output the P groups of equipment status diagnostic results. Then, the P intelligent edge gateways perform anomaly diagnostic filtering on the P groups of equipment status diagnostic results to obtain P key anomaly diagnostic events. Finally, the cloud control center receives the P key anomaly diagnostic events sent by the P intelligent edge gateways, executes dynamic equipment control strategy matching and distribution, and generates log files to record operation events. The P intelligent edge gateways are connected to the cloud control center via standard Ethernet. It solves the technical problems of complex integration and difficult real-time processing caused by the diversification of equipment protocols and insufficient edge computing power in new energy power plants. It achieves the technical effects of reducing system complexity and improving real-time response capabilities through protocol conversion and local intelligent processing, and ensuring system reliability and security through cloud-edge collaboration.

[0032] Example 2, based on the same inventive concept as the adaptive intelligent edge computing method in the foregoing examples, such as... Figure 2 As shown, this application provides an adaptive intelligent edge computing system, the system comprising: Gateway Configuration Component 11: Configures P intelligent edge gateways for P groups of substation equipment entities in the new energy smart power station, wherein the P intelligent edge gateways are communicatively connected to P groups of multimodal sensing units, and the P groups of multimodal sensing units are mapped and deployed on the P groups of substation equipment entities; Anomaly Detection Component 12: After receiving the P groups of multimodal data streams transmitted in real time by the P groups of multimodal sensing units through a heterogeneous industrial communication protocol, the P intelligent edge gateways unify the data format through a protocol conversion module, perform equipment status anomaly detection, and output the P groups of equipment status diagnosis results; Anomaly Filtering Component 13: The P intelligent edge gateways perform anomaly diagnosis filtering on the P groups of equipment status diagnosis results to obtain P key anomaly diagnosis events; Event Recording Component 14: After receiving the P key anomaly diagnosis events sent by the P intelligent edge gateways, the cloud control center performs dynamic equipment control strategy matching and distribution, generates a log file to record operation events, wherein the P intelligent edge gateways are connected to the cloud control center through a standard Ethernet.

[0033] Furthermore, the gateway configuration component 11 is used to perform the following methods: The new energy smart station is aggregated with power transformation equipment to obtain P groups of power transformation equipment entities corresponding to P types of power transformation equipment; fault monitoring sensor demand matching is performed according to the fault state attributes of the P types of power transformation equipment to obtain P multi-modal sensor configurations; sensor installation is performed on the P groups of power transformation equipment entities according to the P multi-modal sensor configurations to obtain the P groups of multi-modal sensor units.

[0034] Further, the anomaly detection component 12 is configured to perform the following method: The first intelligent edge gateway filters device types through heterogeneous industrial communication protocols to directionally receive a first group of multi-modal data streams that are real-time returned by a first group of multi-modal sensor units; type data cleaning is performed on the first group of multi-modal data streams to obtain K device multi-modal data streams of K first power transformation equipment entities; K first anomaly detection AI models are activated, and anomaly batch processing of the K device multi-modal data streams is performed in parallel to output K device state diagnosis results, which constitute a first group of device state diagnosis results.

[0035] Further, the anomaly detection component 12 is configured to perform the following method: The first multi-modal sensor configuration is used as a constraint of heterogeneous data types to perform spatio-temporal correlation mining of the first type of power transformation equipment, to extract multi-modal time sequence fault data and time sequence fault event labels; the multi-modal time sequence fault data and the time sequence fault event labels are used as training data, a plurality of algorithm analysis models are locally called, model training is performed, a plurality of verification set accuracies of a plurality of local diagnosis models are obtained; according to a preset accuracy threshold, the plurality of verification set accuracies are traversed, N local diagnosis models are screened out from the plurality of local diagnosis models; the N local diagnosis models are connected in parallel, and a confidence analysis layer is constructed based on the N verification set accuracies at an output end to complete construction of the first anomaly detection AI model.

[0036] Further, the event recording component 14 is configured to perform the following method: When the log file accumulates to a preset time window or reaches a preset abnormal event quantity threshold, iterative training of P anomaly detection AI models in the P intelligent edge gateways is performed based on the log file to obtain P updated detection AI models; P incremental parameter tuning packages of the P updated detection AI models and the P anomaly detection AI models are compared and output; the cloud control center issues the P incremental parameter tuning packages to the P intelligent edge gateways, and model parameter updating is performed through a lightweight container hot updating technology.

[0037] Further, the event recording component 14 is configured to perform the following method: According to the fault influence range of the P group of power transformation equipment entities, edge gateway security level classification is performed to obtain a gateway reference priority; after a fault influence evaluation matrix is constructed according to the P intelligent edge gateways, the P key abnormal diagnosis events are loaded into the fault influence evaluation matrix for fault urgency quantification, and an event emergency degree sequence is output; the gateway reference priority and the event emergency degree sequence are fused to obtain a decision priority; and according to the decision priority, the P key abnormal diagnosis events are executed to perform dynamic device control strategy matching and issuing.

[0038] Further, the event recording component 14 is configured to execute the following method: A preset response time window is set; if the first intelligent edge gateway does not receive the control strategy issued by the cloud control center within the response time window, a security strategy mapping table is locally called to make a local control decision for the first key abnormal diagnosis event.

[0039] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, without departing from the technical solution of the present application. Any modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application are still within the scope of the technical solution of the present application.

Claims

1. An adaptive intelligent edge computing method, characterized in that, The method includes: In a smart new energy power station, P intelligent edge gateways are configured for P groups of substation equipment entities. The P intelligent edge gateways are communicatively connected to P groups of multimodal sensing units, and the P groups of multimodal sensing units are mapped and deployed on the P groups of substation equipment entities. After receiving the P groups of multimodal data streams transmitted in real time from the P groups of multimodal sensing units through the heterogeneous industrial communication protocol, the P intelligent edge gateways unify the data format through the protocol conversion module, perform equipment status anomaly detection, and output the P groups of equipment status diagnosis results. The P intelligent edge gateways perform abnormal diagnosis filtering on the P groups of device status diagnosis results to obtain P key abnormal diagnosis events. After receiving the P key anomaly diagnostic events sent by the P intelligent edge gateways, the cloud control center executes dynamic device control policy matching and distribution, and generates log files to record operation events. The P intelligent edge gateways are connected to the cloud control center via standard Ethernet.

2. The self-adaptive intelligent edge computing method of claim 1, wherein, The method further includes: When the log files accumulate to a preset time window or reach a preset threshold for the number of abnormal events, iterative training of P anomaly detection AI models in the P smart edge gateways is performed based on the log files to obtain P update detection AI models. Compare and output the P incremental parameter tuning packages of the P update detection AI models and the P anomaly detection AI models; The cloud control center sends the P incremental parameter tuning packages to the P intelligent edge gateways, and performs model parameter updates through lightweight container hot update technology.

3. The self-adaptive intelligent edge computing method of claim 1, wherein, The method involves configuring P intelligent edge gateways for P groups of substation equipment in a smart new energy power station. The new energy smart power station is aggregated with substation equipment to obtain P groups of substation equipment entities corresponding to P types of substation equipment; Based on the fault status attributes of the P types of substation equipment, fault monitoring sensing requirements are matched to obtain P multimodal sensing configurations. Based on the P multimodal sensor configurations, sensors are installed on the P groups of substation equipment entities to obtain the P groups of multimodal sensor units.

4. The self-adaptive intelligent edge computing method of claim 3, wherein, After receiving the P groups of multimodal data streams transmitted in real time from the P groups of multimodal sensing units via a heterogeneous industrial communication protocol, the P intelligent edge gateways perform equipment status anomaly detection and output P groups of equipment status diagnostic results. The method includes: The first intelligent edge gateway filters device types through a heterogeneous industrial communication protocol to receive the first set of multimodal data streams transmitted back in real time by the first set of multimodal sensing units. Typed data cleaning is performed on the first group of multimodal data streams to obtain K device multimodal data streams for K first power equipment entities; Activate K first anomaly detection AI models, perform anomaly batch processing of the K device multimodal data streams in parallel, and output K device status diagnosis results to form the first set of device status diagnosis results.

5. The self-adaptive intelligent edge computing method of claim 4, wherein, The method further includes: Using the first multimodal sensor configuration as a heterogeneous data type constraint, spatiotemporal correlation mining is performed on the first type of substation equipment to extract multimodal time-series fault data and time-series fault event labels; Using the multimodal time-series fault data and time-series fault event labels as training data, multiple algorithm analysis models are locally retrieved, model training is performed, and multiple validation set accuracies of multiple local diagnostic models are obtained. Based on the preset accuracy threshold, the multiple validation sets are iterated through to select N local diagnostic models from the multiple local diagnostic models. The N local diagnostic models are connected in parallel, and a confidence analysis layer is constructed at the output end based on the accuracy of the N validation sets, thus completing the construction of the first anomaly detection AI model.

6. The self-adaptive intelligent edge computing method of claim 3, wherein, The cloud control center receives the P key anomaly diagnostic events sent by the P intelligent edge gateways and executes dynamic device control policy matching and distribution. The method includes: The security level of the edge gateway is classified according to the fault impact range of the P group of substation equipment entities to obtain the gateway baseline priority; After constructing a fault impact assessment matrix based on the P intelligent edge gateways, the P key anomaly diagnostic events are loaded into the fault impact assessment matrix to quantify the fault urgency and output an event urgency sequence. The decision priority is obtained by combining the gateway baseline priority and the event urgency sequence; Based on the decision priority, dynamic device control strategies are matched and issued for the P key abnormal diagnostic events.

7. The adaptive intelligent edge computing method as described in claim 4, characterized in that, The method further includes: Preset response delay window; If the first intelligent edge gateway does not receive the control policy issued by the cloud control center within the response delay window, it will locally call the security policy mapping table to make local control decisions for the first critical anomaly diagnosis event.

8. An adaptive intelligent edge computing system, characterized in that, The system is used to implement the adaptive intelligent edge computing method according to any one of claims 1-7, the system comprising: Gateway configuration component: Configure P intelligent edge gateways for P groups of substation equipment entities in the new energy smart power station, wherein the P intelligent edge gateways are communicatively connected to P groups of multimodal sensing units, and the P groups of multimodal sensing units are mapped and deployed on the P groups of substation equipment entities; Anomaly detection component: After receiving the P groups of multimodal data streams transmitted in real time from the P groups of multimodal sensing units through the heterogeneous industrial communication protocol, the P intelligent edge gateways unify the data format through the protocol conversion module, perform equipment status anomaly detection, and output the P groups of equipment status diagnosis results. Anomaly filtering component: The P smart edge gateways perform anomaly diagnosis filtering on the P groups of device status diagnosis results to obtain P key anomaly diagnosis events; Event logging component: After receiving the P key anomaly diagnostic events sent by the P intelligent edge gateways, the cloud control center executes dynamic device control policy matching and distribution, and generates log files to record operation events. The P intelligent edge gateways are connected to the cloud control center via standard Ethernet.

Images