EMS-based local ai risk assessment system and method

Abstract

This invention relates to a local AI risk assessment system and method based on EMS. The method involves accessing hardware devices to perform lightweight processing and cleaning on raw data to obtain cleaned data; extracting features and classifying the cleaned data to obtain classified data; performing multi-dimensional analysis and quantitative scoring on the classified data to obtain quantitative scoring results; and executing emergency response measures based on the quantitative scoring results and a preset mechanism. Its advantages lie in improving the accuracy of risk assessment, achieving risk quantification, and automatically triggering graded processing.

Description

Technical Field

[0001] This invention relates to the field of intelligent assembly technology applications, and in particular to a local AI risk assessment system and method based on EMS. Background Technology

[0002] With the rapid development of distributed energy storage systems, a large number of outdoor energy storage power stations have been widely deployed in various parts of the power grid, including large power stations in the backbone network and small power stations in the end network. However, these outdoor energy storage power stations face multiple risks and challenges: on the one hand, due to the large number of batteries they carry, there are potential safety hazards such as heat leakage, fire, and explosion; on the other hand, expensive batteries and cables are easy targets for theft. In addition, most power stations are located in remote areas and may also be subject to animal invasions, natural disasters, etc., all of which can cause serious losses to the power stations.

[0003] Currently, there is no effective solution to the problem that existing energy storage power stations cannot integrate multi-source heterogeneous data to achieve accurate risk assessment and automated closed-loop management. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a local AI risk assessment system and method based on EMS, in order to solve the technical problem that existing energy storage power stations are unable to integrate multi-source heterogeneous data to achieve accurate risk assessment and automated closed-loop management.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This invention provides a local AI risk assessment system based on EMS, comprising: a data acquisition module for accessing various hardware devices, collecting multi-source data including numbers, text, images, audio, and video, and performing lightweight data processing and cleaning on the multi-source data to obtain cleaned data; the lightweight data processing and cleaning includes cleaning and filtering of abnormal data, and encryption and compression of normal data; an AI model processing module for performing standardized preprocessing and feature extraction on the cleaned data to obtain fusion processing and deep analysis of heterogeneous data, and transmitting the processed data to EMS; a big data processing module for performing multi-dimensional analysis on the data output by the AI ​​model processing module, establishing a structured label system and performing quantitative scoring to obtain quantitative scoring results; and a processing module for receiving the quantitative scoring results, classifying them in conjunction with the power plant's operating status, and automatically executing emergency response measures according to a preset corresponding mechanism.

[0006] Optional hardware devices include one or more of the following: drones, cameras, weather stations, seismographs, and broadcasting equipment.

[0007] Optionally, the AI ​​model processing module is also used to extract core key information from the cleaned data using image recognition and text recognition technologies, including time points, personnel, number of people, animals, moving objects, vehicle colors, and license plate information, and to perform feature extraction and classification on keywords extracted from broadcast signals.

[0008] Optionally, the big data processing module is also used to quantify and score the data using a cumulative scoring method, and synchronize the quantification and scoring results and evaluation reports to the power plant control system.

[0009] Optionally, the big data processing module is also used to set different classification weights based on device type, time, region and data credibility, and to label the classification results with color.

[0010] Optionally, the processing module includes a rule engine that uses a cloud-local synchronization mode, with the cloud synchronizing to the local system and the local rule engine serving as the execution unit.

[0011] Furthermore, optionally, the rules in the rules engine are jointly generated by humans and AI models, and the rule content includes: labels, weights, activation conditions, and scope of application.

[0012] Optionally, the rules engine accumulates valid scores with lifecycles to form a quantitative score, and then performs tiered processing based on the score range.

[0013] Optionally, the processing module is also used to initiate a tiered processing mechanism based on the quantitative scoring results. The tiered processing mechanism includes: sending SMS or email notifications to relevant personnel, adjusting power plant operating parameters, initiating a safety control mechanism, or contacting a security company for intervention.

[0014] This invention provides a local AI risk assessment method based on EMS, applied to the aforementioned local AI risk assessment system based on EMS, comprising: accessing hardware devices to perform lightweight processing and cleaning on raw data to obtain cleaned data; extracting features and classifying the cleaned data to obtain classified data; performing multi-dimensional analysis and quantitative scoring on the classified data to obtain quantitative scoring results; and executing emergency response measures based on the quantitative scoring results and a preset mechanism.

[0015] This invention employs the above technical solution, which involves accessing hardware devices to perform lightweight processing and cleaning on the original data, resulting in cleaned data; extracting features and classifying the cleaned data, resulting in classified data; performing multi-dimensional analysis and quantitative scoring on the classified data, resulting in quantitative scoring results; and executing emergency response measures based on the quantitative scoring results and a preset mechanism. Compared with existing technologies, this invention has the following technical effects: improving the accuracy of risk assessment, achieving risk quantification, and automatically triggering graded processing. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a local AI risk assessment system based on EMS according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a local AI risk assessment method based on EMS according to an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0018] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0019] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0020] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units (elements) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or apparatus. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The terms “multiple” / “several” used in this application refer to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can indicate: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0021] An illustrative embodiment of the present invention, such as Figure 1 As shown, Figure 1 This is a schematic diagram of a local AI risk assessment system based on EMS according to an embodiment of the present invention. The local AI risk assessment system based on EMS provided in this application includes: The data acquisition module 10 is used to connect to various hardware devices, collect multi-source data including numbers, text, images, audio and video, and perform lightweight data processing and cleaning on the multi-source data to obtain cleaned data. Lightweight data processing and cleaning includes cleaning and filtering of abnormal data, and encryption and compression of normal data. The AI ​​model processing module 12 is used to perform standardized preprocessing and feature extraction on the cleaned data to obtain heterogeneous data fusion processing and deep analysis, and transmit the processed data to EMS. The big data processing module 14 is used to perform multi-dimensional analysis on the data output by the AI ​​model processing module 12, establish a structured label system and perform quantitative scoring to obtain quantitative scoring results. The processing module 16 is used to receive the quantitative scoring results, classify and process them in conjunction with the power plant's operating status, and automatically execute emergency response measures according to the preset corresponding mechanism.

[0022] Optional hardware devices include one or more of the following: drones, cameras, weather stations, seismographs, and broadcasting equipment.

[0023] Optionally, the AI ​​model processing module 12 is also used to extract core key information from the cleaned data using image recognition and text recognition technologies, including time points, personnel, number of people, animals, moving objects, vehicle colors, and license plate information, and to perform feature extraction and classification on the keywords extracted from the broadcast signal.

[0024] Optionally, the big data processing module 14 is also used to quantify and score the data using a cumulative scoring method, and synchronize the quantification and scoring results and evaluation report to the power plant control system.

[0025] Optionally, the big data processing module 14 is also used to set different classification weights based on device type, time, region and data credibility, and to label the classification results with color.

[0026] Optionally, the processing module 16 has a built-in rule engine. The rule engine adopts a cloud-local synchronization mode, with the cloud unidirectionally synchronizing to the local machine, and the local rule engine serving as the execution unit.

[0027] Furthermore, optionally, the rules in the rules engine are jointly generated by humans and AI models, and the rule content includes: labels, weights, activation conditions, and scope of application.

[0028] Optionally, the rules engine accumulates valid scores with lifecycles to form a quantitative score, and then performs tiered processing based on the score range.

[0029] Optionally, the processing module 16 is also used to initiate a graded processing mechanism based on the quantitative scoring results. The graded processing mechanism includes: sending SMS or email notifications to relevant personnel, adjusting power plant operating parameters, initiating a safety control mechanism, or contacting a security company for intervention.

[0030] In summary, the EMS-based local AI risk assessment system provided in this application embodiment is as follows: The data acquisition module 10 can connect to various hardware devices and collect various types of data uploaded by them, including numbers, text, images, audio and video. These various types of data are first preprocessed and feature extracted by a dedicated AI model (i.e., the AI ​​model processing module 12 in this embodiment), and then converted into standardized data before being transmitted to the energy management system (EMS).

[0031] The general-purpose data collector installed on the device performs initial lightweight data processing and cleaning on the raw data, including cleaning and filtering abnormal data, and encrypting and compressing normal data.

[0032] The processed data is pushed to the EMS side, where the AI ​​module performs batch processing, feature extraction, classification, and labeling of the data.

[0033] The specific image recognition and character recognition technologies used are as follows: For example, when a drone takes a picture, AI automatically extracts the core information from the picture, such as the time, people, number of people, animals, moving objects, etc., and then classifies and labels them. If there are repeated objects, such as a car appearing repeatedly, the AI ​​extracts the time of its appearance, vehicle color, license plate information, and automatically counts and classifies them.

[0034] For example, in broadcast signals, frequently mentioned keywords such as earthquake prediction, magnitude, location, and distance from the local area are all extracted and classified as features.

[0035] Within the EMS, the big data processing module 14 conducts multi-dimensional analysis of this massive amount of data, establishes a structured tagging system, and performs quantitative scoring (using a cumulative scoring method). The quantitative scoring results and evaluation report are presented intuitively to the user and simultaneously synchronized to the power plant control system. The system will then automatically execute appropriate emergency response measures based on the preset corresponding mechanisms.

[0036] After processing and classifying the raw collected data, different weights will be set for each category. For example, data from the device itself will be given priority, followed by digitized data, and images will be assigned to the last category of extracted data. In the embodiments of this application, the classification is also labeled with colors.

[0037] In the embodiments of this application, the weighting rules can be divided according to device, time, region, and data reliability.

[0038] The data after weighting is imported into the rule engine for scoring.

[0039] In the rule engine design of this application embodiment, a cloud and local synchronization mode is adopted, with one-way synchronization from the cloud to the local machine; In this context, the local rule engine unit serves only as the execution unit.

[0040] The rule engine in the cloud is generated by both human and AI models. The generated rules are identical in purpose, except that the labels may differ.

[0041] A single rule can be configured with different application conditions, such as proportion, activation conditions, scope of application, etc.

[0042] At the same time, scores themselves have a lifespan; once they expire, they are removed from the points system.

[0043] After summarizing the above, the rules engine accumulates the valid scores in real time, and finally obtains the quantified score.

[0044] The preset corresponding mechanism in this application embodiment will classify the score into interval ratings. Once the score quickly reaches a certain score within a certain time period, or the score of a certain color reaches a certain value, we will carry out corresponding grading processing. For example, a slight red score of 100 points represents all fire alarm-related data, such as cabinet temperature, infrared imaging taken by drones, etc. If the threshold is exceeded, the system will send a text message or email to the relevant contacts.

[0045] For example: A power station has integrated cameras with a 24-hour drone monitoring system to form an intelligent monitoring linkage mechanism: when a camera captures an abnormal situation, but is unable to accurately determine the nature of the abnormality due to factors such as shooting angle and image clarity, the EMS system will first perform AI preprocessing on the information collected by the camera and conduct risk scoring.

[0046] If the assessment results indicate a medium-level risk, EMS will immediately initiate the corresponding handling procedures: on the one hand, put the entire power station into standby operation to ensure safe and controllable operation; on the other hand, automatically dispatch drones to the station for targeted patrol and reconnaissance, focusing on collecting high-definition video of abnormal points.

[0047] After the information collected by the drone is fed back to the system, EMS will combine it with the original camera information to conduct secondary AI analysis and risk reassessment. Finally, based on the latest assessment results, precise control measures will be taken, including adjusting the power plant's operating parameters, activating the safety control mechanism, or contacting the security company to intervene in a timely manner, forming a closed-loop management process of "discovery, assessment, handling, review, and re-handling".

[0048] Building upon the camera-drone linkage mechanism, this application provides a local AI risk assessment system based on EMS, which further introduces a dynamic trigger-based fusion processing mechanism to address the asynchronicity of multi-source heterogeneous data in the temporal dimension, the inconsistency in the spatial dimension, and the lag in risk assessment. This dynamic trigger-based fusion processing mechanism forms a complete closed-loop processing flow through multi-level triggering conditions, dynamic task scheduling, spatiotemporal alignment fusion, and secondary verification feedback, significantly improving the system's ability to identify complex risk scenarios and its response speed.

[0049] In this application embodiment, the preset and hierarchical nature of multi-level dynamic triggering conditions can be: To accurately capture risk events, multi-level dynamic trigger conditions are preset in the rule engine. These trigger conditions are not simple threshold judgments, but rather complex logical expressions built based on multiple dimensions such as device type, data change rate, time window, spatial region, and data credibility.

[0050] Specifically, the EMS-based local AI risk assessment system provided in this application first defines basic trigger parameters for each type of access hardware device. For image data collected by cameras, trigger conditions include target recognition confidence threshold (e.g., personnel recognition confidence > 80%), target continuous appearance time (e.g., the same person staying for more than 5 minutes), and target behavioral characteristics (e.g., abnormal actions such as climbing or entering restricted areas). For meteorological station data, trigger conditions include temperature change rate (e.g., temperature rise exceeding 3°C per minute), wind speed change threshold, and seismic wave intensity level. For broadcast signals, trigger conditions include keyword matching degree (e.g., frequency of occurrence of keywords such as "earthquake prediction" and "fire alarm") and information source credibility level.

[0051] The triggering conditions are divided into three levels according to risk level: Level 1 is the early warning triggering condition, which is used to identify potential risk signs, such as an abnormal rise in temperature but not reaching the alarm threshold; Level 2 is the confirmation triggering condition, which is used for medium-level risks that require further verification, such as a camera capturing a suspicious person but the image is blurry; Level 3 is the emergency triggering condition, which is used for clearly defined emergency situations, such as a smoke detector alarm or a seismograph detecting destructive vibrations.

[0052] Each trigger condition is accompanied by a time window parameter, which limits the validity of the condition within a specific time period. For example, the threshold for triggering intrusion during nighttime hours (22:00-6:00) is automatically lowered to enhance sensitivity to theft risks; during daytime hours, the threshold is appropriately increased to avoid false alarms caused by normal maintenance personnel activities. Additionally, the trigger conditions include a credibility weighting factor, assigning higher weight to data sources with high historical accuracy and setting lower initial credibility for newly connected devices or devices with low historical accuracy.

[0053] In this application embodiment, the dynamic task scheduling mechanism based on triggering conditions can be: When a certain triggering condition is met, the local AI risk assessment system based on EMS provided in this application does not immediately initiate a full emergency response, but instead enters a dynamic task scheduling phase. EMS automatically generates secondary data collection tasks based on the type and level of the triggering condition and the currently available mobile data collection equipment resources, enabling precise verification of abnormal areas.

[0054] The core of the dynamic task scheduling mechanism in this application embodiment lies in the task priority algorithm and resource matching strategy. This application embodiment provides a local AI risk assessment system based on EMS that maintains a real-time updated mobile data acquisition device resource pool, including: the drone's current location, battery status, current task status, and the type of sensor it carries (such as a visible light camera, infrared thermal imager, multispectral sensor, etc.). Upon receiving a trigger signal, the scheduling engine first assesses the urgency of the trigger event and the type of data to be collected: if it is a fire-related temperature anomaly trigger, then drones equipped with infrared thermal imagers are prioritized for scheduling; if it is a personnel intrusion trigger, then drones equipped with high-definition zoom cameras are scheduled.

[0055] The scheduling algorithm comprehensively considers the shortest path for the drone to reach the target area, the impact of current wind speed on flight, and the priority ranking among multiple concurrent tasks. For multiple simultaneous triggering events, the system adopts a dynamic priority queue based on risk scores, with events with higher scores being allocated data collection resources first. If all drones are currently executing tasks and cannot respond immediately, the system will trigger a waiting queue mechanism and simultaneously activate backup data collection schemes, such as retrieving nearby fixed cameras for pre-tracking.

[0056] The task scheduling command is issued using a two-way confirmation mechanism to ensure that the UAV successfully receives the task and begins execution. During the execution of the task, the UAV transmits location information and data collection progress back in real time. The EMS can dynamically adjust the flight path or data collection parameters according to the real-time situation. If the target is detected to be moving, it will automatically track and shoot.

[0057] In the embodiments of this application, the spatiotemporal alignment and consistency verification of multi-source data can be as follows: After mobile data collection devices such as drones return supplementary data, the core technical challenge faced by the EMS-based local AI risk assessment system provided in this application embodiment lies in how to effectively fuse the data collected in the first instance (such as images from a fixed camera) with the data collected in the second instance (such as drone video). These two types of data differ significantly in terms of collection time, spatial coordinates, imaging angle, and resolution, and direct superposition or simple comparison cannot obtain an accurate risk assessment.

[0058] To address this, this application provides a local AI risk assessment system based on EMS that introduces a spatiotemporal alignment and consistency verification mechanism. Firstly, there is alignment in the time dimension: each piece of collected data is accompanied by a timestamp accurate to the millisecond level upon generation. This application's local AI risk assessment system based on EMS uses the occurrence time of the first triggered event as a reference point, aligning all subsequently collected data to a unified timeline using a time interpolation algorithm. For video stream data, the system employs frame synchronization technology to extract keyframes and match them with the time point of the first image.

[0059] Secondly, there is the alignment of spatial dimensions: image data collected by fixed cameras and drones use different spatial coordinate systems. The EMS-based local AI risk assessment system provided in this application first obtains parameters such as the spatial position, orientation, and field of view of the fixed camera through calibration technology, establishing a mapping relationship from image pixel coordinates to actual geographic coordinates. The drone, on the other hand, obtains precise position and attitude information in real time through its onboard GPS / IMU system. The EMS-based local AI risk assessment system provided in this application converts both types of data into the WGS-84 geographic coordinate system, achieving precise spatial positioning.

[0060] After completing spatiotemporal alignment, the local AI risk assessment system based on EMS provided in this application embodiment enters the consistency verification stage. The AI ​​model processing module 12 extracts features from the targets (such as people and vehicles) identified in the first image, including: shape features, color features, motion trajectory, etc.; at the same time, it performs the same feature extraction on the high-definition image captured by the drone. The similarity between the two identified targets is calculated by a feature matching algorithm. If the similarity exceeds a preset threshold (such as 85%), it is determined to be the same target, and target identity is associated.

[0061] For situations involving multiple targets, this application provides a local AI risk assessment system based on EMS that employs a multi-target tracking algorithm to establish the spatiotemporal trajectory of the targets. It verifies whether the target's movement during the two data acquisitions conforms to physical laws, eliminating false targets caused by image noise or recognition errors. After successful verification, the system fuses the feature vectors extracted from the first and second data acquisitions to generate a fused feature vector containing multi-dimensional information.

[0062] In the embodiments of this application, the secondary risk assessment and dynamic feedback closed loop can be: After the feature vector is generated, the local AI risk assessment system based on EMS provided in this application embodiment inputs it into the big data processing module for secondary quantitative scoring. The secondary scoring is not a simple repetition of the first scoring process, but rather employs a weighted fusion algorithm, comprehensively considering factors such as the timeliness, reliability, and information richness of the data from both datasets. For example, high-resolution images collected by drones are given higher weight in the scoring due to their higher resolution and more flexible viewing angle; the first data collection, being earlier, is given higher weight when judging the risk evolution trend.

[0063] The secondary scoring result is dynamically compared with the primary scoring result. The EMS-based local AI risk assessment system provided in this application embodiment presets a scoring change rate threshold. If the increase in the secondary score compared to the primary score exceeds this threshold (e.g., a score increase exceeding 30%), it indicates that the risk does exist and its severity has increased. The EMS-based local AI risk assessment system provided in this application embodiment immediately triggers the emergency response module to perform tiered handling. If the secondary score is basically the same as or slightly lower than the primary score, a decision is made based on the specific circumstances to whether to lower the response level or maintain the observation status.

[0064] More importantly, the results of the secondary evaluation are fed back to the rule engine to optimize future triggering conditions. For example, if a certain type of event is confirmed as a false alarm after multiple triggers and secondary evaluations, the EMS-based local AI risk assessment system provided in this application automatically adjusts the triggering threshold of such events or reduces their initial weight, forming a complete closed loop of "collection, triggering, scheduling, fusion, evaluation, and optimization." This dynamic feedback mechanism enables the system to continuously adapt to environmental changes and new risks, and the accuracy and response efficiency of risk assessment continue to improve over time.

[0065] To ensure the timeliness and dynamic adaptability of risk scoring and avoid interference from outdated historical data in current risk assessments, the big data processing module 14 introduces a dynamic scoring and time decay mechanism with a lifecycle. This mechanism assigns a time attribute to each piece of risk data, establishes a dynamic decay model, and performs real-time integral recalculation to ensure that the scoring system always reflects the latest risk situation.

[0066] In this application embodiment, the construction and lifecycle definition of the timestamp-based scoring unit can be as follows: This application provides a local AI risk assessment system based on EMS that constructs an independent scoring unit for each piece of risk data entering the big data processing module 14. Each scoring unit includes the following core attributes: unique data identifier, data source device type, original data content summary, feature extraction result vector, classification label set, data collection timestamp, data validity period, initial score value, current score value, decay coefficient, and last update time.

[0067] The validity period of data is not set using a uniform fixed value, but is dynamically determined based on the characteristics of the data type. For data with extremely high real-time requirements, such as temperature readings from battery thermal leakage sensors, the validity period is set to 30 minutes; temperature data that has not been updated for more than 30 minutes is considered invalid. For image data captured by cameras, the validity period is set according to the dynamic characteristics of the targets in the image. Images of static targets (such as buildings) can have a validity period of up to 24 hours, while images of dynamic targets (such as people and vehicles) typically have a validity period of 2-4 hours. For environmental data collected by meteorological stations, the validity period is set according to the rate of change of meteorological elements; temperature data has a validity period of 1 hour, while seismic wave data has a validity period of only a few minutes.

[0068] Furthermore, the validity period of data is also related to the risk level. Data from high-risk events (such as fire alarms) remains more valuable than new data from low-risk events, even if it has been stored for a longer period. Therefore, the EMS-based local AI risk assessment system provided in this application sets a longer validity period and a slower decay rate for high-risk data, and a shorter validity period and a faster decay rate for low-risk data.

[0069] In this application embodiment, the mathematical modeling and parameter configuration of the multidimensional time decay function can be as follows: The core of the time decay mechanism is the time decay function, which defines the mathematical law by which the value of a scoring unit decays over time. This application provides a local AI risk assessment system based on EMS that incorporates multiple decay models to adapt to the characteristics of different types of risk data, allowing users to select and configure models according to their specific application scenarios.

[0070] The linear decay model is the most basic form of decay, and its mathematical expression is: S(t) = S0 × max(0, 1 - α × (t - t0) / T), where S(t) is the score value at the current time t, S0 is the initial score value, t0 is the data acquisition time, T is the validity period, and α is the decay rate coefficient. This linear decay model is suitable for risk indicators with relatively stable changing trends, such as continuous monitoring data of ambient temperature and humidity.

[0071] The exponential decay model adopts the form S(t) = S0 × e^(-β × (t - t0)), where β is the decay constant. This exponential decay model is suitable for risk assessment of sudden events, such as aftershock risk after an earthquake or reignition risk after a fire. In the initial stage, the risk value decreases rapidly and then gradually levels off.

[0072] The stepped decay model divides the time axis into multiple intervals, maintaining a constant score within each interval, and decreasing the score abruptly when the interval switches. This stepped decay model is suitable for risks with clear stage characteristics, such as risk events classified according to warning levels, where the score immediately returns to zero after the warning is lifted.

[0073] Hybrid decay models combine multiple decay characteristics. For example, they employ rapid exponential decay in the first hour after data collection, then switch to linear decay, and finally return to zero after a specific period. This composite decay strategy can more accurately simulate the complex patterns of risk evolution in the real world.

[0074] This application provides a local AI risk assessment system based on EMS that supports configuring different attenuation parameters for different types of data and can apply different attenuation rules to the same data source within different time windows. For example, for intrusion detection data collected at night, the attenuation rate is set to half that of daytime data because the intrusion risk is more persistent at night.

[0075] In this application embodiment, the dynamic integral rating and rate of change monitoring mechanism can be: The rules engine maintains a dynamic points pool in real time, which contains only all scoring units that are still valid. The local AI risk assessment system based on EMS provided in this application uses an incremental calculation method. Each time new data is added, old data expires, or existing data scores are updated, the total valid points and the point distribution under each category label are recalculated.

[0076] Dynamic score rating focuses not only on the absolute value of the score, but also on the trend and rate of change of the score. This application provides a preset score change rate monitor in a local AI risk assessment system based on EMS, which monitors the changes in the score for each risk category within a sliding time window in real time. The length of the sliding time window can be dynamically adjusted according to the risk type. For sudden risks (such as fire alarms), the window length is set to 1-5 minutes; for cumulative risks (such as equipment aging), the window length is set to several hours or even several days.

[0077] The core algorithm for rate of change monitoring is based on the concept of differentiation: calculating the change in the integral value at the current moment relative to the previous moment, and the acceleration of this change over time. When the increase in the integral value per unit time exceeds a preset threshold, even if the absolute integral value has not yet reached the alarm line, the local AI risk assessment system based on EMS provided in this application embodiment will issue an early warning. For example, if the temperature of a power plant's cabinet rises rapidly from its normal value within 2 minutes, although the current temperature has not yet exceeded the alarm threshold, the integral rate of change monitor detects an abnormal temperature rise rate and triggers a preventative response in advance, such as activating the auxiliary cooling system or notifying maintenance personnel to check.

[0078] Conversely, for risks that accumulate slowly, such as long-term slight overload of equipment, the EMS-based local AI risk assessment system provided in this application identifies potential chronic risks by monitoring the cumulative changes of integrals over a longer time window and generates maintenance recommendations.

[0079] In this application embodiment, the automatic cleaning of invalid data and archiving of historical data can be: Scoring units that have exceeded their validity period are automatically removed from the dynamic points pool and no longer participate in real-time risk assessment. This process is executed periodically by a background daemon, with the cleanup frequency set according to the system size and data processing volume, typically once per minute or once every ten minutes.

[0080] The processing of failure data is not simply deletion. This application provides a local AI risk assessment system based on EMS that transfers failure data, along with its complete lifecycle record (including initial score, decay process, final failure time, etc.), to a historical data archive. The archived data retains complete time-series information and can be used for subsequent data analysis, model training, audit traceability, and other purposes.

[0081] While archiving data, this application provides a local AI risk assessment system based on EMS to generate a data lifecycle report, intuitively displaying the complete process of various risk data from generation to expiration. Operations personnel can query the archive library to trace the evolution of historical risk events, analyze risk occurrence patterns, and optimize current attenuation parameter settings.

[0082] Furthermore, the EMS-based local AI risk assessment system provided in this application embodiment supports a mechanism for reactivating expired data. In special circumstances, such as the reassessment of historical events or the emergence of related new events, the EMS-based local AI risk assessment system provided in this application embodiment can, according to manual instructions or preset rules, re-import some archived data into the dynamic integration pool and assign it new validity periods and decay parameters. This flexible data management approach ensures that the EMS-based local AI risk assessment system provided in this application embodiment can meet complex practical application needs.

[0083] To further optimize the applicability and accuracy of the rules and enable the rule engine to adapt to the ever-changing risk environment and new threats, this application provides a local AI risk assessment system based on EMS that introduces a two-way feedback and self-learning mechanism. This two-way feedback and self-learning mechanism realizes the evolution of the rule engine from "one-way passive execution" to "two-way active optimization" by constructing an execution effect feedback channel, establishing a cloud-based self-learning optimization model, and implementing abnormal pattern recognition and automatic rule correction.

[0084] In this embodiment of the application, the construction and data collection of the execution effect feedback channel can be as follows: After executing emergency measures, processing module 16 does not end the entire process but initiates an execution effect feedback procedure. This procedure is responsible for collecting data related to the entire process of the emergency response, including: the type and score of the event that triggered the emergency, a list of specific measures executed, the timing and duration of the measures, the actual effects of the measures (e.g., whether the risk was successfully prevented from evolving, whether unnecessary downtime losses were caused), the results of manual review (if any), and subsequent actual events (e.g., whether a fire or theft actually occurred).

[0085] The feedback data collection employs a multi-source verification method. Regarding the effectiveness of the measures, this application's embodiment provides a local AI risk assessment system based on EMS that automatically compares the status data before and after implementation, such as whether the temperature has decreased, whether the intruder has left, and whether the equipment operating parameters have returned to normal. For situations involving human intervention, this application's embodiment provides a local AI risk assessment system based on EMS that records the content and time of the human operation and obtains the human's evaluation of the response (such as whether it was a correct response or a false alarm) through subsequent questionnaires or confirmation mechanisms.

[0086] All feedback data is encapsulated into feedback events in a standardized format and uploaded to the cloud-based rules engine via an encrypted channel. The upload process employs breakpoint resumption and message queue mechanisms to ensure that feedback data is not lost even in cases of network instability. Upon receiving the feedback data, the cloud performs integrity verification and unpacking, then stores it in the feedback history database.

[0087] The feedback database employs a time-series storage structure, with each feedback record forming a complete causal chain with the original triggering event, implemented measures, and execution results. Simultaneously, the database provides multi-dimensional indexing of the feedback data, supporting rapid retrieval and analysis based on criteria such as device type, risk classification, time range, and execution effectiveness.

[0088] In the embodiments of this application, the establishment and training of the cloud-based self-learning optimization model can be carried out as follows: The core of the cloud-based rule engine is a self-learning optimization model. This model continuously optimizes various rule parameters based on historical feedback data and using machine learning algorithms. The inputs to the self-learning optimization model include: the original rule's configuration parameters (weights, thresholds, triggering conditions), real-time data at the time of actual triggering, execution action records, and execution effect scores. The output of the self-learning optimization model is a suggested optimized rule parameter.

[0089] The self-learning optimization model employs a reinforcement learning framework, treating the rule engine as an intelligent agent, each emergency response as a decision-making action, and the execution effect as a reward signal. The goal of the self-learning optimization model is to learn an optimal set of rule parameter configurations that, in long-term operation, maximizes the accuracy of emergency responses, minimizes false alarm rates, and reduces resource consumption.

[0090] Specifically, the self-learning optimization model first performs feature engineering on historical feedback data to extract key features, such as the matching degree between triggering conditions and actual results, the correlation between weight settings and risk severity, and the correlation between threshold levels and false alarm rates. Then, it uses algorithms such as decision trees, random forests, or gradient boosting trees to establish a mapping relationship model between rule parameters and execution effects.

[0091] For complex scenarios, this application provides a local AI risk assessment system based on EMS that incorporates a deep neural network model, particularly a long short-term memory network, to capture the dynamic patterns of rule parameters changing over time and the temporal dependencies of risk events. For example, some risks may behave differently in a specific season or time period, and the deep neural network can automatically learn these periodic patterns and adjust the rule parameters accordingly.

[0092] The model is trained using a combination of offline batch training and online incremental learning. Every morning, the EMS-based local AI risk assessment system provided in this application retrains or fine-tunes the model using newly added feedback data, generating updated model parameters. Simultaneously, for scenarios with high real-time requirements, the model supports online incremental updates, making small adjustments to the model immediately after receiving new feedback data.

[0093] In the embodiments of this application, the dynamic adjustment and parameter optimization of rule weights can be as follows: Based on the output of the self-learning optimization model, the cloud-based rule engine dynamically adjusts the locally synchronized rules. The adjustments include: the weight coefficients of each category label, the thresholds for triggering conditions, the data validity period, the parameters of the decay function, and the classification of emergency response levels.

[0094] The weight coefficients are adjusted using a Bayesian optimization method. This application provides a local AI risk assessment system based on EMS that maintains historical accuracy statistics for each rule. When a rule is confirmed as accurate after multiple triggers (i.e., actual risk has occurred), its weight is appropriately increased; conversely, if a rule is frequently triggered but no actual risk has occurred (false alarm), its weight is gradually decreased. The magnitude of the weight adjustment is related to the historical sample size; the larger the sample size, the more reliable the adjustment.

[0095] The threshold optimization of triggering conditions is based on cost-sensitive learning. The local AI risk assessment system based on EMS provided in this application assigns different cost coefficients to different types of false alarms and false alarms. For example, the cost of false alarms of real fires is much higher than the cost of false alarms of fires. When adjusting the threshold, the optimization algorithm comprehensively considers accuracy, recall and cost factors to find the optimal threshold that minimizes the overall expected cost.

[0096] The optimization of data validity period and decay parameters is based on statistical analysis of risk duration. The local AI risk assessment system based on EMS provided in this application analyzes the average duration of various risks from occurrence to disappearance in historical data, as well as the value of data at different time points for risk prediction, and automatically adjusts the validity period and decay rate to make the scoring system more consistent with the actual evolution of risk.

[0097] All optimized parameters do not take effect immediately. Instead, they are first verified in a simulation testing environment. The cloud-based rule engine maintains a simulation testing platform isolated from the actual operating environment, using historical data playback technology to test the performance of the new parameters in historical scenarios. Only new configurations that pass simulation testing and confirm that their performance is better than or at least no worse than the original parameters will be pushed to the local rule engine.

[0098] In the embodiments of this application, the abnormal pattern recognition and automatic correction mechanism can be: Another important component of the self-learning mechanism is abnormal pattern recognition. The embodiment of this application provides a local AI risk assessment system based on EMS that continuously monitors the running status and feedback data of the rule engine, automatically identifies abnormal patterns, and triggers corresponding corrective actions.

[0099] Abnormal modes include, but are not limited to: repeated triggering of a certain type of risk event without actually causing any loss (continuous false alarm mode); a certain type of risk event actually occurring but never being triggered (missed alarm mode); the triggering frequency of rules fluctuating drastically in a short period of time (unstable mode); and significant discrepancies in feedback data from multiple local devices (inconsistent mode).

[0100] To address different types of anomaly patterns, this application provides a local AI risk assessment system based on EMS, employing corresponding identification algorithms. For false positive patterns, the system uses a clustering algorithm to group historical false positive events, analyzing the common characteristics of each group, such as specific devices, specific time periods, and specific environmental conditions, thereby pinpointing the root cause of the false positives. For missed positive patterns, the system uses an association rule mining algorithm to analyze which associated indicators are abnormal but not captured by rules when missed positive events occur.

[0101] After identifying anomaly patterns, the system automatically generates correction suggestions. For simple and clear anomalies, such as persistent false alarms from a device, the system can directly and automatically correct the rules, such as temporarily reducing the device's weight, increasing its trigger threshold, or temporarily removing it from the trigger source of a specific rule. After the correction action is executed, the EMS-based local AI risk assessment system provided in this application continues to monitor subsequent feedback, evaluate the correction effect, and form a closed-loop control.

[0102] For complex or uncertain anomalies, this application provides a local AI risk assessment system based on EMS that generates detailed anomaly reports and prompts for manual review. The report includes a description of the anomaly, possible causal analysis, suggested corrective measures, and an assessment of expected results. Operations personnel can refer to the report to decide whether to adopt the system's suggestions or intervene manually. The results of manual intervention are also fed back into the self-learning model as samples for subsequent learning.

[0103] In this application embodiment, rule version management and evolution tracking can be as follows: To ensure the controllability and traceability of rule updates, the cloud-based rule engine has established a comprehensive rule version management system. Each adjustment to rule parameters generates a new rule version, with version numbers following semantic versioning conventions (major version number.minor version number.revision number). Each rule version records complete change information, including: change time, change content, reason for change, the range of feedback data upon which the change was based, and a comparison of the effects before and after the change.

[0104] The rule version supports rollback. When a new version of the rule exhibits unexpected adverse effects during actual operation, operations and maintenance personnel can revert to the previous stable version with a single click. The rollback operation is also recorded as a version change, ensuring a complete historical record.

[0105] After long-term operation, the EMS-based local AI risk assessment system provided in this application has accumulated multiple versions of rule parameters and corresponding operational performance data. This data constitutes a complete map of rule evolution, which can be used to analyze changing trends in the risk landscape, evaluate the overall effectiveness of rule optimization, and predict possible future adjustment directions. Through this continuous evolution and optimization, the rule engine constantly adapts to changes in the external environment, maintaining efficient and accurate risk assessment capabilities at all times.

[0106] To effectively reduce losses, establishing a local risk assessment system is crucial. Currently, with the rapid development of AI technology and the increasing prevalence of large-scale models, the local AI risk assessment system based on EMS provided in this application deploys the AI ​​system locally and combines it with existing big data analysis methods. This not only significantly improves the accuracy of risk assessment but also fully meets the stringent safety requirements of the power system. It identifies, analyzes, and assesses potential safety risks in existing technologies, proposes corresponding risk prevention and control measures, and ensures the safe operation of energy storage power stations.

[0107] This invention employs the above technical solution, which involves accessing hardware devices to perform lightweight processing and cleaning on the original data, resulting in cleaned data; extracting features and classifying the cleaned data, resulting in classified data; performing multi-dimensional analysis and quantitative scoring on the classified data, resulting in quantitative scoring results; and executing emergency response measures based on the quantitative scoring results and a preset mechanism. Compared with existing technologies, this invention has the following technical effects: improving the accuracy of risk assessment, achieving risk quantification, and automatically triggering graded processing.

[0108] An illustrative embodiment of the present invention, such as Figure 2 As shown, Figure 2 This is a flowchart illustrating a local AI risk assessment method based on EMS according to an embodiment of the present invention, applied to... Figure 1 The EMS-based local AI risk assessment system shown in this application embodiment provides an EMS-based local AI risk assessment method including: Step S200: Connect the hardware device to perform light processing and cleaning on the raw data to obtain the cleaned data; Step S202: Perform feature extraction and classification on the cleaned data to obtain classified data; Step S204: Perform multi-dimensional analysis and quantitative scoring on the classified data to obtain the quantitative scoring results; Step S206: Implement emergency response measures based on the quantitative scoring results and the preset mechanism.

[0109] Specifically, in conjunction with steps S200 to S206, the local AI risk assessment method based on EMS provided in this application embodiment can be applied to... Figure 1 The local AI risk assessment system based on EMS is shown.

[0110] This invention employs the above technical solution, which involves accessing hardware devices to perform lightweight processing and cleaning on the original data, resulting in cleaned data; extracting features and classifying the cleaned data, resulting in classified data; performing multi-dimensional analysis and quantitative scoring on the classified data, resulting in quantitative scoring results; and executing emergency response measures based on the quantitative scoring results and a preset mechanism. Compared with existing technologies, this invention has the following technical effects: improving the accuracy of risk assessment, achieving risk quantification, and automatically triggering graded processing.

[0111] The above description is merely a preferred embodiment of the present invention and does not limit the implementation and protection scope of the present invention. Those skilled in the art should realize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A local AI risk assessment system based on EMS, characterized in that, include: The data acquisition module is used to connect to various hardware devices, collect multi-source data including numbers, text, images, audio and video, and perform lightweight data processing and cleaning on the multi-source data to obtain cleaned data. The lightweight data processing and cleaning includes cleaning and filtering abnormal data, and encrypting and compressing normal data. The AI ​​model processing module is used to perform standardized preprocessing and feature extraction on the cleaned data to obtain the fusion processing and deep analysis of heterogeneous data, and transmit the processed data to EMS. The big data processing module is used to perform multi-dimensional analysis on the data output by the AI ​​model processing module, establish a structured label system and perform quantitative scoring to obtain quantitative scoring results. The processing module is used to receive the quantitative scoring results, classify and process them in conjunction with the power plant's operating status, and automatically execute emergency response measures according to the preset corresponding mechanism.

2. The local AI risk assessment system based on EMS according to claim 1, characterized in that, The hardware devices include one or more of the following: drones, cameras, weather stations, seismographs, and broadcasting equipment.

3. The local AI risk assessment system based on EMS according to claim 1, characterized in that, The AI ​​model processing module is also used to extract core key information from the cleaned data using image recognition and text recognition technologies, including time points, personnel, number of people, animals, moving objects, vehicle colors, and license plate information, and to perform feature extraction and classification on the keywords extracted from the broadcast signal.

4. The local AI risk assessment system based on EMS according to claim 1, characterized in that, The big data processing module is also used to quantify and score the data using a cumulative scoring method, and to synchronize the quantified scoring results and evaluation report to the power plant control system.

5. The local AI risk assessment system based on EMS according to claim 1, characterized in that, The big data processing module is also used to set different classification weights based on device type, time, region and data credibility, and to label the classification results with color.

6. The local AI risk assessment system based on EMS according to claim 1, characterized in that, The processing module has a built-in rule engine, which adopts a cloud-local synchronization mode, with the cloud unidirectionally synchronizing to the local machine, and the local rule engine serving as the execution unit.

7. The local AI risk assessment system based on EMS according to claim 6, characterized in that, The rules in the rule engine are jointly generated by humans and AI models, and the rule content includes: tags, weights, activation conditions, and scope of application.

8. The local AI risk assessment system based on EMS according to claim 6, characterized in that, The rule engine accumulates valid scores with lifecycles to form a quantitative score, and then performs tiered processing based on the score range.

9. The local AI risk assessment system based on EMS according to claim 1, characterized in that, The processing module is also used to initiate a graded processing mechanism based on the quantitative scoring results. The graded processing mechanism includes: sending SMS or email notifications to relevant personnel, adjusting power plant operating parameters, initiating a safety control mechanism, or contacting a security company for intervention.

10. A local AI risk assessment method based on EMS, characterized in that, The local AI risk assessment system based on EMS, as described in any one of claims 1 to 9, comprises: Connect to hardware devices to perform lightweight processing and cleaning on the raw data, and obtain the cleaned data; The cleaned data is then subjected to feature extraction and classification to obtain classified data. The categorized data is subjected to multi-dimensional analysis and quantitative scoring to obtain quantitative scoring results. Emergency response measures will be implemented based on the quantitative scoring results and the preset mechanism.

Images