AI agent-based user-side power quality dynamic optimization method and system

Abstract

The application discloses a user-side power quality dynamic optimization method and system based on an AI agent, and relates to the technical field of power quality optimization. The method comprises the following steps: collecting real-time power consumption data of a user-side power distribution system, including time series data of voltage and current, and synchronously acquiring peak-valley electricity price information and user production plan information of a power grid; based on the power consumption data, performing multi-dimensional deep analysis through an AI agent to identify power quality risk types; the AI agent combines the power quality risk types, the peak-valley electricity price information and the user production plan information to dynamically generate optimization suggestions with power quality standard reaching as the core. The technical problem that the user-side power quality is difficult to fuse multi-source data for collaborative analysis and dynamic adjustment in the prior art, resulting in large power quality fluctuation, is solved, active identification and dynamic optimization adjustment of power quality risks are realized, and the technical effects of improving power quality stability and controllability are achieved.

Description

Technical Field

[0001] This invention relates to the field of power quality optimization technology, specifically to a user-side power quality dynamic optimization method and system based on AI intelligent agents. Background Technology

[0002] With the continuous improvement of industrial automation, the application of high-power frequency converters, power electronic devices, and nonlinear loads in user-side power distribution systems is becoming increasingly widespread. While this has improved production efficiency, it has also exacerbated power quality issues such as voltage fluctuations, harmonic pollution, three-phase imbalance, and voltage sags. Existing user-side power quality management methods typically rely on fixed-parameter compensation devices or rule-based control strategies based on single monitoring indicators. They lack the ability to deeply analyze time-series data such as voltage and current, and are difficult to coordinate and optimize adjustments in conjunction with peak-valley electricity pricing information and user production plans. Therefore, under conditions of frequent changes in load structure or dynamic adjustments in production rhythm, traditional control methods often suffer from response lag and insufficient adjustment accuracy, making it difficult to achieve continuous and stable power quality compliance. Furthermore, existing technologies largely focus on post-event management, lacking the ability to proactively identify and predict power quality risks, leading to increased power quality fluctuations and even affecting equipment safety and production stability. Summary of the Invention

[0003] This application provides a method and system for dynamic optimization of user-side power quality based on AI intelligent agents, which solves the technical problem in the prior art that it is difficult to integrate multi-source data for collaborative analysis and dynamic adjustment of user-side power quality, resulting in large fluctuations in power quality.

[0004] A first aspect of this application provides a user-side power quality dynamic optimization method based on an AI agent, the method comprising: Real-time collection of electricity consumption data from the user-side power distribution system, including time-series data of voltage and current, and simultaneous acquisition of peak-valley electricity price information from the power grid and user production plan information; based on the electricity consumption data, multi-dimensional in-depth analysis is performed through an AI agent to identify power quality risk types; the AI ​​agent combines the power quality risk types, the peak-valley electricity price information, and the user production plan information to dynamically generate optimization suggestions centered on achieving power quality standards.

[0005] A second aspect of this application provides a user-side power quality dynamic optimization system based on an AI intelligent agent, the system comprising: Data Acquisition Module: Collects real-time electricity consumption data from the user-side power distribution system, including time-series data of voltage and current, and simultaneously acquires peak-valley electricity price information from the power grid and user production plan information; Deep Analysis Module: Based on the electricity consumption data, performs multi-dimensional deep analysis through an AI agent to identify power quality risk types; Suggestion Generation Module: The AI ​​agent combines the power quality risk types, peak-valley electricity price information, and user production plan information to dynamically generate optimization suggestions centered on achieving power quality standards.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, real-time electricity consumption data from the user-side power distribution system is collected, including time-series data of voltage and current, and peak-valley electricity price information from the power grid and user production plan information are acquired simultaneously. Then, based on the electricity consumption data, a multi-dimensional in-depth analysis is performed using an AI agent to identify power quality risk types. Finally, the AI ​​agent combines power quality risk types, peak-valley electricity price information, and user production plan information to dynamically generate optimization suggestions centered on achieving power quality standards. This solves the technical problem in existing technologies where it is difficult to integrate multi-source data for collaborative analysis and dynamic adjustment of user-side power quality, leading to large fluctuations in power quality. It achieves proactive identification and dynamic optimization of power quality risks, improving the stability and controllability of power quality. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A schematic diagram of the user-side power quality dynamic optimization method based on AI intelligent agents provided in the embodiments of this application; Figure 2 A schematic diagram of the user-side power quality dynamic optimization system based on AI intelligent agents provided in this application embodiment.

[0009] Figure labeling: Data acquisition module 11, in-depth analysis module 12, suggested generation module 13. Detailed Implementation

[0010] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0011] Example 1, as Figure 1 As shown, this application provides a user-side power quality dynamic optimization method based on AI intelligent agents, wherein the method includes: Real-time collection of electricity consumption data from the user-side power distribution system, including time-series data of voltage and current, and simultaneous acquisition of peak-valley electricity price information from the power grid and user production plan information.

[0012] Intelligent power quality monitoring terminals with a metering accuracy of no less than 0.5S are deployed at the incoming line cabinets, key branches, and high-power load nodes of the user-side power distribution system. Each monitoring terminal includes a voltage transformer, a current transformer, and a data acquisition module. It continuously samples the three-phase voltage and three-phase current at a preset sampling frequency (e.g., 1kHz~10kHz) to generate voltage and current time-series data. The data acquisition module performs analog-to-digital conversion, timestamp marking, and data caching on the sampled data, and uploads it to a local edge computing gateway or cloud data server via an industrial Ethernet or 5G communication network. To ensure data consistency, both the voltage and current time-series data are synchronized and calibrated based on a unified time source (such as the NTP network time protocol or a GPS timing module). Simultaneously, by establishing API communication connections with the power grid dispatching platform or power information service interface, peak-valley electricity price information within the current and future preset time windows is periodically acquired and stored in structured data format. Furthermore, by connecting to enterprise production management systems or enterprise resource planning systems, user production plan information is obtained, including production process arrangements, equipment start-up and shutdown plans, and load allocation plans. This production plan information is then parsed and its fields are standardized. Finally, the voltage time-series data, current time-series data, peak-valley electricity price information, and production plan information are uniformly encapsulated into a multi-source fusion dataset and stored in a database for subsequent analysis and decision-making by the AI ​​agent.

[0013] Based on the electricity consumption data, a multi-dimensional in-depth analysis is performed using an AI agent to identify types of power quality risks.

[0014] In this embodiment, voltage timing data and current timing data are input to the data processing module of the AI ​​agent. The AI ​​agent includes a data preprocessing unit, a feature extraction unit, a risk discrimination unit, and a result output unit. First, the data preprocessing unit performs denoising, outlier removal, and sliding time window segmentation on the collected voltage and current time-series data. The sliding time window length can be set to 1s~10s, and the data within each time window is normalized. Second, the feature extraction unit extracts multi-dimensional features in the time domain, frequency domain, and statistical domain, including features such as voltage RMS fluctuation rate, current peak factor, total harmonic distortion rate, negative sequence voltage ratio, zero sequence current ratio, and voltage imbalance coefficient. Then, the multi-dimensional features are input into a pre-trained risk discrimination model. The risk discrimination model can be constructed using deep neural networks, convolutional neural networks, or time-series prediction models based on attention mechanisms, and the model weight parameters are obtained by training on historical labeled samples. The risk discrimination model classifies the input features and outputs different types of risk probability values. When the probability value corresponding to a certain risk type is greater than a preset risk judgment threshold, it is determined that there is a corresponding power quality risk, and the power quality risk type and corresponding risk level are output.

[0015] Furthermore, the power quality risk types include at least one of harmonic exceedance, voltage fluctuation, and three-phase imbalance.

[0016] Among them, harmonic exceedance risk is used to characterize the state where the content of each harmonic or the total harmonic distortion rate in the current or voltage signal exceeds the national or industry standard limit; voltage fluctuation risk is used to characterize the abnormal amplitude change of the effective voltage value within a preset time window, including events such as voltage dips, voltage swells, or short-term interruptions; three-phase imbalance risk is used to characterize the state where there is an abnormal increase in negative or zero-sequence components due to amplitude or phase deviations between the three-phase voltages or currents. These risk types can occur individually or multiple risk types can coexist; when multiple risk types exist, the AI ​​agent classifies and labels them according to the severity index corresponding to each risk, and generates combined risk labels for subsequent priority ranking and decision-making in optimization strategies.

[0017] Furthermore, based on the aforementioned electricity consumption data, a multi-dimensional in-depth analysis is performed using an AI agent to identify types of power quality risks, including: The system invokes a harmonic analysis model within the AI ​​intelligent entity to assess the risk of exceeding limits in the current time-series data of the electricity consumption data, and outputs a first assessment result. It then invokes a voltage fluctuation analysis model within the AI ​​intelligent entity to assess the risk of voltage fluctuations based on the voltage time-series data of the electricity consumption data, and outputs a second assessment result. Finally, it invokes a three-phase imbalance analysis model within the AI ​​intelligent entity to assess the risk of three-phase imbalance in the current time-series data of the electricity consumption data, and outputs a third assessment result. Combining the first assessment result, the second assessment result, and the third assessment result, the system outputs the power quality risk type.

[0018] First, the time-synchronized and preprocessed current time-series data is input into the harmonic analysis model within the AI ​​intelligent body. This model performs spectral analysis on the current signal based on a frequency domain decomposition algorithm, extracting the total harmonic distortion rate and the content of each harmonic, and compares these with preset standard limits, outputting a first judgment result characterizing whether there is a risk of harmonic exceedance and its severity. Second, the voltage time-series data is input into the voltage fluctuation analysis model within the AI ​​intelligent body. This model dynamically tracks the effective voltage value based on a sliding time window and identifies abnormal fluctuation events such as voltage dips, swells, or short-term interruptions, outputting a characterization of the voltage fluctuation risk type and wave characteristics. The second judgment result of the amplitude level; next, the current time series data is input into the three-phase imbalance analysis model in the AI ​​intelligent body, and the amplitude deviation rate, negative sequence component ratio and zero sequence component ratio of the three-phase current are calculated in real time. The risk is judged according to the preset imbalance threshold, and the third judgment result representing the degree and level of three-phase imbalance is output; finally, the risk fusion unit of the AI ​​intelligent body performs rule fusion or weighted fusion on the first judgment result, the second judgment result and the third judgment result to generate a comprehensive risk vector. The final power quality risk type is determined according to the dominance of the risk indicators in the comprehensive risk vector, and the corresponding risk label and risk level are output.

[0019] Furthermore, the harmonic analysis model within the AI ​​intelligent body is invoked to assess the risk of exceeding limits in the current time-series data of the electricity consumption data, and a first assessment result is output, including: Perform a Fast Fourier Transform on the current time-series data to calculate the total harmonic distortion (THD) and the content of each harmonic; compare the THD and the content of each harmonic with the standard limits to determine whether there is any exceedance, and output the first judgment result.

[0020] First, the time-synchronized current timing data is segmented according to a preset time window (e.g., 20ms~200ms), and windowing is applied to the current sampling points within each time window to reduce spectral leakage errors. Then, a Fast Fourier Transform is performed on the current data within each time window to obtain the corresponding spectral amplitude sequence. The fundamental effective value is extracted based on the fundamental frequency component, and the effective values ​​of the 2nd to Nth harmonic components are also extracted. Based on the fundamental effective value and the effective values ​​of each harmonic, the total harmonic distortion rate is calculated, and its calculation formula is: The ratio of the square root of the sum of squares of the effective values ​​to the fundamental effective value is calculated, along with the harmonic content of each order. Then, the total harmonic distortion rate and the harmonic content of each order are compared with preset standard limits, which are set according to national or industry power quality standards. When the total harmonic distortion rate or any harmonic content exceeds the corresponding limit, a harmonic exceedance risk is determined, and a risk level coefficient is calculated based on the exceedance magnitude to generate a first judgment result that includes the exceedance type, the number of exceedances, and the risk level. If the limit is not exceeded, a first judgment result indicating normal harmonic risk is output.

[0021] Furthermore, the voltage fluctuation analysis model within the AI ​​intelligent body is invoked to assess voltage fluctuation risk based on voltage time-series data in the electricity consumption data, and a second assessment result is output, including: For the voltage timing data, a morphological filtering method is used to identify the start and end times and amplitude changes of voltage sags, swells, or short-term interruptions, generating event identification results; load current data at the same time is extracted, and if there is an internal load current with a sudden change greater than a preset sudden change threshold within the start and end times corresponding to the event identification results, it is determined to be a voltage fluctuation caused by an internal impact load; otherwise, it is determined to be a voltage fluctuation caused by an external power grid disturbance, and the second judgment result is output.

[0022] First, the voltage time-series data is segmented into continuous sliding windows according to a preset sampling frequency, and the effective voltage value sequence is calculated within each time window. Then, morphological filtering is used to detect anomalies in the effective voltage value sequence. This involves constructing structuring elements to perform opening and closing operations on the voltage curve to remove random noise and highlight abrupt change edges. The difference between the original signal and the morphological filtering result is used to identify abrupt changes in voltage amplitude. When the effective voltage value is detected to decrease or increase by more than a preset fluctuation ratio threshold within a preset time threshold, it is marked as a voltage dip or voltage swell event. When the effective voltage value drops below a preset ratio of the rated voltage and the duration exceeds the short-term interruption determination time, it is marked as a short-term interruption event, and the corresponding start time, end time, and amplitude change are recorded to generate event identification results.

[0023] Based on this, load current time-series data within the time interval corresponding to the event identification result is extracted, and the current change rate or current amplitude difference sequence is calculated. If the current change rate is detected to be greater than the preset mutation threshold within the start and end time of the voltage event, and the current change direction is correlated with the voltage change trend, it is determined to be voltage fluctuation caused by internal impact load. If no significant current mutation is detected or the current change is not correlated with the voltage change, it is determined to be voltage fluctuation caused by external power grid disturbance. Finally, a second judgment result including voltage fluctuation type, duration, fluctuation amplitude, and disturbance source category is output.

[0024] Furthermore, the three-phase imbalance analysis model within the AI ​​intelligent body is invoked to assess the three-phase imbalance risk of the current time-series data in the electricity consumption data, and a third assessment result is output, including: Three-phase voltage and current time-series data are extracted from the collected current and voltage time-series data, and negative sequence imbalance and zero sequence imbalance are calculated in real time. Based on the negative sequence imbalance and zero sequence imbalance, a three-phase imbalance risk assessment is performed, and the third assessment result is output.

[0025] First, the effective current and voltage sequences of phases A, B, and C are extracted from the collected current and voltage time-series data, and synchronized within a unified time window. Then, the three-phase voltage and current are decomposed using the symmetrical component method, calculating the positive-sequence, negative-sequence, and zero-sequence components. The negative-sequence unbalance is defined as the ratio of the effective value of the negative-sequence voltage (or negative-sequence current) to the effective value of the positive-sequence voltage (or positive-sequence current), and the zero-sequence unbalance is defined as the ratio of the effective value of the zero-sequence current to the average value of the three-phase current. After calculating the negative-sequence and zero-sequence unbalance in real time, they are compared with preset unbalance limits. When the negative-sequence or zero-sequence unbalance exceeds the corresponding limit, a three-phase unbalance risk is identified, and an unbalance risk level coefficient is calculated based on the excess ratio. Finally, a third judgment result is output, including the unbalance type (negative-sequence dominant or zero-sequence dominant), the unbalance value, and the risk level.

[0026] Furthermore, the identification of power quality risk types not only relies on voltage and current time series data, but also on the real-time collection of more power quality-related parameters, including active power, reactive power, apparent power, power factor, frequency, and infrared temperature measurement data, to conduct a comprehensive quality risk assessment.

[0027] Specifically, multi-functional power sensors and infrared temperature sensors are installed at key nodes of the user-side power distribution system to collect power parameters and equipment temperature data in real time. All collected power parameters (including active power, reactive power, apparent power, power factor, and frequency) are recorded and transmitted by high-precision energy meters or power analysis instruments. Infrared temperature sensors monitor the surface temperature of key equipment, with a sampling frequency of no less than 1Hz. All collected data are synchronized using a unified timestamp to ensure data consistency in the time domain. Time synchronization uses Network Time Protocol (NTP) or Global Positioning System (GPS). After noise reduction, outlier removal, and standardization, the power quality data is input to the multi-dimensional deep analysis module of the AI ​​agent for further analysis. The AI ​​agent conducts risk assessment through multiple sub-models, including: harmonic analysis model, voltage fluctuation analysis model, three-phase imbalance analysis model, power factor analysis model, and temperature monitoring model. These models analyze current time series data, voltage time series data, power factor, and temperature data respectively, and output their respective risk judgment results. When multiple risk types coexist, the AI ​​agent integrates the judgment results of various sub-models to generate a comprehensive risk assessment and outputs the corresponding power quality risk type and risk level, providing a basis for subsequent optimization and adjustment. Finally, based on the comprehensive assessment results, the AI ​​agent generates optimization and adjustment suggestions, such as adjusting the electricity usage time of non-critical loads, optimizing the power factor, and adjusting equipment operating temperature, to achieve power quality compliance and maximize electricity cost-effectiveness. The AI ​​agent dynamically generates optimization suggestions centered on achieving power quality compliance, combining the power quality risk types, peak-valley electricity price information, and user production plan information.

[0028] First, the power quality risk types are converted into structured risk vectors, which include risk category identifiers, risk severity coefficients, and risk duration parameters. Simultaneously, the peak-valley electricity price information is parsed into time-of-use price sequences and aligned with a preset future time window. The user production plan information is parsed into equipment start-up and shutdown plans, load power demand curves, and a set of process time constraints. Based on this, a multi-objective optimization model is constructed, where meeting power quality indicators is used as a hard constraint. These hard constraints include total harmonic distortion not exceeding a preset limit, voltage fluctuation amplitude not exceeding a preset proportion, and three-phase imbalance not exceeding a preset threshold. The optimization objective functions are minimizing electricity costs and maximizing production stability. Electricity costs are calculated by integrating the product of time-of-use prices and load power curves, while production stability is comprehensively evaluated through a combination of key equipment operation continuity indicators and process completion indicators. Subsequently, the AI ​​agent identifies the set of adjustable loads based on the risk vector, and generates load adjustment decision schemes by solving the multi-objective optimization model under the premise of meeting the production process time constraints. The decision schemes include equipment start-up and shutdown timing adjustments, operating power limit adjustments, or reactive power compensation equipment switching strategies. When real-time monitoring data changes, the AI ​​agent updates the risk vector and triggers rolling optimization calculations to generate new optimization suggestions. Finally, the output includes optimization suggestion results containing the adjustment object, adjustment range, adjustment time interval, and expected power quality improvement effect.

[0029] Furthermore, the AI ​​agent, combining the power quality risk type, peak-valley electricity price information, and user production plan information, dynamically generates optimization suggestions centered on achieving power quality standards, including: Based on the power quality risk type and the user's production plan information, correlation analysis is used to match the power quality risk type with the nonlinear loads on the production line to generate a set of associated loads. Based on the user's production plan information, the production criticality of each associated load in the set of associated loads is identified to determine a set of non-critical associated loads. Taking the set of non-critical associated loads as the adjustment object, and combining the peak-valley electricity price information with the multi-objective optimization model in the AI ​​intelligent body, the system uses the achievement of power quality indicators as the core constraint and the minimization of electricity costs and the maximization of production stability as the optimization objectives to make decisions on the operating parameters of the non-critical associated loads and generate the optimization suggestions.

[0030] First, a risk feature vector is constructed based on the power quality risk type. This vector includes a risk category identifier, an impact frequency range, an occurrence time interval, and a severity coefficient. Simultaneously, production equipment operation plans, equipment rated power, and equipment load type identifiers are extracted from the user's production plan information. By calculating the correlation coefficient between the risk feature vector and the load characteristics of each device, or by constructing a risk-load mapping model based on historical operating data, nonlinear loads that trigger or amplify corresponding risks are identified, generating a set of associated loads. Second, based on the user's production plan information, the production criticality of each associated load in the associated load set is identified. Specifically, the process sequence relationships and delivery node constraints in the production plan are analyzed, a directed graph model of the production process is constructed, and the time margin of each production unit is calculated using critical path analysis. If a production unit to which an associated load belongs is on the critical path, or its shutdown would lead to delivery delays or a complete production line shutdown, it is marked as a critical load. Other loads with room for delay or derated operation are marked as non-critical loads, forming the set of non-critical associated loads. Subsequently, taking the set of non-critical loads as the adjustment object, a multi-objective optimization model is constructed based on the peak-valley electricity price information. The model uses the achievement of power quality standards as a hard constraint, including harmonic content, voltage fluctuation amplitude, and three-phase imbalance not exceeding preset limits. The optimization objective functions are minimizing electricity costs and maximizing production stability. Electricity cost indicators are calculated by integrating time-of-use electricity prices with load power curves, and production stability indicators are calculated by using equipment operation continuity coefficients and production cycle deviation coefficients. A multi-objective evolutionary algorithm or weighted solution method is used to solve the model, obtaining decision schemes for the operating parameters of each non-critical load, including start-stop timing adjustments, power limit settings, or reactive power compensation configuration strategies, ultimately generating the optimization recommendations.

[0031] Furthermore, based on the user production plan information, the production criticality of each associated load in the associated load set is identified, and the set of non-critical associated loads is determined, including: The user production plan information is parsed to extract production process metadata, including process sequence, process dependencies, and product delivery nodes. Based on the production process metadata, a directed graph model with production units as nodes is constructed, where the edges of the directed graph represent the predecessor-successor relationships and time constraints between processes. For each load in the associated load set, the production unit to which it belongs and the key process links of the service are mapped. Based on the directed graph model, critical path analysis is performed to identify loads that, if interrupted or derated, will directly lead to product delivery delays or entire line shutdowns. These loads are marked as critical loads and excluded from the associated load set. The remaining loads constitute the non-critical associated load set.

[0032] First, the user's production plan information is structured and parsed to extract production process metadata such as production process number, process start time, process end time, process dependencies, equipment resource allocation, and product delivery node time. Then, a directed graph model with production units or processes as nodes is constructed based on the production process metadata. The nodes represent production units or process links, and the directed edges in the graph represent the preceding and following logical relationships between processes and the corresponding time constraints. The time constraints include the earliest start time, the latest finish time, and the allowable delay margin. After the directed graph model is constructed, the earliest start time, latest start time, and time fluctuation of each node are calculated using a critical path analysis algorithm. Critical path nodes are identified by comparing the time fluctuation. Then, each load in the associated load set is mapped and associated with its corresponding production unit or process step to determine whether the load is located on a critical path node or supports a critical process step. If the production unit corresponding to a load is located on the critical path, or its power derating operation will cause the time fluctuation of the unit to be less than a preset safety margin threshold, then the load is determined to be a critical load and is excluded from the associated load set. The remaining loads that are not located on the critical path and have adjustable time margin or power adjustment space are determined as a non-critical associated load set for subsequent optimization and adjustment.

[0033] Furthermore, among the optimization objectives, maximizing production stability takes precedence over minimizing electricity costs.

[0034] When constructing the multi-objective function, hierarchical optimization or weighted coefficient method is used to prioritize and constrain each optimization objective. First, production stability index is used as the primary optimization objective. Under the premise of meeting the hard constraint of power quality compliance, the production cycle deviation rate, continuous operation time of key equipment, and number of process interruptions are comprehensively evaluated to find the candidate solution set that makes the production stability index optimal or meets the preset stability threshold. Within the candidate solution set, the minimization of electricity cost is used as the secondary optimization objective. The electricity cost index is determined by calculating the integral value of the time-of-use electricity price and load power curve. The optimization scheme with the lowest electricity cost is selected from the candidate solution set as the final decision result.

[0035] In summary, the embodiments of this application have at least the following technical effects: First, real-time electricity consumption data from the user-side power distribution system is collected, including time-series data of voltage and current, and peak-valley electricity price information from the power grid and user production plan information are acquired simultaneously. Then, based on the electricity consumption data, a multi-dimensional in-depth analysis is performed using an AI agent to identify power quality risk types. Finally, the AI ​​agent combines power quality risk types, peak-valley electricity price information, and user production plan information to dynamically generate optimization suggestions centered on achieving power quality standards. This solves the technical problem in existing technologies where it is difficult to integrate multi-source data for collaborative analysis and dynamic adjustment of user-side power quality, leading to large fluctuations in power quality. It achieves proactive identification and dynamic optimization of power quality risks, improving the stability and controllability of power quality.

[0036] Example 2, based on the same inventive concept as the AI-based user-side power quality dynamic optimization method in the aforementioned examples, such as... Figure 2 As shown, this application provides a user-side power quality dynamic optimization system based on AI intelligent agents, wherein the system includes: Data acquisition module 11: Real-time acquisition of electricity consumption data from the user-side power distribution system, including time-series data of voltage and current, and simultaneous acquisition of peak-valley electricity price information and user production plan information from the power grid; Deep analysis module 12: Based on the electricity consumption data, multi-dimensional deep analysis is performed through an AI agent to identify the types of power quality risks; Suggestion generation module 13: The AI ​​agent combines the types of power quality risks, the peak-valley electricity price information, and the user production plan information to dynamically generate optimization suggestions with power quality compliance as the core.

[0037] Furthermore, the depth analysis module 12 is used to perform the following methods: The power quality risk types include at least one of harmonic exceedance, voltage fluctuation, and three-phase imbalance.

[0038] Furthermore, the depth analysis module 12 is used to perform the following methods: The system invokes a harmonic analysis model within the AI ​​intelligent entity to assess the risk of exceeding limits in the current time-series data of the electricity consumption data, and outputs a first assessment result. It then invokes a voltage fluctuation analysis model within the AI ​​intelligent entity to assess the risk of voltage fluctuations based on the voltage time-series data of the electricity consumption data, and outputs a second assessment result. Finally, it invokes a three-phase imbalance analysis model within the AI ​​intelligent entity to assess the risk of three-phase imbalance in the current time-series data of the electricity consumption data, and outputs a third assessment result. Combining the first assessment result, the second assessment result, and the third assessment result, the system outputs the power quality risk type.

[0039] Furthermore, the depth analysis module 12 is used to perform the following methods: Perform a Fast Fourier Transform on the current time-series data to calculate the total harmonic distortion (THD) and the content of each harmonic; compare the THD and the content of each harmonic with the standard limits to determine whether there is any exceedance, and output the first judgment result.

[0040] Furthermore, the depth analysis module 12 is used to perform the following methods: For the voltage timing data, a morphological filtering method is used to identify the start and end times and amplitude changes of voltage sags, swells, or short-term interruptions, generating event identification results; load current data at the same time is extracted, and if there is an internal load current with a sudden change greater than a preset sudden change threshold within the start and end times corresponding to the event identification results, it is determined to be a voltage fluctuation caused by an internal impact load; otherwise, it is determined to be a voltage fluctuation caused by an external power grid disturbance, and the second judgment result is output.

[0041] Furthermore, the depth analysis module 12 is used to perform the following methods: Three-phase voltage and current time-series data are extracted from the collected current and voltage time-series data, and negative sequence imbalance and zero sequence imbalance are calculated in real time. Based on the negative sequence imbalance and zero sequence imbalance, a three-phase imbalance risk assessment is performed, and the third assessment result is output.

[0042] Furthermore, the suggestion generation module 13 is used to perform the following method: Based on the power quality risk type and the user's production plan information, correlation analysis is used to match the power quality risk type with the nonlinear loads on the production line to generate a set of associated loads. Based on the user's production plan information, the production criticality of each associated load in the set of associated loads is identified to determine a set of non-critical associated loads. Taking the set of non-critical associated loads as the adjustment object, and combining the peak-valley electricity price information with the multi-objective optimization model in the AI ​​intelligent body, the system uses the achievement of power quality indicators as the core constraint and the minimization of electricity costs and the maximization of production stability as the optimization objectives to make decisions on the operating parameters of the non-critical associated loads and generate the optimization suggestions.

[0043] Furthermore, the suggestion generation module 13 is used to perform the following method: The user production plan information is parsed to extract production process metadata, including process sequence, process dependencies, and product delivery nodes. Based on the production process metadata, a directed graph model with production units as nodes is constructed, where the edges of the directed graph represent the predecessor-successor relationships and time constraints between processes. For each load in the associated load set, the production unit to which it belongs and the key process links of the service are mapped. Based on the directed graph model, critical path analysis is performed to identify loads that, if interrupted or derated, will directly lead to product delivery delays or entire line shutdowns. These loads are marked as critical loads and excluded from the associated load set. The remaining loads constitute the non-critical associated load set.

[0044] Furthermore, the suggestion generation module 13 is used to perform the following method: Among the optimization objectives, maximizing production stability takes precedence over minimizing electricity costs.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A user-side power quality dynamic optimization method based on AI intelligent agents, characterized in that, The method includes: Real-time collection of electricity consumption data from the user-side power distribution system, including time-series data of voltage and current, and simultaneous acquisition of peak-valley electricity price information from the power grid and user production plan information; Based on the electricity consumption data, a multi-dimensional in-depth analysis is performed using an AI agent to identify the types of power quality risks. The AI ​​agent combines the power quality risk type, peak-valley electricity price information, and user production plan information to dynamically generate optimization suggestions with power quality compliance as the core.

2. The user-side power quality dynamic optimization method based on AI intelligent agents as described in claim 1, characterized in that, The power quality risk types include at least one of harmonic exceedance, voltage fluctuation, and three-phase imbalance.

3. The user-side power quality dynamic optimization method based on AI intelligent agents as described in claim 2, characterized in that, Based on the aforementioned electricity consumption data, a multi-dimensional in-depth analysis is performed using an AI agent to identify types of power quality risks, including: The harmonic analysis model within the AI ​​intelligent body is invoked to assess the risk of exceeding the standard in the current time series data of the electricity consumption data, and the first assessment result is output. The voltage fluctuation analysis model within the AI ​​intelligent body is invoked to assess voltage fluctuation risk based on voltage time-series data in the electricity consumption data, and a second assessment result is output. The three-phase imbalance analysis model in the AI ​​intelligent body is invoked to perform a three-phase imbalance risk assessment on the current time series data in the power consumption data, and a third assessment result is output. The power quality risk type is output by combining the first judgment result, the second judgment result, and the third judgment result.

4. The user-side power quality dynamic optimization method based on AI intelligent agents as described in claim 3, characterized in that, The harmonic analysis model within the AI ​​intelligent body is invoked to assess the risk of exceeding limits in the current time-series data of the electricity consumption data, and a first assessment result is output, including: Perform a fast Fourier transform on the current time series data to calculate the total harmonic distortion rate and the content of each harmonic. The total harmonic distortion rate and the content of each harmonic are compared with the standard limit to determine whether there is any exceedance, and the first judgment result is output.

5. The user-side power quality dynamic optimization method based on AI intelligent agents as described in claim 3, characterized in that, The voltage fluctuation analysis model within the AI ​​intelligent body is invoked to assess voltage fluctuation risk based on voltage time-series data in the electricity consumption data, and a second assessment result is output, including: For the voltage timing data, morphological filtering is used to identify the start and end times and amplitude changes of voltage sags, swells or short-term interruptions, and event identification results are generated. Extract load current data at the same time. If there is an internal load current with a change degree greater than the preset change threshold during the start and end time corresponding to the event identification result, it is determined to be a voltage fluctuation caused by internal impact load. Otherwise, it is determined to be a voltage fluctuation caused by external power grid disturbance, and the second judgment result is output.

6. The user-side power quality dynamic optimization method based on AI intelligent agents as described in claim 3, characterized in that, The three-phase imbalance analysis model within the AI ​​intelligent body is invoked to assess the three-phase imbalance risk of the current time-series data in the electricity consumption data, and a third assessment result is output, including: Three-phase voltage and current time series data are extracted from the collected current and voltage time series data, and negative sequence unbalance and zero sequence unbalance are calculated in real time. Based on the negative sequence imbalance degree and the zero sequence imbalance degree, a three-phase imbalance risk assessment is performed, and the third assessment result is output.

7. The user-side power quality dynamic optimization method based on AI intelligent agents as described in claim 1, characterized in that, The AI ​​agent, combining the power quality risk type, peak-valley electricity price information, and user production plan information, dynamically generates optimization suggestions centered on achieving power quality standards, including: Based on the power quality risk type and the user production plan information, the power quality risk type is associated with the nonlinear load on the production line through correlation analysis to generate an associated load set. Based on the user production plan information, the production criticality of each associated load in the associated load set is identified, and the set of non-critical associated loads is determined. Using the set of non-critical loads as the adjustment object, and combining the peak-valley electricity price information with the multi-objective optimization model within the AI ​​intelligent body, the system takes the achievement of power quality indicators as the core constraint and the minimization of electricity costs and the maximization of production stability as the optimization objectives to make decisions on the operating parameters of the non-critical loads and generate the optimization suggestions.

8. The user-side power quality dynamic optimization method based on AI intelligent agents as described in claim 7, characterized in that, Based on the user production plan information, the production criticality of each associated load in the associated load set is identified, and the set of non-critical associated loads is determined to include: Parse the user's production plan information and extract production process metadata, including process sequence, process dependencies, and product delivery nodes; Based on the production process metadata, a directed graph model with production units as nodes is constructed, where the edges of the directed graph represent the preceding and succeeding relationships and time constraints between processes. For each load in the associated load set, map the production unit and key process links of the service to which it belongs. Based on the directed graph model, perform critical path analysis to identify loads that, if interrupted or derated, will directly lead to product delivery delays or production line shutdowns. Mark these loads as critical loads and exclude them from the associated load set. The remaining loads constitute the non-critical associated load set.

9. The user-side power quality dynamic optimization method based on AI intelligent agents as described in claim 7, characterized in that, Among the optimization objectives, maximizing production stability takes precedence over minimizing electricity costs.

10. A user-side power quality dynamic optimization system based on AI intelligent agents, characterized in that, The system is used to implement the user-side power quality dynamic optimization method based on any one of claims 1-9, the system comprising: Data acquisition module: Real-time acquisition of electricity consumption data from the user-side power distribution system, including time-series data of voltage and current, and synchronous acquisition of peak and valley electricity price information from the power grid and user production plan information; Deep analysis module: Based on the electricity consumption data, it performs multi-dimensional deep analysis through AI intelligent agents to identify the types of power quality risks; Suggestion generation module: The AI ​​agent combines the power quality risk type, peak-valley electricity price information and user production plan information to dynamically generate optimization suggestions with power quality compliance as the core.

Images