Zero-carbon factory intelligent operation and maintenance central control system and power quality optimization method and device

Abstract

The present application relates to the field of photovoltaic power generation technology, and discloses a zero-carbon factory intelligent operation and maintenance central control system and a power quality optimization method and device. The method is as follows: a central control platform is set up in a preset zero-carbon factory, including a photovoltaic monitoring module and a power quality analysis module; a factory energy digital twin model is constructed according to the characteristics of photovoltaic components, load equipment parameters and grid access point characteristics; real-time meteorological data and historical power generation data corresponding to the zero-carbon factory are obtained; power generation data, power quality parameters, real-time meteorological data and historical power generation data are input into the factory energy digital twin model to generate a future power generation curve; the best treatment plan is matched in a preset fault mode self-learning database according to the power generation data and future power generation curve of the zero-carbon factory; the three-phase imbalance of the zero-carbon factory power supply network is adjusted according to the best treatment plan to complete the power quality optimization of the zero-carbon factory. It effectively solves the power quality control problem of distributed photovoltaic system grid connection.

Description

Technical Field

[0001] The present application relates to the field of photovoltaic power generation technology, and in particular to a zero-carbon factory intelligent operation and maintenance central control system and a power quality optimization method and device. Background Art

[0002] Driven by the "dual carbon" goal, zero-carbon factories are widely adopting distributed photovoltaic power generation systems. However, the intermittent nature of photovoltaic output and nonlinear load operation lead to problems such as excessive total harmonic distortion (THD) in the power grid and increased three-phase imbalance. Traditional methods have the following drawbacks:

[0003] 1. Disconnect between prediction and control: Existing PV monitoring systems only collect power generation data and do not dynamically link weather forecasts with power quality parameters. This results in the configuration of harmonic control equipment (such as APF) lagging behind actual demand.

[0004] 2. Insufficient model accuracy: By using a single physical model to construct the factory energy system, without integrating the attenuation characteristics of photovoltaic modules and grid impedance parameters, a power prediction error of more than 15% can be generated when the load suddenly changes.

[0005] 3. Rigid governance solutions: Using a fixed threshold-triggered compensation strategy, the impact of PV output fluctuations on grid harmonic characteristics is not considered, which can easily lead to over-compensation or under-compensation.

[0006] 4. Insufficient data utilization: Existing methods do not use machine learning to match historical fault patterns with real-time power generation curves, resulting in a three-phase unbalanced regulation response time exceeding 200ms.

[0007] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0008] This application provides a zero-carbon factory intelligent operation and maintenance central control system and power quality optimization method and device, aiming to solve the problems of existing zero-carbon factories generally using distributed photovoltaic power generation systems, such as disconnection between prediction and management, insufficient model accuracy, rigid management solutions and insufficient data utilization.

[0009] In a first aspect, the present application provides a method for optimizing power quality in intelligent operation and maintenance of a zero-carbon factory, comprising:

[0010] A central control platform is set up in the preset zero-carbon factory; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory area, which is used to collect power generation data; the power quality analysis module includes a multi-parameter power quality detector on the zero-carbon factory distribution bus, which is used to collect power quality parameters;

[0011] Constructing a factory energy digital twin model based on the photovoltaic module characteristics, load equipment parameters, and grid access point characteristics corresponding to the zero-carbon factory;

[0012] Obtaining real-time meteorological data and historical power generation data corresponding to the zero-carbon plant;

[0013] Inputting the power generation data, power quality parameters, real-time meteorological data, and historical power generation data into the plant energy digital twin model to generate a future power generation curve;

[0014] When the power quality analysis module detects that the total harmonic distortion rate exceeds a preset threshold, the optimal control plan is matched in the preset fault mode self-learning database based on the power generation data of the zero-carbon factory and the future power generation power curve; the three-phase imbalance of the zero-carbon factory power supply network is adjusted according to the optimal control plan to complete the power quality optimization of the zero-carbon factory; the optimal control plan includes the compensation capacitor group switching strategy of the zero-carbon factory.

[0015] In some embodiments, the factory energy digital twin model is constructed according to the photovoltaic component characteristics, load equipment parameters and grid access point characteristics corresponding to the zero-carbon factory, including: constructing a three-dimensional simulation model of the photovoltaic power generation unit of the zero-carbon factory according to the photovoltaic component characteristics; constructing a load dynamic model according to the load equipment parameters; constructing a grid equivalent model according to the grid access point characteristics; based on a hybrid modeling method of a mechanism model and a data-driven model, the three-dimensional simulation model of the photovoltaic power generation unit, the load dynamic model and the grid equivalent model are coupled by multi-physical fields to form a digital twin model with real-time simulation capabilities.

[0016] Exemplarily, the method of constructing a three-dimensional simulation model of a photovoltaic power generation unit of a zero-carbon factory according to the characteristics of the photovoltaic modules; constructing a load dynamic model according to the load device parameters; and constructing a grid equivalent model according to the characteristics of the grid access point includes: parsing the characteristics of the photovoltaic modules, obtaining the temperature coefficient of the photovoltaic modules and the photoelectric conversion efficiency curve, which are used to construct the three-dimensional simulation model of the photovoltaic power generation unit; parsing the load device parameters, obtaining the power consumption characteristic curve and harmonic spectrum distribution of the load equipment of the zero-carbon factory, which are used to construct the load dynamic model; and obtaining the short-circuit capacity and system impedance ratio of the grid access point according to the characteristics of the grid access point, which are used to construct the grid equivalent model.

[0017] In some embodiments, the power generation data, power quality parameters, real-time meteorological data and historical power generation data are input into the plant energy digital twin model to generate a future power generation curve, including: normalizing and preprocessing the power generation data, power quality parameters, real-time meteorological data and historical power generation data to establish three-dimensional feature space information including irradiance, temperature and power; using a long short-term memory neural network to construct a time series prediction model corresponding to the plant energy digital twin model; the time series prediction model performs rolling prediction through a sliding time window mechanism; inputting the three-dimensional feature space information into the time series prediction model to output the future power generation curve.

[0018] Exemplarily, after inputting the three-dimensional feature space information into the time series prediction model and outputting the future power generation curve, it also includes: parsing the real-time meteorological data to obtain the cloud motion vector; and correcting the photovoltaic output prediction value of the future power generation curve according to the cloud motion vector based on a particle filtering algorithm.

[0019] In some embodiments, the matching of the optimal control plan in a preset fault mode self-learning database based on the power generation data and future power generation power curve of the zero-carbon plant includes: obtaining harmonic spectrum feature vectors and transient process recording data based on the power generation data and future power generation power curve; using a dynamic time warping algorithm to calculate the waveform similarity between the harmonic spectrum feature vectors and transient process recording data and the fault mode self-learning database to obtain multiple candidate control plans in the fault mode self-learning database; establishing a load demand response constraint condition based on the predicted slope change rate of the future power generation curve; and performing multi-objective optimization on the candidate control plans according to a genetic algorithm based on the load demand response constraint condition to determine the optimal control plan among the multiple candidate control plans.

[0020] In some embodiments, the three-phase imbalance of the zero-carbon factory power supply network is adjusted according to the optimal governance scheme to complete the optimization of the power quality of the zero-carbon factory, and also includes: controlling the static VAR generator of the zero-carbon factory power supply network for dynamic compensation; adjusting the grid-connected power factor of the energy storage system of the zero-carbon factory power supply network to compensate for the residual imbalance of the preset SVG equipment; wherein, during the compensation process, the neutral line current changes of the zero-carbon factory power supply network are monitored in real time to dynamically optimize the switching strategy of the compensation capacitor group.

[0021] In some embodiments, after completing the power quality optimization of the zero-carbon factory, it also includes: generating an encrypted electronic certificate with a timestamp from the photovoltaic power generation, reactive compensation and equivalent carbon emission reduction data of the zero-carbon factory; the electronic certificate complies with the certification standards of the carbon trading market.

[0022] In a second aspect, the present application provides a zero-carbon factory intelligent operation and maintenance central control system, the system comprising:

[0023] Central control platform; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory area, which is used to collect power generation data; the power quality analysis module includes a multi-parameter power quality detector on the zero-carbon factory distribution bus, which is used to collect power quality parameters;

[0024] A control device comprises a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method provided in any embodiment of the present application when executing the computer program.

[0025] In a third aspect, the present application provides a zero-carbon factory intelligent operation and maintenance power quality optimization device, comprising:

[0026] A platform setting unit is used to set up a central control platform in a preset zero-carbon factory; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory area, which is used to collect power generation data; the power quality analysis module includes a multi-parameter power quality detector on the zero-carbon factory distribution bus, which is used to collect power quality parameters;

[0027] A model building unit, configured to build a factory energy digital twin model based on photovoltaic module characteristics, load device parameters, and grid access point characteristics corresponding to the zero-carbon factory;

[0028] A data acquisition unit, configured to acquire real-time meteorological data and historical power generation data corresponding to the zero-carbon plant;

[0029] a curve generating unit, configured to input the power generation data, power quality parameters, real-time meteorological data, and historical power generation data into the plant energy digital twin model to generate a future power generation curve;

[0030] The optimization completion unit is used to match the optimal governance plan in the preset fault mode self-learning database according to the power generation data of the zero-carbon factory and the future power generation power curve when the power quality analysis module detects that the total harmonic distortion rate exceeds the preset threshold; adjust the three-phase imbalance of the zero-carbon factory power supply network according to the optimal governance plan to complete the power quality optimization of the zero-carbon factory; the optimal governance plan includes the compensation capacitor group switching strategy of the zero-carbon factory.

[0031] This application provides a zero-carbon factory intelligent operation and maintenance central control system and power quality optimization method and device. This platform integrates photovoltaic monitoring and power quality analysis. PV array sensors collect real-time power generation data, and multi-parameter power quality detectors monitor distribution bus parameters, achieving a closed-loop linkage between the data collection layer and the governance execution layer. This innovative application utilizes digital twin modeling technology to integrate photovoltaic module characteristics (such as conversion efficiency and attenuation coefficient), load device dynamic parameters (such as nonlinear load characteristics), and grid access point characteristics (such as short-circuit capacity and impedance characteristics) to construct a three-dimensional energy model, overcoming the limitations of traditional single-physics modeling. By establishing a four-dimensional data fusion mechanism (real-time power generation data + historical operating data + weather forecast data + power quality parameters), the digital twin model performs multivariate time series analysis to generate minute-level power generation forecast curves, achieving over 15% accuracy improvement over traditional LSTM prediction models. This application also pioneers a self-learning database for fault modes, employing a transfer learning algorithm to extract features from historical governance cases and establish a mapping database between THD exceedance events and governance strategies. When total harmonic distortion (THD) exceeds the IEC 61000-3-6 threshold, a similarity matching algorithm dynamically generates an optimal switching strategy for the compensation capacitor bank. A genetic algorithm-based three-phase imbalance optimization model is proposed to simultaneously optimize interphase current distribution during reactive power compensation, keeping imbalance within the ≤2% range specified in GB / T 15543.

[0032] By integrating multi-dimensional information such as meteorological data (irradiance, temperature, and wind speed), equipment status data (inverter efficiency, module temperature), and grid parameters (voltage fluctuations and harmonic spectrum), the digital twin model achieves rapid simulation in 500ms, three times the response speed of traditional SCADA systems. A two-tiered optimization strategy is employed: the upper tier uses Model Predictive Control (MPC) for 24-hour rolling optimization, while the lower tier utilizes a fuzzy PID controller for millisecond-level dynamic compensation, addressing power quality degradation caused by intermittent photovoltaic output and load fluctuations.

[0033] This system can control voltage fluctuations to within ±2% (better than the national standard of ±7%), reduce THD from the typical 8-12% to below 4%, reduce harmonic losses by approximately 18-25% annually, and extend the service life of sensitive equipment by over 30%. Precise capacitor bank switching strategies maintain the power factor between 0.95 and 1.0, reducing the risk of reactive power penalties. Combined with photovoltaic predictive optimization scheduling, this system reduces energy storage system capacity by 15-20%, saving approximately 2 million yuan per MW in initial investment. This system has increased the photovoltaic utilization rate to 98.5%, a 12 percentage point increase compared to traditional control methods. It can also reduce diesel generator standby time by over 600 hours annually, equivalent to reducing carbon emissions by 1,500 tons per year. The developed equipment health assessment model provides 72-hour advance warning of power quality deterioration trends, shortening operation and maintenance response time to less than 10 minutes, improving fault location accuracy to 95%, and reducing unplanned downtime losses by approximately 40%.

[0034] In summary, this technical solution effectively solves the power quality control problems brought about by the grid connection of distributed photovoltaic systems through the collaborative innovation of digital twin modeling, multi-source data fusion and intelligent optimization algorithms, and provides a technical implementation path that is both real-time and economical for the carbon neutrality goals in the industrial field, and has outstanding industrial application value.

[0035] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0037] Figure 1 This is a schematic block diagram of the structure of the zero-carbon factory intelligent operation and maintenance central control system provided in one embodiment of the present application;

[0038] Figure 2 This is a schematic flow chart of the steps of a method for optimizing power quality for intelligent operation and maintenance of a zero-carbon factory provided in one embodiment of the present application;

[0039] Figure 3 This is a schematic block diagram of the structure of a control device provided in one embodiment of the present application.

[0040] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0042] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0043] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish between identical or similar items having substantially the same functions and effects. Those skilled in the art will understand that terms such as "first" and "second" do not limit the quantity or order of execution, and that terms such as "first" and "second" do not necessarily define differences.

[0044] It should be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0045] It will also be understood that the term "and / or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0046] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0047] Driven by the "dual carbon" goal, zero-carbon factories are widely adopting distributed photovoltaic power generation systems. However, the intermittent nature of photovoltaic output and nonlinear load operation lead to problems such as excessive total harmonic distortion (THD) in the power grid and increased three-phase imbalance. Traditional methods have the following drawbacks:

[0048] 1. Disconnect between prediction and control: Existing PV monitoring systems only collect power generation data and do not dynamically link weather forecasts with power quality parameters. This results in the configuration of harmonic control equipment (such as APF) lagging behind actual demand.

[0049] 2. Insufficient model accuracy: By using a single physical model to construct the factory energy system, without integrating the attenuation characteristics of photovoltaic modules and grid impedance parameters, a power prediction error of more than 15% can be generated when the load suddenly changes.

[0050] 3. Rigid governance solutions: Using a fixed threshold-triggered compensation strategy, the impact of PV output fluctuations on grid harmonic characteristics is not considered, which can easily lead to over-compensation or under-compensation.

[0051] 4. Insufficient data utilization: Existing methods do not use machine learning to match historical fault patterns with real-time power generation curves, resulting in a three-phase unbalanced regulation response time exceeding 200ms.

[0052] Therefore, a method is urgently needed to solve at least one of the above problems.

[0053] To solve the above problems, please refer to Figure 1 The present application provides a zero-carbon factory intelligent operation and maintenance central control system, which is characterized by including: a central control platform; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory area, for collecting power generation data; the power quality analysis module includes a multi-parameter power quality detector on the distribution bus of the zero-carbon factory, for collecting power quality parameters; a control device includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method provided in any embodiment of the present application when executing the computer program.

[0054] Exemplarily, the control device is equipped to perform the following method: a central control platform is set up in a preset zero-carbon factory; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory, for collecting power generation data; the power quality analysis module includes a multi-parameter power quality detector on the distribution bus of the zero-carbon factory, for collecting power quality parameters; a factory energy digital twin model is constructed according to the photovoltaic component characteristics, load equipment parameters and grid access point characteristics corresponding to the zero-carbon factory; and real-time meteorological data corresponding to the zero-carbon factory are obtained. The power generation data, power quality parameters, real-time meteorological data and historical power generation data are input into the plant energy digital twin model to generate a future power generation curve; when the power quality analysis module detects that the total harmonic distortion rate exceeds a preset threshold, the optimal control plan is matched in the preset fault mode self-learning database according to the power generation data and future power generation curve of the zero-carbon plant; the three-phase imbalance of the power supply network of the zero-carbon plant is adjusted according to the optimal control plan to complete the power quality optimization of the zero-carbon plant; the optimal control plan includes the compensation capacitor group switching strategy of the zero-carbon plant.

[0055] Specifically, the present invention provides a zero-carbon factory intelligent operation and maintenance central control system, which realizes the dynamic coordination of power generation prediction and power quality management by constructing a digital twin model and an adaptive learning mechanism. The central control platform integrates a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module: a photovoltaic array sensor group (including irradiance sensor, temperature sensor, current / voltage sensor) deployed on the roof of the factory area, collects power generation data (such as DC side voltage, current, power) in real time; the power quality analysis module installs a multi-parameter power quality detector (supporting IEC 61000-4-30 standard) on the distribution bus side to collect THD, three-phase imbalance, harmonic spectrum and voltage fluctuation data; the control device is configured with a memory and a processor to perform digital twin modeling, dynamic prediction and management strategy optimization; the factory energy digital twin model integrates the photovoltaic component attenuation model (based on the Arrhenius equation to simulate the aging rate), load equipment parameters (such as nonlinear load harmonic emission coefficient) and grid impedance parameters (online identification through the sweep frequency method) to construct a multi-physics field coupling model;

[0056] The fault mode self-learning database stores historical fault modes (such as APF overload events and capacitor bank switching failure records) and corresponding treatment plans, and dynamically updates the weights through a reinforcement learning algorithm.

[0057] For example, a distributed sensor group is deployed in the photovoltaic array to upload power generation data to the central control platform with a sampling period of 1 second. The THD and three-phase imbalance are monitored in real time through the power quality detector on the distribution bus, and the data is synchronized to the digital twin model. Based on the attenuation curve provided by the photovoltaic module manufacturer (such as an annual attenuation rate of 0.5%) and real-time meteorological data (wind speed, cloud coverage), the photovoltaic output prediction model is dynamically corrected.

[0058] Real-time meteorological data (such as irradiance forecasts for the next 15 minutes) and historical power generation curves are input into the digital twin model to generate a high-precision power generation curve (error ≤ 5%). When the THD exceeds a preset threshold (such as 5%), fault pattern matching is initiated: the current harmonic spectrum characteristics (such as the proportion of the fifth harmonic exceeding 40%) are extracted and combined with the future power generation curve to match the optimal treatment plan in the self-learning database.

[0059] If PV output is predicted to drop by 30% within 10 minutes, the APF dynamic capacity increase mode is prioritized over capacitor bank switching to avoid overcompensation. Based on the real-time power generation curve, the load deviation rate of each phase is calculated. A particle swarm optimization algorithm is used to dynamically adjust the compensation capacitor bank switching strategy (for example, switching two capacitor banks on phase C), reducing the three-phase imbalance from 8% to within 2% and shortening the response time to 50ms. After each treatment is completed, the deviation between the actual THD rate of change and the predicted value is recorded, and the weight coefficients of the digital twin model are updated using the backpropagation algorithm. When PV module efficiency degradation is detected (for example, a 3% year-on-year decrease in output power), the model parameters are automatically adjusted to match the current state.

[0060] By real-time correlation of meteorological data with power quality parameters, the APF configuration response time is shortened by 60%; by integrating a multi-source data model of photovoltaic attenuation characteristics and grid impedance, the power prediction error is reduced from 15% to less than 5%; dynamic threshold adjustment based on power generation fluctuation prediction reduces the probability of overcompensation by 40%; matching historical fault patterns with real-time data shortens the three-phase unbalance adjustment response time from 200ms to 50ms; and by optimizing the capacitor bank switching strategy, the number of switching operations is reduced by 30%, extending the service life of the compensation device.

[0061] Take a zero-carbon automobile factory as an example: a 500kW photovoltaic array was installed on the factory roof, equipped with a temperature and humidity composite sensor and a string current monitoring unit; power quality detectors were deployed on the distribution bus side to monitor the content of harmonics below 19 in real time; when midday clouds caused a sudden drop in irradiance, the system enabled the APF expansion mode in advance based on the predicted curve, reducing the THD from 6.2% to 4.1%, avoiding the THD exceeding the standard (7.5%) caused by response delays in traditional solutions.

[0062] Through the deep integration of digital twin and machine learning technologies, this invention solves the problem of coordination between photovoltaic power generation and power quality management in zero-carbon factories, and provides a reliable solution for smart factories with a high proportion of renewable energy access.

[0063] See also Figure 2 , Figure 2 This is a schematic flow chart of a method for optimizing power quality of a zero-carbon factory intelligent operation and maintenance provided by an embodiment of the present application. The execution device of the method is the control device of the zero-carbon factory intelligent operation and maintenance central control system provided by any embodiment of the present application.

[0064] like Figure 2 As shown, the provided method includes steps S101 to S105. The control device may be a handheld terminal, a laptop computer, a wearable device, or a robot, etc., for implementing steps S101 to S105 and their corresponding embodiments.

[0065] Step S101. Set up a central control platform in a preset zero-carbon factory; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory, which is used to collect power generation data; the power quality analysis module includes a multi-parameter power quality detector on the distribution bus of the zero-carbon factory, which is used to collect power quality parameters.

[0066] Specifically, this step aims to establish a core control architecture within the zero-carbon factory, integrating photovoltaic monitoring modules with power quality analysis modules to form a dynamic energy perception network for the entire factory. The photovoltaic monitoring module is deployed on the photovoltaic array on the factory roof and consists of a high-precision irradiance sensor, a photovoltaic panel temperature sensor, and a current / voltage transformer. It collects real-time data on the DC-side power generation power, component surface temperature, and ambient light intensity of the photovoltaic modules. The power quality analysis module uses a multi-parameter power quality detector (integrated with a high-speed sampling chip and FFT analysis unit) installed on the distribution bus to simultaneously monitor key parameters such as three-phase voltage / current harmonic content, phase offset, power factor, and imbalance. The two modules are interconnected with the central control platform via industrial Ethernet, achieving millisecond-level synchronous transmission of data between the power generation and consumption sides.

[0067] The PV monitoring module deployment includes, for example, a set of sensor nodes per 100 square meters of PV array. These include: Irradiance sensor: An amorphous silicon sensor with a spectral response range of 300-1100 nm, mounted vertically on the PV panel surface with a sampling frequency of 1 Hz. Temperature sensor: A PT1000 surface-mount platinum resistor embedded in the PV backplane monitors the operating temperature of the module. Current / voltage acquisition unit: A Hall effect sensor (accuracy of ±0.5%) connected in parallel to the string combiner box collects DC output. Communication architecture: Sensor nodes connect to an edge gateway via an RS-485 bus. The gateway has a built-in Modbus TCP protocol stack, which aggregates data into the central control platform's time series database.

[0068] The power quality analysis module deployment includes: Monitoring point placement: A multi-parameter power quality monitor (such as the Fluke 1750 series) is installed on the outgoing low-voltage busbar side of the distribution room. Each phase is equipped with an independent sampling channel (sampling rate of 256 points / cycle), which calculates THD (total harmonic distortion), interharmonic content, and negative sequence components in real time. Data processing: The monitor's built-in DSP chip performs real-time harmonic analysis (based on the IEC 61000-4-30 standard) and uploads the data to the central control platform via the OPC UA protocol.

[0069] By synchronously collecting power output parameters on the generator side and power quality parameters on the consumer side, this system breaks down traditional data silos and provides a data foundation for dynamic governance. Millisecond-level sampling and normalization algorithms ensure THD detection error ≤ 0.2% and temperature monitoring accuracy ±0.5°C, providing reliable data for model input. Industrial Ethernet communication reduces data latency to <10ms, eliminating the delay in governance device response caused by data lag in traditional systems.

[0070] Step S102: Construct a factory energy digital twin model based on the photovoltaic module characteristics, load equipment parameters and grid access point characteristics corresponding to the zero-carbon factory.

[0071] Specifically, this step constructs a multi-physics field coupled digital twin model by integrating the physical characteristics of photovoltaic modules, the dynamic response characteristics of the load, and the grid impedance parameters. The model includes a photovoltaic power generation prediction sub-model, a grid harmonic propagation sub-model, and a load disturbance response sub-model, and uses modular modeling technology to achieve interactive simulation of each subsystem. The photovoltaic power generation sub-model is based on the IV curve data provided by the photovoltaic module manufacturer (including the attenuation coefficient β = 0.5% / year) to construct a single diode equivalent circuit model. The formula is:

[0072] ;

[0073] in, is the photocurrent, The reverse saturation current is dynamically corrected based on the real-time temperature and irradiance. Attenuation compensation is achieved by introducing a service life factor α=1-β×T to compensate for the annual attenuation of the module output power (T is the number of years of operation). ≈5 to 8 A (standard test conditions: irradiance 1000 W / m², temperature 25°C). This is slightly lower than for polycrystalline silicon modules, approximately 5%. is the internal resistance of the battery, including the electrode contact resistance and the bulk resistance of the material. It is calculated from the slope of the IV curve near the open-circuit point under illumination. n is the diode ideality factor, a correction factor that reflects the non-ideal characteristics of the PN junction. n=1 represents an ideal diode. is the thermovoltage, the temperature-dependent thermodynamic voltage. Characterizes the equivalent resistance of battery edge leakage current and surface defects. V represents the leakage current of a diode under reverse bias, reflecting material defects and recombination losses. V represents the DC terminal voltage of a photovoltaic module under specific operating conditions. It is formed by the potential difference generated by the separation of photogenerated carriers at the PN junction. This voltage varies with the load current, following the IV characteristic curve. V, together with the output current I, constitutes the power output characteristic of a photovoltaic module (P = V × I), which is the core variable for power prediction in the digital twin model.

[0074] The grid harmonic propagation sub-model includes the calculation of the grid impedance frequency characteristics based on the distribution network topology using the node admittance matrix method. The formula is:

[0075] ;

[0076] Where R, L, and C are equivalent line parameters, measured using the swept-frequency method. R is the equivalent AC resistance of the power line, including skin effect and proximity effect. The impedance-frequency curve is used to identify system resonance points and predict the amplification risk of specific harmonics (such as the 5th and 7th harmonics). For example, for low-voltage cables (copper core), R = 0.1 to 0.3 Ω / km. For medium-voltage lines (aluminum stranded wire), R = 0.2 to 0.5 Ω / km.

[0077] The load disturbance response sub-model establishes a time-domain model based on the VI characteristics for equipment such as inverters and arc furnaces, and uses state-space equations to describe their harmonic emission characteristics:

[0078] ;

[0079] Where u is the input voltage disturbance and y is the output current harmonic spectrum.

[0080] Step S103: Obtain real-time meteorological data and historical power generation data corresponding to the zero-carbon factory.

[0081] Specifically, this step uses multi-source data fusion technology to integrate meteorological satellite data, local weather station observations, and historical power generation records to construct a meteorological-power generation correlation database covering both the short-term (0-4 hours) and the medium- to long-term (4-72 hours). This satellite data is integrated with shortwave radiation forecast data from the China Meteorological Administration's Fengyun-4 satellite (with a spatial resolution of 1 km and a temporal resolution of 15 minutes) to extract the GHI (global horizontal irradiance) forecast for the power plant's grid. Local observations involve deploying a Total Sky Imager (TSI-880) at a commanding height within the plant area to monitor cloud movement speed and direction in real time and correct for local errors in the satellite data.

[0082] Data cleaning involves removing outliers (using the 3σ principle) and interpolating missing values ​​(using the KNN algorithm) on the past five years of power generation data. Feature engineering involves extracting the peak-to-valley difference, volatility, and Pearson correlation coefficients of daily power generation curves with temperature and irradiance to construct a feature vector library.

[0083] The fusion of multi-source meteorological data reduces the irradiance prediction error from 20% to 8%. The historical data feature library supports similar day matching, improving the model's generalization ability in extreme weather conditions.

[0084] Step S104: Input power generation data, power quality parameters, real-time meteorological data, and historical power generation data into the plant energy digital twin model to generate a future power generation curve.

[0085] Real-time data is input into the digital twin model, and a physical model and data-driven hybrid prediction method is used to generate a minute-by-minute power generation curve for the next 24 hours and quantify the uncertainty of the prediction results.

[0086] Based on the PV sub-model in step S102, the theoretical power generation Pphy is calculated using an LSTM neural network. The inputs are the historical power series, weather forecast values, and module temperature. The correction factor α is output, resulting in a final power of Pphy*α. Monte Carlo simulation generates 1000 sets of disturbance samples and outputs a 90% confidence interval for the power forecast. The latest weather data is received every 15 minutes, triggering a model recalibration to ensure that the forecast curve dynamically adjusts to weather changes.

[0087] The hybrid model ensures a 15-minute short-term forecast error of ≤3% and a 4-hour forecast error of ≤7%. The confidence interval output can identify the risk of sudden power generation drops (such as cloud cover) in advance, supporting the pre-activation of control equipment.

[0088] Step S105. When the power quality analysis module detects that the total harmonic distortion rate exceeds a preset threshold, the optimal control plan is matched in the preset fault mode self-learning database based on the power generation data of the zero-carbon factory and the future power generation power curve; the three-phase imbalance of the zero-carbon factory power supply network is adjusted according to the optimal control plan to complete the power quality optimization of the zero-carbon factory; the optimal control plan includes the compensation capacitor group switching strategy of the zero-carbon factory.

[0089] Specifically, when THD exceeds the standard, the system combines the real-time power generation curve with the prediction results, matches the optimal control strategy from the fault mode self-learning database, and dynamically adjusts the APF (active power filter) compensation current and capacitor bank switching plan.

[0090] The knowledge base is constructed based on historical fault records (such as THD exceeding limits and voltage swells), extracting feature vectors (harmonic spectrum and power fluctuation rate), and using hierarchical clustering to generate typical fault modes. The Euclidean distance between the current operating condition and each cluster center is calculated in real time, and the closest mode is selected as the basis for treatment. Based on the predicted PV output fluctuations, the model predictive control (MPC) algorithm is used to continuously optimize the harmonic current injection amount of the APF. The objective function is:

[0091] ;

[0092] in is the change in APF output current, and λ is the weight coefficient. The optimal switching combination is solved based on a genetic algorithm, with the goal of minimizing three-phase imbalance and network losses. The constraints include the life of the capacitor bank (switching frequency ≤ 5 times / hour).

[0093] In some embodiments, the factory energy digital twin model is constructed according to the photovoltaic component characteristics, load equipment parameters and grid access point characteristics corresponding to the zero-carbon factory, including: constructing a three-dimensional simulation model of the photovoltaic power generation unit of the zero-carbon factory according to the photovoltaic component characteristics; constructing a load dynamic model according to the load equipment parameters; constructing a grid equivalent model according to the grid access point characteristics; based on a hybrid modeling method of a mechanism model and a data-driven model, the three-dimensional simulation model of the photovoltaic power generation unit, the load dynamic model and the grid equivalent model are coupled by multi-physical fields to form a digital twin model with real-time simulation capabilities.

[0094] The three-dimensional simulation model of the photovoltaic power generation unit is constructed based on the electrical characteristics (such as open-circuit voltage and short-circuit current) and physical layout parameters (such as tilt angle and azimuth) of the photovoltaic components. The geometric model of the photovoltaic array is established using three-dimensional modeling software (such as ANSYS Twin Builder), and the equivalent circuit model of the photovoltaic cell is integrated to realize the dynamic simulation of the effect of light intensity and temperature changes on the power generation.

[0095] The load dynamic model is constructed by analyzing the nameplate parameters of the load equipment (such as rated power and power factor) and operation log data to establish a time domain simulation model that includes motor starting impact and nonlinear load harmonic emission characteristics, and uses state space equations to describe the dynamic response characteristics of the load.

[0096] The grid equivalent model is constructed based on the short-circuit capacity and impedance ratio of the grid access point, and a Thevenin equivalent circuit model is established. The grid harmonic background distortion rate parameters are embedded to simulate the impact of grid voltage fluctuations on the power quality of the factory.

[0097] Multi-physics field coupling couples the above models through a joint simulation platform (such as COMSOL Multiphysics), adopts the finite element method to solve the interaction between the electromagnetic field, thermal field and mechanical field, and introduces historical operation data to perform online calibration of the mechanism model parameters to enhance the real-time simulation capability of the model.

[0098] Through multi-physics field coupling modeling, the interaction between photovoltaic power generation, load dynamics and the grid environment is accurately reflected, improving the adaptability of the digital twin model to complex working conditions;

[0099] The hybrid modeling approach combines the theoretical rigor of the mechanism model with the flexibility of the data-driven model, significantly improving simulation accuracy and providing a reliable decision-making basis for power quality optimization.

[0100] Exemplarily, the method of constructing a three-dimensional simulation model of a photovoltaic power generation unit of a zero-carbon factory according to the characteristics of the photovoltaic modules; constructing a load dynamic model according to the load device parameters; and constructing a grid equivalent model according to the characteristics of the grid access point includes: parsing the characteristics of the photovoltaic modules, obtaining the temperature coefficient of the photovoltaic modules and the photoelectric conversion efficiency curve, which are used to construct the three-dimensional simulation model of the photovoltaic power generation unit; parsing the load device parameters, obtaining the power consumption characteristic curve and harmonic spectrum distribution of the load equipment of the zero-carbon factory, which are used to construct the load dynamic model; and obtaining the short-circuit capacity and system impedance ratio of the grid access point according to the characteristics of the grid access point, which are used to construct the grid equivalent model.

[0101] The photovoltaic monitoring module collects real-time module backsheet temperature data. The photovoltaic conversion efficiency curve is corrected using a temperature coefficient (e.g., -0.45% / °C) to construct a three-dimensional simulation model that accounts for hot-spot effects. A Fourier transform analysis is performed on historical electricity consumption data to extract the characteristic harmonic spectra of the load devices (e.g., the proportion of the 5th and 7th harmonics). This allows for the development of a dynamic load model that includes harmonic current injection capabilities. The power quality analysis module measures the three-phase voltage imbalance at the grid access point. The system impedance ratio (X / R ratio) is calculated using the short-circuit capacity, allowing for the construction of a grid equivalent model capable of simulating voltage sags and harmonic resonances.

[0102] The photovoltaic model is dynamically corrected based on the temperature coefficient and irradiance to accurately predict the power drop caused by shadow shading; the harmonic spectrum characteristics are integrated into the load model to provide a data basis for matching subsequent harmonic control solutions; the grid equivalent model reflects the actual impedance characteristics and effectively predicts the risk of harmonic amplification.

[0103] For example, the energy digital twin models of multiple zero-carbon factories are distributedly and collaboratively trained through a federated learning framework, specifically including: local training of the LSTM prediction network of the digital twin model in each factory, encrypting and uploading the model gradient to the central server; aggregating the global gradient and sending it to each node to achieve model parameter updates and privacy protection.

[0104] Federated Learning Architecture: Participants: Multiple zero-carbon factories act as clients, and a central server aggregates model parameters using the Paillier homomorphic encryption protocol. Each factory only shares the LSTM network weights ΔW and does not transmit raw power generation data.

[0105] The training process includes: Local training: Each plant trains the LSTM prediction model based on its own historical data (such as a 100-day photovoltaic power generation sequence) and calculates the gradient gi = L(W); Gradient encryption: Gradients are noised using differential privacy (noise standard deviation σ = 0.1) and encrypted using the SM9 algorithm before upload. Global aggregation: The server performs a weighted average of the encrypted gradients (weighted by the data volume share of each plant) to update the global model Wglobal = W - η∑gi. Federated learning iterations are performed every 24 hours to dynamically adapt to regional weather and load variations. Cross-plant data collaboration improves the generalization of the prediction model, reducing prediction errors in cloudy areas by 12%. Encryption and differential privacy technologies ensure data security and meet GDPR compliance requirements.

[0106] In some embodiments, the power generation data, power quality parameters, real-time meteorological data and historical power generation data are input into the plant energy digital twin model to generate a future power generation curve, including: normalizing and preprocessing the power generation data, power quality parameters, real-time meteorological data and historical power generation data to establish three-dimensional feature space information including irradiance, temperature and power; using a long short-term memory neural network to construct a time series prediction model corresponding to the plant energy digital twin model; the time series prediction model performs rolling prediction through a sliding time window mechanism; inputting the three-dimensional feature space information into the time series prediction model to output the future power generation curve.

[0107] The power generation data (such as DC side voltage and inverter efficiency) and meteorological data (such as irradiance and wind speed) are normalized by Min-Max to eliminate dimensional differences. The time series prediction model is constructed using an LSTM neural network architecture, with the number of input layer nodes set to the feature dimension (irradiance, temperature, power), the number of hidden layer neurons set to 64, and the output layer being the power generation sequence for the next hour. The rolling prediction mechanism uses a 15-minute time window step, and updates the input data after each prediction to achieve continuous 72-hour power curve generation.

[0108] The power generation data collected by the photovoltaic monitoring module, such as the DC side current / voltage and inverter AC output power, as well as the irradiance (unit: W / m²) and ambient temperature (unit: °C) data obtained by the weather station, are each subjected to Min-Max normalization processing, and each parameter is mapped to the [0, 1] interval. The formula is:

[0109] ;

[0110] Historical power generation data and real-time data are aligned by timestamp to construct a three-dimensional dataset containing time series features. The dimensions are time step (such as 15-minute interval), number of features (irradiance, temperature, power), and sample batch.

[0111] The time series forecasting model uses a two-layer LSTM neural network. The input layer has 3 nodes (corresponding to irradiance, temperature, and power), the hidden layer has 128 neurons, and a dropout rate of 0.2 to prevent overfitting. The output layer is a fully connected layer, generating a series of power generation forecasts for the next 24 hours (96 time points). The model is trained using the Adam optimizer, the mean squared error (MSE) loss function, the initial learning rate set to 0.001, and early stopping (patience = 10) to dynamically adjust the number of training rounds.

[0112] The rolling forecast mechanism uses a sliding window length of 24 hours and a step size of 1 hour. Each forecast is made by concatenating the most recent hour's real-time data (four time points) with the historical 23 hours of data to form an input sequence. The power value for the next hour is then predicted, and the input window is cyclically updated to achieve a continuous 72-hour rolling forecast.

[0113] Normalization eliminates dimensional differences and improves model convergence speed. The dual-layer structure of the LSTM network effectively captures long-term and short-term dependencies, making it particularly suitable for predicting power fluctuations caused by sudden changes in irradiance in cloudy weather.

[0114] The sliding window mechanism dynamically integrates real-time data, reducing the prediction error by more than 15% compared with traditional static models.

[0115] Exemplarily, after inputting the three-dimensional feature space information into the time series prediction model and outputting the future power generation curve, it also includes: parsing the real-time meteorological data to obtain the cloud motion vector; and correcting the photovoltaic output prediction value of the future power generation curve according to the cloud motion vector based on a particle filtering algorithm.

[0116] Cloud motion vector extraction uses meteorological radar data or all-sky imagers to obtain the cloud movement speed and direction, and construct a cloud shading distribution heat map for the next 30 minutes. Particle filter correction uses photovoltaic array partitions as particle groups, adjusts the weight of each particle according to the cloud shading probability, performs Monte Carlo simulation on the predicted power curve, and outputs the corrected confidence interval power value.

[0117] Sky images are acquired using an all-sky imager, and the optical flow method is used to calculate the speed and direction of cloud movement. This generates a two-dimensional motion vector field with the center of the photovoltaic array as the origin. The cloud position for the next 15 minutes is extrapolated based on the vector field, and the photovoltaic array is divided into a 10×10 grid. The probability of each grid cell being obscured by clouds is calculated (in the range of 0-1).

[0118] Each photovoltaic array string is defined as a particle, and a total of 1000 particles are generated. The status of each particle includes position, shading probability, and output power correction factor. The particle weight is adjusted according to the cloud motion vector. For example, when the cloud moves westward, the particle weight of the western array increases. The residual between the actual power and the predicted power is used as the observation value, and Monte Carlo sampling is used to redistribute the particles to generate the corrected power prediction value and its confidence interval (such as 95% confidence level).

[0119] The cloud motion vector quantifies the impact of short-term occlusion, solving the prediction lag problem caused by the rapid movement of clouds in traditional models; the particle filter integrates probability prediction and physical model, reducing the power prediction error within 15 minutes from 8% to below 3%.

[0120] In some embodiments, the matching of the optimal control plan in a preset fault mode self-learning database based on the power generation data and future power generation power curve of the zero-carbon plant includes: obtaining harmonic spectrum feature vectors and transient process recording data based on the power generation data and future power generation power curve; using a dynamic time warping algorithm to calculate the waveform similarity between the harmonic spectrum feature vectors and transient process recording data and the fault mode self-learning database to obtain multiple candidate control plans in the fault mode self-learning database; establishing a load demand response constraint condition based on the predicted slope change rate of the future power generation curve; and performing multi-objective optimization on the candidate control plans according to a genetic algorithm based on the load demand response constraint condition to determine the optimal control plan among the multiple candidate control plans.

[0121] Waveform similarity calculation performs dynamic time warping (DTW) on transient recording data (such as voltage swell waveforms), calculates its morphological distance with historical waveforms in the database, and screens candidate solutions with a similarity higher than 90%. Multi-objective optimization uses the NSGA-II genetic algorithm to conduct a Pareto frontier search for candidate solutions, with the lowest control cost, the largest reduction in harmonic distortion rate, and the longest life of compensation equipment as the objective functions, and ultimately selects the best control solution through fuzzy decision-making.

[0122] Extract the harmonic spectrum feature vectors H3, H5, ..., H50 (unit: %) from transient recording data (such as voltage swell waveforms) collected by the power quality analysis module. Use the dynamic time warping (DTW) algorithm to calculate the morphological distance between the current harmonic spectrum and the historical waveforms in the database. The expression includes:

[0123] ;

[0124] Where π is the alignment path, and the candidate solutions whose DTW distance is less than a threshold (such as 0.1) are screened.

[0125] The objective functions are set as follows: treatment cost f1(x), THD reduction rate f2(x), and capacitor bank life loss f3(x). The NSGA-II genetic algorithm is used: the population size is set to 200, the crossover probability is 0.8, and the mutation probability is 0.05. The Pareto front solution set is generated by non-dominated sorting. The comprehensive optimal solution is selected through the fuzzy membership function, for example, the weight distribution is f1:0.4, f2:0.4, and f3:0.2.

[0126] The DTW algorithm overcomes the problem of waveform time axis scaling and deformation, and the similarity matching accuracy is increased to 92%; multi-objective optimization balances economic efficiency and governance effect, avoiding overcompensation caused by a single solution (for example, THD is reduced from 8% to 2%, and the cost is controlled within ￥5000).

[0127] In some embodiments, the three-phase imbalance of the zero-carbon factory power supply network is adjusted according to the optimal governance scheme to complete the optimization of the power quality of the zero-carbon factory, and also includes: controlling the static VAR generator of the zero-carbon factory power supply network for dynamic compensation; adjusting the grid-connected power factor of the energy storage system of the zero-carbon factory power supply network to compensate for the residual imbalance of the preset SVG equipment; wherein, during the compensation process, the neutral line current changes of the zero-carbon factory power supply network are monitored in real time to dynamically optimize the switching strategy of the compensation capacitor group.

[0128] SVG dynamic compensation controls the SVG's IGBT bridge arm to generate reverse harmonic current based on the harmonic current command value in the optimal control solution, compensating for the 5th and 7th characteristic harmonics in real time. When the SVG capacity is insufficient, the energy storage system collaboratively adjusts the reactive output mode of the energy storage converter to compensate for the remaining imbalance. At the same time, it monitors the zero-sequence component of the neutral line current and dynamically adjusts the switching order of the capacitor banks to avoid resonance.

[0129] Based on the harmonic current command generated by the optimal control solution (such as the 5th harmonic compensation amount = 20A), the SVG's IGBT module is controlled to generate reverse harmonic current and track and compensate in real time. Dq rotating coordinate system decoupling control is adopted, and zero-static-error tracking of harmonic current is achieved through a PI regulator (proportional coefficient Kp = 0.5, integral coefficient Ki = 0.1).

[0130] When the SVG capacity is insufficient (for example, the compensation demand exceeds 80% of its rated capacity), the energy storage converter switches to reactive power priority mode and outputs the remaining compensation ΔQ = Qdemand - QSVG. The zero-sequence component I0 of the neutral current is monitored in real time. If I0 > 10% IN (IN is the rated current), the capacitor bank switching strategy is optimized: a binary search method is used to dynamically adjust the capacitor bank switching sequence to prevent capacitive compensation from causing the resonant frequency to fall into the harmonic frequency band (such as near the 11th harmonic).

[0131] Through the coordinated control of SVG and energy storage, the three-phase imbalance is reduced from 15% to less than 2%; the neutral line current feedback mechanism prevents the compensation equipment from causing resonance, and the system stability is improved by 30%.

[0132] In some embodiments, after completing the power quality optimization of the zero-carbon factory, it also includes: generating an encrypted electronic certificate with a timestamp from the photovoltaic power generation, reactive compensation and equivalent carbon emission reduction data of the zero-carbon factory; the electronic certificate complies with the certification standards of the carbon trading market.

[0133] The data of photovoltaic power generation, reactive power compensation, etc. are stored in a distributed manner through IPFS, and a unique hash value is generated. The hash value is bound to the timestamp based on the smart contract, and the national secret SM2 algorithm is used to sign it to generate carbon emission reduction certificates that comply with VCS (Verified Carbon Standard), supporting atomic transactions on exchanges.

[0134] For example, the power quality analysis module extracts daily photovoltaic power generation (EPV) and reactive power compensation to calculate equivalent carbon emissions reductions (assuming a grid carbon emission factor of 0.8). This data is uploaded to the Hyperledger Fabric blockchain network, where a unique data fingerprint is generated using the SHA-256 algorithm and attribute-encrypted using the national SM9 algorithm to ensure decryption only by authorized parties. Based on the smart contract, a Verified Emission Reduction (VER) certificate compliant with the VCS standard is automatically generated. The certificate contains the following fields: Project ID: CN-2023-ZCFACTORY-001; Timestamp: ISO 8601 format (e.g., 2023-07-20T08:30:00Z); Emission reduction amount; and a digital signature: the certificate hash value is signed using the SM2 elliptic curve algorithm.

[0135] Blockchain technology ensures that data cannot be tampered with, meeting the carbon trading market's requirements for data traceability; encrypted electronic certificates support one-click order transactions, shortening the carbon asset realization cycle from 30 days to within 24 hours.

[0136] For example, when carbon emission reduction electronic certificates are traded on the blockchain market, a smart contract automatically triggers the expansion of the zero-carbon factory's reactive power compensation equipment. Specifically, this involves calculating the available SVG capacity based on the carbon trading proceeds and invoking the central control platform via chaincode to execute the equipment upgrade. The smart contract's logic is designed as follows: if the carbon certificate transaction amount exceeds a threshold (e.g., ¥10,000), the expansion command is triggered. Upon receiving the expansion command, the central control platform automatically sends an order to the SVG manufacturer (e.g., adding a 100kVar module). When new equipment is added, it is automatically registered with the power supply network control system via the IEC 61850 protocol. 80% of the carbon trading proceeds are earmarked for equipment procurement, and 20% is deposited into the smart contract's reserve pool. This creates a positive feedback loop of "carbon asset monetization - equipment upgrade - enhanced emission reduction capabilities," promoting the sustainable development of zero-carbon factories. The automated execution of the smart contract reduces manual intervention, shortening the equipment expansion cycle from 30 days to 7.

[0137] In some embodiments, the collaborative compensation strategy of SVG and energy storage system is dynamically optimized through deep reinforcement learning algorithm (DRL), specifically including: constructing a state space with three-phase imbalance, harmonic distortion rate, and equipment loss rate, taking SVG output current instruction and energy storage power adjustment as action space, and maximizing the improvement of comprehensive power quality indicators as reward function; using proximal policy optimization (PPO) algorithm to train the intelligent agent and generate optimal compensation instructions in real time. The state space includes parameters such as real-time three-phase current imbalance (ϵ), total harmonic distortion rate (THD), SVG module temperature (TSVG), capacitor bank switching times (Nswitch), etc., which are normalized into a 10-dimensional vector. Action space: defines the SVG's d-axis / q-axis harmonic compensation current (Id, Iq) and the energy storage system reactive power adjustment (ΔQ), a total of 3-dimensional continuous actions. Reward function:

[0138] ;

[0139] The weight coefficients are w1=0.4, w2=0.3, w3=0.2, and w4=0.1.

[0140] The PPO algorithm is trained using an actor-critic network structure. The actor network is a three-layer fully connected network (256-128-64 nodes) and the critic network is a two-layer fully connected network (128-64 nodes). The discount factor γ is set to 0.99, the clip range ϵ is set to 0.2, and the policy is updated every 1000 steps. Random harmonic disturbances (e.g., a 3% to 15% THD change) are injected into the simulation environment, and training is performed for 500,000 iterations until convergence. The trained policy network is embedded in the central control platform, which receives power quality parameters in real time, outputs compensation commands, and controls device operations.

[0141] Dynamically adapting to complex harmonic scenarios, it improves response speed by 40% compared to traditional PI control strategies. It also balances power quality and equipment life through a multi-objective reward function, extending the service life of capacitor banks by over 30%.

[0142] In some embodiments, edge computing nodes are deployed on the distribution bus side of the zero-carbon factory, and power quality anomalies are detected in real time through a lightweight convolutional neural network (CNN). Specifically, the following steps are performed: wavelet transform is performed on the voltage / current waveforms collected by the multi-parameter power quality detector to generate a time-frequency graph; and a pruned MobileNetV3 model is used to classify the anomaly type (such as voltage sag and harmonic resonance).

[0143] Edge computing devices with 128 CUDA cores and power consumption of less than 15W were used. A lightweight CNN model (model size less than 5MB) was deployed, enabling inference within 20ms. Six-layer Db4 wavelet decomposition was performed on the voltage signal to generate a time-frequency energy distribution map (size 64×64). MobileNetV3 was subjected to channel pruning (removing 50% redundant convolution kernels) and knowledge distillation technology (using a ResNet50 teacher model) was used to improve accuracy. Seven types of anomalies were defined: voltage sag, swell, harmonics, flicker, interruption, oscillation, and resonance. Output confidence levels greater than 90% trigger an alarm on the central control platform.

[0144] Localized processing at the edge nodes reduces data transmission latency, with detection response time less than 50ms. The pruned model reduces memory usage by 70% while maintaining 95% accuracy.

[0145] The embodiments of the present application also provide a zero-carbon factory intelligent operation and maintenance power quality optimization device. This zero-carbon factory intelligent operation and maintenance power quality optimization device is used to execute the steps of the zero-carbon factory intelligent operation and maintenance power quality optimization method shown in the above embodiments. This zero-carbon factory intelligent operation and maintenance power quality optimization device can be a single server or a server cluster, or this zero-carbon factory intelligent operation and maintenance power quality optimization device can be a terminal, which can be a handheld terminal, a laptop computer, a wearable device, or a robot.

[0146] The zero-carbon factory intelligent operation and maintenance power quality optimization device includes:

[0147] A platform setting unit is used to set up a central control platform in a preset zero-carbon factory; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory area, which is used to collect power generation data; the power quality analysis module includes a multi-parameter power quality detector on the zero-carbon factory distribution bus, which is used to collect power quality parameters;

[0148] A model building unit, configured to build a factory energy digital twin model based on photovoltaic module characteristics, load device parameters, and grid access point characteristics corresponding to the zero-carbon factory;

[0149] A data acquisition unit, configured to acquire real-time meteorological data and historical power generation data corresponding to the zero-carbon plant;

[0150] a curve generating unit, configured to input the power generation data, power quality parameters, real-time meteorological data, and historical power generation data into the plant energy digital twin model to generate a future power generation curve;

[0151] The optimization completion unit is used to match the optimal governance plan in the preset fault mode self-learning database according to the power generation data of the zero-carbon factory and the future power generation power curve when the power quality analysis module detects that the total harmonic distortion rate exceeds the preset threshold; adjust the three-phase imbalance of the zero-carbon factory power supply network according to the optimal governance plan to complete the power quality optimization of the zero-carbon factory; the optimal governance plan includes the compensation capacitor group switching strategy of the zero-carbon factory.

[0152] It should be noted that technical personnel in the relevant field can clearly understand that for the convenience and conciseness of description, the specific working processes of the zero-carbon factory intelligent operation and maintenance power quality optimization device and each unit described above can refer to the corresponding processes in the zero-carbon factory intelligent operation and maintenance power quality optimization method embodiments described in the above embodiments, and will not be repeated here.

[0153] The above-mentioned method for optimizing power quality for intelligent operation and maintenance of a zero-carbon factory is implemented in the form of a computer program, which can be run on the above-mentioned device.

[0154] See also Figure 3 , Figure 3 1 is a schematic block diagram of the structure of a control device provided in an embodiment of the present application. The control device includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and an internal memory.

[0155] The storage medium may store an operating device and a computer program. The computer program includes program instructions, which, when executed, may cause a processor to execute any embodiment of the method for optimizing power quality for intelligent operation and maintenance of a zero-carbon plant.

[0156] The processor is used to provide computing and control capabilities to support the operation of the entire control device.

[0157] The internal memory provides an environment for the operation of computer programs in non-volatile storage media. When the computer program is executed by the processor, the processor can execute any one of the zero-carbon factory intelligent operation and maintenance central control system methods.

[0158] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the terminal to which the solution of the present application is applied. The specific control device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0159] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0160] In one embodiment, the processor is configured to execute a computer program stored in the memory to implement the following steps:

[0161] A central control platform is set up in the pre-set zero-carbon factory. The central control platform includes a photovoltaic monitoring module and a power quality analysis module. The photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory to collect power generation data. The power quality analysis module includes a multi-parameter power quality detector on the zero-carbon factory's distribution bus to collect power quality parameters.

[0162] Build a factory energy digital twin model based on the photovoltaic module characteristics, load equipment parameters, and grid access point characteristics corresponding to the zero-carbon factory;

[0163] Obtain real-time meteorological data and historical power generation data corresponding to zero-carbon factories;

[0164] Input power generation data, power quality parameters, real-time meteorological data, and historical power generation data into the plant energy digital twin model to generate future power generation curves;

[0165] When the power quality analysis module detects that the total harmonic distortion rate exceeds the preset threshold, it matches the optimal control plan in the preset fault mode self-learning database based on the zero-carbon factory's power generation data and future power generation power curve; according to the optimal control plan, the three-phase imbalance of the zero-carbon factory's power supply network is adjusted to complete the power quality optimization of the zero-carbon factory; the optimal control plan includes the compensation capacitor group switching strategy of the zero-carbon factory.

[0166] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the processor described above can refer to the corresponding process in the method embodiments described in the above embodiments, and will not be repeated here.

[0167] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement the steps of the zero-carbon factory intelligent operation and maintenance power quality optimization method provided in the above embodiments of the present application.

[0168] The computer-readable storage medium may be an internal storage unit of the control device described in the aforementioned embodiment, such as a hard disk or memory of the control device. The computer-readable storage medium may also be an external storage device of the control device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the control device.

[0169] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for optimizing power quality in intelligent operation and maintenance of a zero-carbon factory, characterized in that: include: A central control platform is set up in the preset zero-carbon factory; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; The photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory area, which is used to collect power generation data; the power quality analysis module includes a multi-parameter power quality detector on the zero-carbon factory distribution bus, which is used to collect power quality parameters; Constructing a factory energy digital twin model based on the photovoltaic module characteristics, load equipment parameters, and grid access point characteristics corresponding to the zero-carbon factory; Obtaining real-time meteorological data and historical power generation data corresponding to the zero-carbon plant; The power generation data, power quality parameters, real-time meteorological data and historical power generation data are input into the plant energy digital twin model to generate a future power generation curve, including: normalizing and preprocessing the power generation data, power quality parameters, real-time meteorological data and historical power generation data to establish three-dimensional feature space information including irradiance, temperature and power; using a long short-term memory neural network to construct a time series prediction model corresponding to the plant energy digital twin model; the time series prediction model performs rolling prediction through a sliding time window mechanism; inputting the three-dimensional feature space information into the time series prediction model to output the future power generation curve; after inputting the three-dimensional feature space information into the time series prediction model and outputting the future power generation curve, it also includes: parsing real-time meteorological data to obtain cloud motion vectors; based on a particle filter algorithm, correcting the photovoltaic output prediction value of the future power generation curve according to the cloud motion vector; cloud motion vector extraction obtains the cloud movement speed and direction through meteorological radar data or an all-sky imager to construct A heat map of cloud shading distribution for the next 30 minutes was constructed. The particle filter correction partitioned the photovoltaic array into particle groups, adjusted the weight of each particle according to the cloud shading probability, performed a Monte Carlo simulation on the predicted power curve, and output the corrected confidence interval power value. The sky image was acquired through the all-sky imager, and the optical flow method was used to calculate the cloud movement speed and direction, generating a two-dimensional motion vector field with the center of the photovoltaic array as the origin. The cloud position for the next 15 minutes was extrapolated based on the two-dimensional motion vector field, and the photovoltaic array was divided into 10×10 grids. The probability of each grid unit being blocked by clouds was calculated. Each group of photovoltaic array strings was defined as a particle, and a total of 1000 particles were generated. The status of each particle included position, shading probability, and output power correction coefficient. The particle weight was adjusted according to the cloud motion vector. When the cloud moved westward, the weight of the western array particle increased. The residual between the actual power and the predicted power was used as the observation value, and the particles were redistributed using Monte Carlo sampling to generate the corrected power prediction value and its confidence interval. When the power quality analysis module detects that the total harmonic distortion rate exceeds a preset threshold, the optimal control plan is matched in the preset fault mode self-learning database according to the power generation data and future power generation power curve of the zero-carbon plant, including: obtaining harmonic spectrum feature vectors and transient process recording data according to the power generation data and future power generation power curve; using a dynamic time warping algorithm to calculate the waveform similarity between the harmonic spectrum feature vectors and transient process recording data and the fault mode self-learning database, so as to obtain multiple candidate control plans in the fault mode self-learning database; establishing a load demand response constraint condition according to the predicted slope change rate of the future power generation curve; performing multi-objective optimization on the candidate control plans according to the genetic algorithm based on the load demand response constraint condition to determine the optimal control plan among multiple candidate control plans; waveform similarity The transient recording data is dynamically time-warped using a degree calculation method, the morphological distance with the historical waveform in the database is calculated, and candidate treatment solutions with a similarity of more than 90% are screened; the multi-objective optimization takes the lowest treatment cost, the largest reduction in harmonic distortion rate, and the longest life of compensation equipment as the objective functions, and the NSGA-II genetic algorithm is used to perform a Pareto front search on the candidate solutions, and the optimal treatment solution is selected through fuzzy decision-making; the objective functions set include: treatment cost, THD reduction rate, and capacitor group life loss; the NSGA-II genetic algorithm is used: the population size is set to 200, the crossover probability is 0.8, and the mutation probability is 0.05; the Pareto front solution set is generated by non-dominated sorting; the comprehensive optimal solution is selected through the fuzzy membership function, and the weight distribution is that the weight corresponding to the treatment cost is 0.4, the weight corresponding to the THD reduction rate is 0.4, and the weight corresponding to the capacitor group life loss is 0.2; the three-phase imbalance of the zero-carbon plant power supply network is adjusted according to the optimal treatment solution to complete the power quality optimization of the zero-carbon plant; the optimal treatment solution includes the compensation capacitor group switching strategy of the zero-carbon plant.

2. The method according to claim 1, characterized in that The plant energy digital twin model is constructed according to the photovoltaic module characteristics, load equipment parameters and grid access point characteristics corresponding to the zero-carbon plant, including: Constructing a three-dimensional simulation model of a photovoltaic power generation unit of a zero-carbon factory based on the characteristics of the photovoltaic modules; Constructing a load dynamic model according to the load equipment parameters; constructing a power grid equivalent model according to the characteristics of the power grid access point; Based on a hybrid modeling method of mechanism model and data-driven model, the three-dimensional simulation model of the photovoltaic power generation unit, the load dynamic model and the grid equivalent model are multi-physical field coupled to form a digital twin model with real-time simulation capabilities.

3. The method according to claim 2, characterized in that The method of constructing a three-dimensional simulation model of a photovoltaic power generation unit of a zero-carbon factory according to the characteristics of the photovoltaic modules; constructing a load dynamic model according to the load device parameters; and constructing a grid equivalent model according to the characteristics of the grid access point includes: Analyzing the characteristics of the photovoltaic module to obtain the temperature coefficient and photoelectric conversion efficiency curve of the photovoltaic module for constructing a three-dimensional simulation model of the photovoltaic power generation unit; Analyze the load equipment parameters to obtain the load equipment power consumption characteristic curve and harmonic spectrum distribution of the zero-carbon factory, which are used to construct the load dynamic model; The short-circuit capacity and system impedance ratio of the grid access point are obtained according to the characteristics of the grid access point, and are used to construct the grid equivalent model.

4. The method according to claim 1, wherein The method of adjusting the three-phase imbalance of the zero-carbon factory power supply network according to the optimal control solution to optimize the power quality of the zero-carbon factory further includes: Controlling the static VAR generators in the zero-carbon power supply network for dynamic compensation; The grid-connected power factor of the energy storage system of the zero-carbon factory power supply network is adjusted to compensate for the residual imbalance of the preset SVG equipment; wherein, during the compensation process, the changes in the neutral line current of the zero-carbon factory power supply network are monitored in real time to dynamically optimize the switching strategy of the compensation capacitor group.

5. The method according to claim 1, wherein After the power quality optimization of the zero-carbon plant is completed, the method further includes: The photovoltaic power generation, reactive power compensation and equivalent carbon emission reduction data of the zero-carbon factory are generated into an encrypted electronic certificate with a timestamp; the electronic certificate complies with the certification standards of the carbon trading market.

6. A zero-carbon factory intelligent operation and maintenance central control system, characterized by: include: Central control platform; The central control platform includes a photovoltaic monitoring module and a power quality analysis module; the photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory area, which is used to collect power generation data; the power quality analysis module includes a multi-parameter power quality detector on the zero-carbon factory distribution bus, which is used to collect power quality parameters; A control device comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement the method according to any one of claims 1 to 5 when executing the computer program.

7. A zero-carbon factory intelligent operation and maintenance power quality optimization device, characterized in that: include: A platform setting unit, used to set up a central control platform in a preset zero-carbon factory; the central control platform includes a photovoltaic monitoring module and a power quality analysis module; The photovoltaic monitoring module includes a photovoltaic array sensor group deployed on the roof of the zero-carbon factory area, which is used to collect power generation data; the power quality analysis module includes a multi-parameter power quality detector on the zero-carbon factory distribution bus, which is used to collect power quality parameters; A model building unit, configured to build a factory energy digital twin model based on photovoltaic module characteristics, load device parameters, and grid access point characteristics corresponding to the zero-carbon factory; A data acquisition unit, configured to acquire real-time meteorological data and historical power generation data corresponding to the zero-carbon plant; A curve generation unit is used to input the power generation data, power quality parameters, real-time meteorological data and historical power generation data into the plant energy digital twin model to generate a future power generation curve, including: normalizing and preprocessing the power generation data, power quality parameters, real-time meteorological data and historical power generation data to establish three-dimensional feature space information including irradiance, temperature and power; using a long short-term memory neural network to construct a time series prediction model corresponding to the plant energy digital twin model; the time series prediction model performs rolling prediction through a sliding time window mechanism; inputting the three-dimensional feature space information into the time series prediction model to output the future power generation curve; after inputting the three-dimensional feature space information into the time series prediction model and outputting the future power generation curve, it also includes: parsing real-time meteorological data to obtain cloud motion vectors; correcting the photovoltaic output prediction value of the future power generation curve according to the cloud motion vector based on the particle filter algorithm; cloud motion vector extraction obtains cloud movement speed through meteorological radar data or all-sky imager The cloud shading distribution heat map for the next 30 minutes is constructed based on the cloud shading probability. The particle filter correction divides the photovoltaic array into particle groups, adjusts the weight of each particle according to the cloud shading probability, performs Monte Carlo simulation on the predicted power curve, and outputs the corrected confidence interval power value. The sky image is obtained by the all-sky imager, and the optical flow method is used to calculate the cloud movement speed and direction to generate a two-dimensional motion vector field with the center of the photovoltaic array as the origin. The cloud position in the next 15 minutes is extrapolated based on the two-dimensional motion vector field, and the photovoltaic array is divided into 10×10 grids to calculate the probability of each grid unit being blocked by clouds. Each group of photovoltaic arrays is defined as a particle, and a total of 1000 particles are generated. The state of each particle includes position, shading probability, and output power correction coefficient. The particle weight is adjusted according to the cloud motion vector. When the cloud moves westward, the particle weight of the western array increases. The residual between the actual power and the predicted power is used as the observation value, and the particles are redistributed by Monte Carlo sampling to generate the corrected power prediction value and its confidence interval. The optimization completion unit is used to match the best governance plan in the preset fault mode self-learning database according to the power generation data and future power generation power curve of the zero-carbon plant when the power quality analysis module detects that the total harmonic distortion rate exceeds the preset threshold, including: obtaining the harmonic spectrum feature vector and transient process recording data according to the power generation data and the future power generation power curve; using the dynamic time warping algorithm to calculate the waveform similarity between the harmonic spectrum feature vector and the transient process recording data and the fault mode self-learning database, so as to obtain multiple candidate governance plans in the fault mode self-learning database; establishing the load demand response constraint condition according to the predicted slope change rate of the future power generation curve; performing multi-objective optimization on the candidate governance plan according to the genetic algorithm based on the load demand response constraint condition, so as to obtain the best governance plan among the multiple candidate governance plans. Determine the optimal control solution; perform dynamic time warping of transient recording data using waveform similarity calculation, calculate the morphological distance with historical waveforms in the database, and screen candidate control solutions with a similarity greater than 90%; multi-objective optimization uses the lowest control cost, the largest reduction in harmonic distortion rate, and the longest compensation equipment life as the objective functions, and uses the NSGA-II genetic algorithm to perform a Pareto front search on candidate solutions, and select the optimal control solution through fuzzy decision making; set the objective functions to include: control cost, THD reduction rate, and capacitor bank life loss; use the NSGA-II genetic algorithm with a population size of 200, a crossover probability of 0.8, and a mutation probability of 0.05; generate a Pareto front solution set through non-dominated sorting; select the comprehensive optimal solution through a fuzzy membership function, and assign weights such that the control cost corresponds to a weight of 0.4, the THD reduction rate corresponds to a weight of 0.4, and the capacitor bank life loss corresponds to a weight of 0.2; adjust the three-phase imbalance of the zero-carbon plant power supply network according to the optimal control solution to optimize the power quality of the zero-carbon plant; the optimal control solution includes the compensation capacitor bank switching strategy of the zero-carbon plant.

Images