Power grid load dynamic balance and fluctuation suppression control system and method

Abstract

The invention discloses a power grid load dynamic balance and fluctuation suppression control system and method, and belongs to the technical field of power plant automatic control. The technical problems that a traditional power plant control system is low in response speed, incomplete in equipment protection and insufficient in-plant adjustable resource collaboration are solved. According to the technical principle, a three-level architecture of a cloud collaborative optimization layer, an edge collaborative control layer and a resource execution layer is constructed, wherein the cloud collaborative optimization layer generates an in-plant adjustment strategy by predicting the load and auxiliary engine power demand of the whole plant and adopting distributed robust optimization; the edge cooperative control layer decomposes a strategy into specific instructions for an energy storage system, a standby auxiliary machine and an interruptible load based on local state data; and the resource execution layer realizes rapid power adjustment through the advanced controller. According to the system, collaborative optimization and second-level regulation and control of in-plant resources are realized, power fluctuation is effectively suppressed, and the stability, safety and economical efficiency of operation of a power plant are remarkably improved through digital twin verification and a communication interruption autonomous mechanism.

Description

Technical Field

[0001] This invention belongs to the field of power plant automatic control technology, specifically a power grid load dynamic balance and fluctuation suppression control system and method. Background Technology

[0002] In the production and operation of traditional power plants such as thermal and hydropower plants, maintaining dynamic load balance and suppressing power fluctuations have always been key technical challenges to ensure the safe and stable operation of the units. With the increasing demands of the power grid on the peak-shaving and frequency regulation performance of power plants, and the increasing complexity of auxiliary systems within the plants, the stability control problems faced by power plants are becoming increasingly prominent.

[0003] Traditional power plants generally employ a hierarchical control architecture based on programmable logic controllers (PLCs), which has significant technical limitations. In terms of control response, existing systems lack sufficient coordinated control capabilities for auxiliary systems within the plant (such as feedwater pumps, circulating water pumps, and fans). When the unit load changes rapidly, uneven power distribution among the auxiliary systems leads to power fluctuations in the plant's power system, severely impacting the stable operation of the units. Particularly during grid frequency regulation, frequent changes in unit output commands exacerbate load fluctuations within the plant, and the response speed of traditional control systems struggles to meet the demands for second-level regulation.

[0004] In terms of equipment protection, existing technologies lack real-time capacity verification and protection for critical electrical equipment such as plant transformers, switchgear, and cables. When auxiliary systems start simultaneously or the load changes abruptly, it can easily lead to overload of the plant power system, causing protection devices to activate or even damaging the equipment. Although some power plants are equipped with power limiting functions, they mostly use fixed thresholds and cannot be dynamically adjusted according to the real-time status of the equipment, resulting in low equipment utilization or potential safety hazards.

[0005] Regarding resource utilization, power plants possess a large number of potential adjustable resources, such as energy storage systems, backup auxiliary equipment, and interruptible loads. However, the existing control system lacks unified coordination and management of these resources. Each resource system operates independently, failing to create a synergistic effect, resulting in the plant's internal regulation capacity not being fully utilized. Furthermore, the existing system's coordination and control capabilities over the plant's microgrid are insufficient, making it difficult to maintain frequency and voltage stability of the plant's auxiliary systems during islanded operation or external grid failures.

[0006] Therefore, there is an urgent need for a new type of control system that can achieve dynamic load balancing and fluctuation suppression within the plant, in order to solve the problems of slow response speed, imperfect equipment protection, and insufficient resource coordination in the existing technology, and improve the overall operational safety and economy of the power plant. Summary of the Invention

[0007] To address the above problems, this invention provides a power grid load dynamic balance and fluctuation suppression control system and method to solve the problems of slow response speed and insufficient resource coordination.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] A power grid load dynamic balancing and fluctuation suppression control system includes a cloud-based collaborative optimization layer, an edge-based collaborative control layer, and a resource execution layer; the cloud-based collaborative optimization layer and the edge-based collaborative control layer are connected through a first communication network; the edge-based collaborative control layer and the resource execution layer are connected through a second communication network.

[0010] It includes a cloud-based collaborative optimization layer, an edge-based collaborative control layer, and a resource execution layer; the cloud-based collaborative optimization layer and the edge-based collaborative control layer are connected through a first communication network; the edge-based collaborative control layer and the resource execution layer are connected through a second communication network.

[0011] The cloud-based collaborative optimization layer includes a prediction module, an optimization calculation module, and a strategy generation module. The prediction module is used to perform load prediction for the entire power plant and auxiliary equipment power demand prediction, with a prediction time range of 5 to 60 minutes. The optimization calculation module, connected to the prediction module, is used to perform distributed robust optimization calculations based on the prediction results of the prediction module to obtain optimization calculation results. The strategy generation module, connected to the optimization calculation module, is used to generate virtual inertial scheduling strategies based on the optimization calculation results.

[0012] The edge collaborative control layer includes a state perception module and an instruction decomposition module. The state perception module is used to collect plant power system data within the area, including frequency, voltage, and power data. The instruction decomposition module is connected to the state perception module and is used to perform optimization calculations based on local state data and plant regulation strategies to generate control instructions for virtual inertial resources.

[0013] The resource execution layer includes adjustable resources within the plant, including energy storage systems, backup auxiliary equipment, and interruptible loads. Each adjustable resource within the plant is equipped with a power electronic interface and an advanced controller. The advanced controller has a built-in control algorithm for adjusting the active power output according to the frequency changes of the plant power system.

[0014] Furthermore, the cloud-based collaborative optimization layer also includes a digital twin module, which includes an in-plant power system model library, an auxiliary machine system model library, and a resource model library. The digital twin module is data-connected to the optimization calculation module and is used to perform online simulation verification of the generated in-plant regulation strategy.

[0015] Furthermore, the edge collaborative control layer also includes a communication status monitoring module and an autonomous control module; the communication status monitoring module is used to monitor the communication connection status with the cloud collaborative optimization layer; when the autonomous control module detects a communication interruption, it maintains the operation of the plant's power system based on the last received effective plant regulation strategy and the real-time collected local status data.

[0016] Furthermore, the control algorithm is based on the rotor motion equations of the synchronous generator, which include: ;in For angular frequency deviation, The virtual inertial time constant. This is a reference value for mechanical power. Electromagnetic power, is the damping coefficient.

[0017] Furthermore, it also includes a blockchain evidence storage module, which comprises a resource registration unit, a contribution record unit, and a smart contract unit. The resource registration unit is connected to the contribution record unit, and the contribution record unit is connected to the smart contract unit. The resource registration unit is used to register the basic parameters and performance indicators of the adjustable resources within the plant. The contribution record unit is used to record the actual power support, response time, and duration provided by the adjustable resources within the plant. The smart contract unit is used to execute automatic economic compensation based on the recorded contribution data.

[0018] Furthermore, the state awareness module includes a data aggregation unit and a data quality assessment unit; the data aggregation unit is used to synchronize the time and unify the format of the collected plant power system data to obtain processed data; the data quality assessment unit is used to assess the credibility of the processed data, and triggers an anomaly handling mechanism when the data quality is lower than a preset threshold.

[0019] A method for dynamic load balancing and fluctuation suppression control in power grids includes the following steps:

[0020] S1: The cloud-based collaborative optimization layer performs plant-wide load forecasting and auxiliary equipment power demand forecasting through the forecasting module, generating forecasting results on a time scale of 5 to 60 minutes;

[0021] S2: The cloud-based collaborative optimization layer obtains the optimization results by performing distributed robust optimization calculations based on the prediction results through the optimization calculation module. The calculation process takes into account the uncertainty boundary of the prediction results. The strategy generation module generates in-plant adjustment strategies based on the optimization calculation results.

[0022] S3: The cloud-based collaborative optimization layer generates in-plant adjustment strategies through the strategy generation module and sends them to the edge collaborative control layer through the first communication network.

[0023] S4: The edge collaborative control layer collects plant power system data within the area through the state awareness module;

[0024] S5: The edge collaborative control layer performs rapid optimization calculations based on local status data and received in-plant regulation strategies through the instruction decomposition module, generating specific power instructions for energy storage systems, backup auxiliary equipment, and interruptible loads;

[0025] S6: The edge collaborative control layer sends power commands to the resource execution layer through the second communication network;

[0026] S7: The plant's adjustable resources in the resource execution layer autonomously adjust their active power output based on received power commands or through control algorithms.

[0027] Furthermore, in S1, the prediction adopts a spatiotemporal graph neural network model. The input of the model includes historical load data, weather forecast data, and date type, and the output is a probability distribution prediction of load and auxiliary power demand.

[0028] Furthermore, in S2, the objective function of the distributed robust optimization computation is: ,in This is the traditional power generation cost coefficient. For traditional power generation, This is the cost coefficient for adjustable resource allocation within the factory. Power for adjustable resource allocation within the factory. The fluctuation penalty coefficient, This represents the power fluctuation.

[0029] Furthermore, in S5, the rapid optimization calculation includes the following sub-steps:

[0030] S41: Transient stability verification: Simulate the power angle swing curve of the system after disturbance, and calculate the maximum power angle difference. S42: Voltage stability verification: The nose point of the PV curve is obtained through continuous power flow calculation. When the power margin between the operating point and the nose point is >15%, it is considered qualified; S43: Equipment capacity verification: Verify whether the power command in the strategy is met. ,in The maximum allowable power of the equipment. This is a power command.

[0031] The beneficial effects of this invention are as follows: This system decouples minute-level (5-60 minute) predictive optimization from second-level real-time control through a layered architecture of "cloud optimization" and "edge control." The cloud is responsible for forward-looking strategy generation, while the edge is responsible for instruction decomposition and execution based on local real-time data. This solves the problem of slow response speed in traditional control systems, which cannot meet the second-level frequency regulation requirements of the power grid, and improves the power plant's ability to respond quickly to load fluctuations.

[0032] By embedding a control algorithm based on the synchronous generator rotor motion equation into the advanced controller at the resource execution layer, power electronic interface devices such as energy storage and auxiliary equipment can simulate the inertial response and damping characteristics of a synchronous generator. When the frequency of the plant power system fluctuates, these resources can autonomously and quickly inject or absorb active power, providing instantaneous support to the system, effectively suppressing the rate of change of frequency (RoCoF) and amplitude, and enhancing the stability of the plant's microgrid. Attached Figure Description

[0033] Figure 1 This is a block diagram of a power grid load dynamic balance and fluctuation suppression control system;

[0034] Figure 2 This is a flowchart of a power grid load dynamic balance and fluctuation suppression control method. Detailed Implementation

[0035] To enable those skilled in the art to better understand the technical solution, the present invention will be described in detail below with reference to embodiments. The description in this part is only exemplary and explanatory, and should not be used to limit the scope of protection of the present invention in any way.

[0036] Example 1

[0037] This embodiment provides a basic architecture implementation method for a power grid load dynamic balance and fluctuation suppression control system.

[0038] See attached document Figure 1 The system includes a cloud-based collaborative optimization layer, an edge-based collaborative control layer, and a resource execution layer.

[0039] The cloud-based collaborative optimization layer and the edge-based collaborative control layer are connected through a first communication network. This first communication network is a low-bandwidth, non-real-time communication network with a communication latency greater than 1 second and a bandwidth less than 10 Mbps, implemented using fiber optic communication technology.

[0040] The edge collaboration control layer and the resource execution layer are connected through a second communication network. This second communication network is a high-bandwidth, low-latency communication network with a communication latency of less than 100 milliseconds and a bandwidth of more than 100 Mbps, implemented using 5G communication technology.

[0041] The cloud-based collaborative optimization layer comprises a forecasting module, an optimization calculation module, and a strategy generation module. The forecasting module performs plant-wide load forecasting and auxiliary equipment power demand forecasting, with a forecasting time range of 5 to 60 minutes. The optimization calculation module, connected to the forecasting module, performs distributed robust optimization calculations based on the forecasting results. The strategy generation module, connected to the optimization calculation module, generates plant-wide regulation strategies.

[0042] The edge collaborative control layer includes a state awareness module and a command decomposition module. The state awareness module collects data on the plant's power system within the area, including frequency, voltage, and power data. The command decomposition module, connected to the state awareness module, performs rapid optimization calculations based on local state data and received plant regulation strategies to generate control commands for adjustable resources within the plant.

[0043] The resource execution layer comprises multiple plant-level adjustable resources, including energy storage systems, backup auxiliary equipment, and interruptible loads. Each plant-level adjustable resource is equipped with a power electronic interface and an advanced controller. The advanced controller has a built-in control algorithm used to adjust active power output according to frequency changes in the plant power system.

[0044] The edge collaborative control layer also includes a communication status monitoring module and an autonomous control module.

[0045] The communication status monitoring module is used to monitor the communication connection status with the cloud-based collaborative optimization layer in real time. The module periodically calculates the average packet loss rate and maximum latency over the past 30 seconds. A communication interruption is declared when the average packet loss rate exceeds 5% and the maximum latency exceeds 30 seconds.

[0046] When a communication interruption is detected, the autonomous control module maintains the operation of the plant's power system based on the last received valid plant regulation strategy and real-time collected local status data. The autonomous control module's operating logic includes: recording the last received valid plant regulation strategy before the communication interruption; performing local optimization calculations at 1-second control cycles based on real-time data from the status awareness module; activating a preset emergency control strategy when a severe frequency anomaly is detected, i.e., the absolute value of the frequency deviation is greater than 0.5 Hz; continuously monitoring the communication status, and performing data synchronization and mode switching after communication is restored.

[0047] The status awareness module includes a data aggregation unit and a data quality assessment unit. The data aggregation unit performs time synchronization and format standardization processing on the collected plant power system data. The data quality assessment unit evaluates the reliability of the processed data, specifically by calculating a comprehensive quality score based on data integrity and accuracy scores. When the comprehensive quality score falls below a preset threshold of 0.8, an anomaly handling mechanism is triggered.

[0048] Example 2

[0049] See attached document Figure 2 A method for dynamic load balancing and fluctuation suppression control in power grids includes the following steps:

[0050] S1: The cloud-based collaborative optimization layer performs plant-wide load forecasting and auxiliary equipment power demand forecasting through the forecasting module, generating forecasting results on a time scale of 5 to 60 minutes;

[0051] S2: The cloud-based collaborative optimization layer performs distributed robust optimization calculations based on the prediction results through the optimization calculation module. The calculation process takes into account the uncertainty boundary of the prediction results.

[0052] S3: The cloud-based collaborative optimization layer generates in-plant adjustment strategies through the strategy generation module and sends them to the edge collaborative control layer through the first communication network.

[0053] S4: The edge collaborative control layer collects plant power system data within the area through the state awareness module;

[0054] S5: The edge collaborative control layer performs rapid optimization calculations based on local status data and received in-plant regulation strategies through the instruction decomposition module, generating specific power instructions for energy storage systems, backup auxiliary equipment, and interruptible loads;

[0055] S6: The edge collaborative control layer sends power commands to the resource execution layer through the second communication network;

[0056] S7: The plant's adjustable resources in the resource execution layer autonomously adjust their active power output based on received power commands or through control algorithms.

[0057] Step S1 includes at least steps S110 to S130:

[0058] S110. Acquire multi-source power plant operation data, perform sampling alignment and data cleaning processing to obtain a standardized data sequence; the acquired multi-source power plant operation data includes total plant load data, auxiliary equipment power demand data, plant power system frequency data, voltage data, plant transformer monitoring data, key auxiliary equipment operation data, energy storage system status data, standby auxiliary equipment status data, interruptible load operation data, and environmental meteorological data. Specifically, a data acquisition link is established between the cloud-based collaborative optimization layer and the edge-based collaborative control layer. This data acquisition link simultaneously covers the total plant load forecasting module, auxiliary equipment power monitoring module, plant power system frequency measurement unit, voltage measurement unit, plant transformer monitoring unit, key auxiliary equipment status monitoring unit, energy storage system status monitoring unit, standby auxiliary equipment status monitoring unit, interruptible load control unit, and meteorological monitoring station. Specifically, measurement units are configured at key nodes of the plant power system, including measurement points MU1 to MU6, to generate time-series curves of frequency and voltage for the plant power system; power monitoring modules are configured in the main auxiliary equipment circuits to generate time-series curves of auxiliary equipment power demand; status monitoring units are configured on the energy storage system side to collect the state of charge and power output capacity of the energy storage system; status monitoring units are configured on the standby auxiliary equipment side to record the start-up and shutdown status and dispatchable capacity of the standby auxiliary equipment; control units are configured on the interruptible load side to collect the load operating status and adjustable range; and meteorological monitoring stations are configured on the environmental side to collect data on temperature, humidity, wind speed, and light intensity.

[0059] After acquisition, the multi-source power plant operation data synchronously enters the sampling alignment and data cleaning processing flow. Specifically, the timestamps of the data from each channel are first uniformly processed, and a master clock is established according to the power plant's unified time synchronization system. Early and late samples are incorporated into the master clock through linear interpolation and repetition imputation. Missing samples are marked with missing tags and filled or corrected by sliding window averaging and outlier removal without changing the data trend. Then, continuous quantities such as plant load data, auxiliary equipment power demand data, frequency data, and voltage data are subjected to low-pass filtering and noise suppression processing. Discrete quantities such as energy storage system status, standby auxiliary equipment status, and interruptible load operation status are subjected to status consistency verification and abnormal status correction processing. The measurement unit data of the plant power system is segmented, and frequency fluctuation segments and voltage fluctuation segments are marked for direct reference in subsequent quality assessment. After the above alignment and cleaning processes are completed, the dimensions and ranges are further standardized: continuous quantities such as plant load data, auxiliary machine power demand data, frequency data, and voltage data are uniformly converted into standard engineering dimensions, and discrete quantities such as energy storage system status, standby auxiliary machine status, and interruptible load operation status are uniformly encoded into consistent status flags, and the valid value set of the status flags is registered.

[0060] A standardized data sequence is formed, which includes the plant-wide load time series, auxiliary equipment power demand time series, frequency time series, voltage time series, plant transformer monitoring data series, key auxiliary equipment operation data series, energy storage system status series, standby auxiliary equipment status series, interruptible load operation status series, and environmental meteorological data series, all processed with a unified master clock and unified dimensions. This standardized data sequence is uniformly stored in a real-time database and includes a time window index to ensure seamless referencing in subsequent windowed processing steps.

[0061] S120. Extract power plant operation feature elements from the standardized data sequence, perform feature extraction processing to obtain a feature element sequence; read the standardized data sequence, perform segmentation processing according to a fixed-length time window, and extract feature elements related to power plant operation within each time window. Specifically, the system extracts load change rate, load fluctuation amplitude, and load prediction deviation from the plant's overall load time series, and binds these quantities to time window markers to form characteristic elements corresponding to load characteristics; it extracts demand volatility, demand predictability, and demand change trend characteristics from the auxiliary equipment power demand time series, and records whether the demand is within the preset range; it extracts frequency deviation, frequency change rate, and frequency stability indicators from the frequency time series based on the spatial arrangement relationship of MU1 to MU6, and divides the voltage time series into normal and abnormal segments at the markers, recording the voltage statistics within each segment and the change characteristics between segments; it extracts the state of charge change rate, chargeable and dischargeable power, and health status indicators from the energy storage system state series; it extracts dispatchable capacity, start-up time, and continuous operating time from the standby auxiliary equipment state series; it extracts adjustable range, adjustment rate, and priority indicators from the interruptible load operation state series; and it performs correlation analysis on environmental meteorological data to extract characteristic elements related to the plant's overall load and auxiliary equipment power demand.

[0062] Preferably, at the level related to power plant balance, the prediction error, fluctuation amplitude, and correlation strength are calculated for the overall plant load and auxiliary equipment power demand, respectively, to form the operating characteristic elements related to power plant balance; for frequency and voltage stability, the duration of frequency deviation, the number of voltage overruns, and the stability margin are statistically analyzed, and these characteristic elements are aligned and registered with the overall plant load characteristics and auxiliary equipment power demand characteristics in the corresponding time period; for the dispatchability of energy storage systems, standby auxiliary equipment, and interruptible loads, the available capacity, response speed, and adjustment accuracy within the time window are statistically analyzed to support the threshold comparison of resource scheduling in subsequent optimization calculations.

[0063] Once all the aforementioned feature elements are generated within each time window, they are arranged in chronological order to form a feature element sequence. This feature element sequence includes plant-wide load feature elements, auxiliary equipment power demand feature elements, frequency feature elements, voltage feature elements, energy storage system feature elements, standby auxiliary equipment feature elements, interruptible load feature elements, and environmental meteorological feature elements. Time window indexes and equipment type indexes are established in the data structure to ensure that elements between different equipment and different features within the same time window can be accurately associated and invoked by downstream steps.

[0064] S130. Perform data quality assessment processing on the feature element sequence to generate quality scoring indicators; read the feature element sequence and establish quality assessment channels according to data type and equipment type, including a prediction quality assessment channel for the overall plant load and auxiliary power demand, a measurement quality assessment channel for frequency and voltage, and a status quality assessment channel for energy storage system, standby auxiliary equipment and interruptible load. Specifically, within the prediction quality assessment channel, based on factors such as the overall plant load prediction deviation and auxiliary equipment power demand prediction deviation within each time window, combined with the overall plant load fluctuation amplitude and auxiliary equipment power demand fluctuation within that time window, a consistency judgment on prediction accuracy is established and accumulated in chronological order. Within the measurement quality assessment channel, based on the duration of frequency deviation, the number of voltage overruns, and stability margin, combined with the frequency change rate and voltage statistics within that segment, a judgment on the reliability of measurement data is established and accumulated in segment sequence. Within the condition quality assessment channel, based on the energy storage system health status indicators, standby auxiliary equipment start-up time, and the statistical values ​​of interruptible load regulation accuracy within the time window, combined with the changes in available capacity and regulation rate, a judgment on the accuracy of condition data is established and accumulated in chronological order.

[0065] To ensure the comparability of quality assessment processing across different data types and devices, the dimensional differences of feature elements are normalized in the processing of the three channels mentioned above, and a unified score range is established. For elements related to data reliability, such as the long-term trend of prediction deviation, the noise level of measurement data, and the update frequency of status data, a sliding window tracker and a steady-state benchmark are set for time series data to ensure that changes in data quality are reflected within the statistical scope. For elements related to transient anomalies, such as transient frequency exceedances, transient voltage fluctuations, and transient state jumps, short-term windows are set to capture transient anomalies, and the results are merged and registered with the results of the long-term trend tracker within the same score range. After the above merging, a quality scoring index is generated, which consists of prediction quality score, measurement quality score, and status quality score. The data structure retains data type indexes and device type indexes to support differentiated use of different data in subsequent steps.

[0066] After the quality assessment process is completed, the quality scoring indicators are bound to the key identifiers in the feature element sequence and registered in the quality assessment database. To ensure seamless integration across steps, explicit references are established during registration: First, the quality scoring index is provided to the overall plant load and auxiliary power demand forecast, enabling the forecast to read scores within the corresponding time window for adjusting the configuration of forecast model parameters and forecast time scales. Second, the quality scoring index is provided to distributed robust optimization computation, allowing it to configure optimization constraints based on forecast quality scores and measurement quality scores when setting uncertainty boundaries. Third, the quality scoring index is provided to the edge collaborative control layer, enabling it to reference the time series of state quality scores during rapid optimization computation. Fourth, the quality scoring index is associated with the time window index of the standardized data sequence, facilitating consistent referencing of data quality-related windows during digital twin verification. Fifth, the quality scoring index is associated with the equipment type index of the feature element sequence, enabling the use of energy storage system feature elements, standby auxiliary equipment feature elements, and interruptible load feature elements as context constraints during resource execution layer control.

[0067] S2 includes at least steps S210 to S230:

[0068] S210. Construct a distributed robust optimization model and set the objective function and constraints.

[0069] In the optimization computation module of the cloud-based collaborative optimization layer, an optimization model based on probability distribution is established. The objective function of the distributed robust optimization computation is: ,in This is the traditional power generation cost coefficient, with a value ranging from 0.1 to 0.5 yuan / kWh; For traditional power generation, This is the cost coefficient for adjustable resource allocation within the plant, with a value ranging from 0.05 to 0.2 yuan / kWh; Power for adjustable resource allocation within the factory. This is the fluctuation penalty coefficient, with a value ranging from 0.8 to 1.5. The power fluctuation is calculated from the power difference between the preceding and following time steps.

[0070] The constraints include: power balance constraints: Traditional generator set output upper and lower limit constraints: Constraints on the plant's adjustable resource capacity: System ramp rate constraint: .

[0071] S220. Solve the optimization model using a stochastic optimization algorithm to generate a preliminary scheduling strategy. Use a scenario-based stochastic programming method to generate 1000 - 5000 scenarios of auxiliary power demand through Monte Carlo simulation. Specifically, the probability weight of each scenario is determined by the probability distribution output by the spatio-temporal graph neural network model in S1. During the solution process, use the Benders decomposition algorithm to decompose the main problem into multiple sub-problems for parallel solution, set the number of iterations to 50 - 200 times, and set the convergence tolerance to 0.001 - 0.01.

[0072] During the solution process, monitor the optimization gap in real time. When the relative gap is less than the convergence tolerance, it is determined to converge, and the preliminary scheduling strategy is output. The preliminary scheduling strategy includes a traditional power generation plan curve and an in-plant adjustable resource scheduling curve, and the time resolution is set to 1 minute - 5 minutes.

[0073] S230. Conduct a robustness test and correction on the preliminary scheduling strategy to generate the final in-plant regulation strategy. Construct an extreme scenario set (such as critical auxiliary equipment tripping, sudden load increase) in the digital twin module for multi-scenario verification tests. Calculate the frequency response of the in-plant power system under each scenario through time-domain simulation, and require the frequency deviation not to exceed ±0.5 Hz. When the strategy does not meet the requirements in some scenarios, perform strategy correction: increase the standby capacity of in-plant adjustable resources by 5% - 15%, and re-optimize the calculation. After iterative correction, generate the final in-plant regulation strategy including the benchmark plan and the correction plan.

[0074] Step S5 at least includes steps S510 - S530, that is, perform a quick online verification on the issued strategy:

[0075] S510. Transient stability verification: Simulate and calculate the power angle swing curve of the system after the disturbance. When the maximum power angle difference δmax < 90°, it is determined to be qualified. In the instruction decomposition module, establish an equivalent model of the in-plant power system and set typical disturbances (such as in-plant line N - 1 fault, large auxiliary equipment tripping). Simulate and calculate the power angle swing curve of the system after the disturbance through numerical integration method, and set the sampling time interval to 0.01 - 0.05 seconds. Calculate the transient stability margin using the equal area criterion. When the maximum power angle difference δ_max < 90° and the stability margin is greater than 15%, it is determined to be qualified. If not qualified, adjust the power distribution ratio of in-plant adjustable resources and preferentially increase the output of the energy storage system.

[0076] S520. Voltage Stability Verification: The nose point of the PV curve is obtained through continuous power flow calculation. A pass is deemed achieved when the power margin between the operating point and the nose point is > 15%. Using the power distribution in the plant's regulation strategy as the initial operating point, the PV curve is plotted through continuous power flow calculation to determine the nose point voltage and the corresponding power margin. Voltage stability is deemed passable when the power margin between the operating point and the nose point is > 15% and all node voltages are within ±5% of the rated voltage. If not passable, reactive power compensation adjustment is initiated to optimize the reactive power output of adjustable resources within the plant.

[0077] S530, Equipment Capacity Verification: Verify whether the power command in the verification strategy is satisfied. ,in This refers to the maximum allowable power of the equipment.

[0078] Perform device-level security checks on each power command in the strategy.

[0079] When all power commands pass verification, the policy is marked as executable; when an out-of-limit command is detected, command pruning is performed, and the power of the out-of-limit portion is redistributed to other available resources according to priority.

[0080] Example 3

[0081] This embodiment is based on Embodiment 2, providing procedures under communication interruption and emergency frequency fluctuation conditions. The procedure under communication interruption conditions may include at least steps S8-S10:

[0082] S8. The edge collaboration control layer monitors the communication connection status and triggers the autonomous operation mode when communication is interrupted.

[0083] S9. Based on the last received plant regulation strategy and local status data, perform local optimization calculations to generate autonomous power commands;

[0084] S10: The resource execution layer adjusts the active power output according to the autonomous power command to maintain the stable operation of the plant's power system.

[0085] Step S8 includes at least steps S810 to S830:

[0086] S810, the edge collaborative control layer monitors the communication connection status with the cloud collaborative optimization layer in real time and calculates communication quality parameters;

[0087] A communication status monitoring module is configured at the edge collaborative control layer. This module periodically collects communication channel indicators, including packet loss rate, latency, bandwidth utilization, and connection status flags. Specifically, the communication status monitoring module calculates the average packet loss rate, maximum latency, minimum latency, and bandwidth utilization within the most recent 30-second time window every 5 seconds, while simultaneously monitoring changes in the connection status flags of the communication link. Specifically, a heartbeat detection mechanism is established between the edge node and the cloud, periodically sending heartbeat packets to detect communication link connectivity; data packet transmission statistics are collected at the network interface layer, including the number of packets sent, received, retransmitted, and timed-out packets; and data transmission latency, specifically the round-trip time from sending a request to receiving a response, is recorded at the application layer.

[0088] The communication quality parameters are collected and then processed in a status determination process. Specifically, each indicator is first compared against thresholds: when the average packet loss rate exceeds 5%–10% and the maximum latency exceeds 30–60 seconds, communication quality is considered degraded; when the connection status flag is continuously lost for more than three heartbeat cycles, communication connection is considered interrupted. Then, the communication quality degradation and interruption events are time-stamped and recorded to generate a communication status event sequence. After the above determination process is completed, the status is further classified: the communication status is divided into normal status, degraded status, and interrupted status, and a status code and priority are assigned to each status.

[0089] Communication quality parameters are generated, including average packet loss rate, maximum latency, bandwidth utilization, connection status flags, and a sequence of communication status events. These communication quality parameters are uniformly stored in a local cache with a timestamp index to ensure time consistency in subsequent steps.

[0090] S820: Determines the communication status based on communication quality parameters and generates a status switching command;

[0091] The communication quality parameters are read, and logical judgments are made according to predefined state transition rules. Specifically, when the communication state is normal and the average packet loss rate is below 5% and the maximum latency is below 30 seconds, the current state is maintained; when the communication state is normal but the average packet loss rate is between 5% and 10% or the maximum latency is between 30 and 60 seconds, it is converted to a degraded state; when the communication state is degraded and the average packet loss rate exceeds 10% or the maximum latency exceeds 60 seconds or the connection status flag is lost for more than 3 heartbeat cycles, it is converted to an interrupted state; when the communication state is interrupted and the average packet loss rate recovers to below 2% and the maximum latency recovers to below 10 seconds and remains below 10 seconds for 60 seconds, it is converted to a normal state.

[0092] Preferably, during the state determination process, a corresponding state switching instruction is generated for each state transition. Specifically, when transitioning from a normal state to a degraded state, a warning instruction is generated; when transitioning from a degraded state to an interrupted state, a switch to autonomous mode instruction is generated; and when transitioning from an interrupted state to a normal state, a restore to normal mode instruction is generated. The state switching instruction includes a target state code, a switching timestamp, and a priority flag.

[0093] Once the state determination is complete, the state transition command is sent to the autonomous control module of the edge collaborative control layer. To ensure timely execution of the command, the state transition command is transmitted through a high-priority interrupt mechanism and recorded in the command log.

[0094] S830: Trigger the autonomous operation mode according to the state switching command and initialize the autonomous control parameters; receive the state switching command, and when the command is to switch to autonomous mode, start the autonomous operation mode initialization process. Specifically, read the last received valid plant regulation strategy from the local cache, extract the key parameters in the strategy, including power setpoint, time range, and resource allocation weight; read the plant load data, auxiliary machine power demand data, frequency data, voltage data, energy storage system status data, standby auxiliary machine status data, and interruptible load operation data within the current time window from the local status database; calculate the initial parameters required for autonomous control based on the above data, including local power demand, available resource capacity, and regulation priority.

[0095] During initialization, the operating parameters of the autonomous control algorithm are configured: the control cycle is set to 1 to 5 seconds, the optimization objective is to minimize frequency and voltage deviations, and the constraints include equipment capacity limitations and regulation rate limitations. Simultaneously, a time base for autonomous control is established to ensure time synchronization and data consistency.

[0096] After initialization, autonomous control parameters are generated, including power demand sequence, resource capacity matrix, regulation priority list, and control cycle setting. These autonomous control parameters are stored in the runtime memory of the autonomous control module, and the autonomous control loop is started.

[0097] The procedure for emergency frequency fluctuation conditions shall include at least steps S11 to S13:

[0098] S11. Monitor the frequency of the plant power system in real time and calculate the frequency change rate and frequency deviation;

[0099] S12. When the frequency is abnormal, the power adjustment amount is calculated through the control algorithm;

[0100] S13. Quickly perform power adjustment, suppress frequency fluctuations, and record event data.

[0101] Step S11 includes at least steps S1110 to S1130:

[0102] S1110. Configure a frequency monitoring unit at the resource execution layer to collect real-time frequency data of the plant power system. Configure a high-precision frequency monitoring unit at each adjustable resource node within the resource execution layer. This unit is based on synchronous phasor measurement technology or zero-crossing detection technology. Specifically, the sampling frequency of the frequency monitoring unit is set to 100Hz~1000Hz, the measurement accuracy reaches 0.001Hz~0.01Hz, and the response time is no more than 10 milliseconds. Specifically, integrate frequency measurement circuits at the inverter interface, backup auxiliary power interface, and interruptible load control interface of the energy storage system to calculate the instantaneous frequency of the plant power system through voltage sampling and phase tracking. Configure backup frequency monitoring units at key nodes to form a redundant measurement system.

[0103] The frequency data of the plant power system enters the preprocessing process after acquisition. Specifically, the raw frequency data is first filtered by using a digital low-pass filter to suppress high-frequency noise, with the cutoff frequency set to 10Hz~50Hz; then, outlier detection is performed on the filtered data, using a sliding window standard deviation method to identify and remove outliers; finally, the data is timestamped to ensure the time consistency of data at each node.

[0104] A real-time frequency data sequence is generated, which includes a timestamp, frequency value, and data quality flag. The real-time frequency data sequence is transmitted to the local controller via a high-speed communication link, with a sampling period of 10 milliseconds to 100 milliseconds.

[0105] S1120. Calculate the rate of frequency change and frequency deviation based on real-time frequency data sequences;

[0106] The real-time frequency data sequence is read, and the dynamic frequency parameters are calculated according to a fixed time window. Specifically, the instantaneous frequency change rate is calculated using the central difference method, with the time window width set to 100 milliseconds to 500 milliseconds; when calculating the frequency deviation, the difference between the instantaneous frequency and the reference value is calculated using the rated frequency of 50Hz or 60Hz as a reference.

[0107] Preferably, the calculated frequency change rate and frequency deviation are smoothed using a moving average or exponential smoothing method to reduce fluctuations, with the smoothing window width set to 1 to 5 seconds. Simultaneously, limit detection is performed on the frequency change rate and frequency deviation; when the absolute value of the frequency change rate exceeds 0.1 Hz / s to 0.5 Hz / s or the absolute value of the frequency deviation exceeds 0.2 Hz to 0.5 Hz, it is marked as a frequency anomaly event.

[0108] Generate frequency dynamic parameters, including a frequency change rate sequence, a frequency deviation sequence, and anomaly event flags. The frequency dynamic parameters are updated every 100 milliseconds and stored in a circular buffer for use in subsequent steps.

[0109] S1130. Determine anomalies in frequency dynamic parameters and generate frequency anomaly alarms.

[0110] The frequency dynamic parameters are read, and anomaly status is determined based on predefined thresholds. Specifically, the frequency change rate threshold is set to 0.1Hz / s~0.5Hz / s, and the frequency deviation threshold is set to 0.2Hz~0.5Hz. When the frequency change rate or frequency deviation exceeds the threshold for 3 to 5 consecutive times, it is determined to be a continuous anomaly. When the frequency change rate or frequency deviation instantaneously exceeds the threshold by 2 to 3 times, it is determined to be an emergency anomaly.

[0111] During the anomaly detection process, anomalies are classified according to their severity: Level 1 anomalies are minor limit violations, Level 2 anomalies are persistent limit violations, and Level 3 anomalies are urgent limit violations. Corresponding alarm information is generated for each level of anomaly, including the anomaly level, occurrence time, duration, and limit violation magnitude.

[0112] After generating a frequency anomaly alarm, it is sent to the edge collaborative control layer and resource execution layer via a high-speed communication interface. Alarm information is prioritized, with urgent anomaly alarms transmitted first, and the transmission delay is no more than 50 milliseconds.

[0113] The specific working principle of this invention is as follows: The system first collects multi-source power plant operation data through a sensor network (including plant power system measurement units, key auxiliary equipment status monitoring devices, etc.) deployed at key nodes within the power plant. Based on a spatiotemporal graph neural network model, it integrates historical data, meteorological information, and quality scores to generate a probability distribution prediction of the total plant load and auxiliary equipment power demand on a 5-60 minute timescale. The model captures spatiotemporal correlations and provides prediction results with confidence levels. A distributed robust optimization algorithm is adopted, considering the prediction uncertainty boundary, to solve for the plant regulation strategy with the objective of minimizing the total system cost (including traditional power generation costs, costs of utilizing adjustable resources within the plant, and fluctuation penalty costs). The generated scheduling strategy is verified through multi-scenario simulation using a digital twin system, including transient stability verification, voltage stability assessment, and equipment capacity verification, to ensure the safety and feasibility of the strategy. The system receives the plant regulation strategy from the cloud and combines it with local real-time status data (frequency, voltage, available capacity of adjustable resources within the plant, etc.), and decomposes the macro strategy into power commands for specific resources (energy storage system, backup auxiliary equipment, interruptible loads) through rapid optimization calculations.

[0114] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the present invention. These examples are merely for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or variations without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, variations, or combinations, or the direct application of the concept and technical solution of the present invention to other situations without modification, should all be considered within the scope of protection of the present invention.

Claims

1. A power grid load dynamic balance and fluctuation suppression control system, characterized in that, It includes a cloud-based collaborative optimization layer, an edge-based collaborative control layer, and a resource execution layer; the cloud-based collaborative optimization layer and the edge-based collaborative control layer are connected through a first communication network; The edge collaboration control layer and the resource execution layer are connected via a second communication network; The cloud-based collaborative optimization layer includes a prediction module, an optimization calculation module, and a strategy generation module. The prediction module is used to perform load prediction for the entire power plant and auxiliary equipment power demand prediction, with a prediction time range of 5 to 60 minutes. The optimization calculation module, connected to the prediction module, is used to perform distributed robust optimization calculations based on the prediction results of the prediction module to obtain optimization calculation results. The strategy generation module, connected to the optimization calculation module, is used to generate virtual inertial scheduling strategies based on the optimization calculation results. The edge collaborative control layer includes a state perception module and an instruction decomposition module. The state perception module is used to collect plant power system data within the area, including frequency, voltage, and power data. The instruction decomposition module is connected to the state perception module and is used to perform optimization calculations based on local state data and plant regulation strategies to generate control instructions for virtual inertial resources. The resource execution layer includes adjustable resources within the plant, including energy storage systems, backup auxiliary equipment, and interruptible loads. Each adjustable resource within the plant is equipped with a power electronic interface and an advanced controller. The advanced controller has a built-in control algorithm for adjusting the active power output according to the frequency changes of the plant power system.

2. The power grid load dynamic balance and fluctuation suppression control system according to claim 1, characterized in that, The cloud-based collaborative optimization layer also includes a digital twin module, which includes an in-plant power system model library, an auxiliary machine system model library, and a resource model library. The digital twin module is data-connected to the optimization calculation module and is used to perform online simulation verification of the generated in-plant regulation strategies.

3. The power grid load dynamic balance and fluctuation suppression control system according to claim 1, characterized in that, The edge collaborative control layer also includes a communication status monitoring module and an autonomous control module; the communication status monitoring module is used to monitor the communication connection status with the cloud collaborative optimization layer; when the autonomous control module detects a communication interruption, it uses the last received effective in-plant adjustment strategy and real-time collected local status data.

4. The power grid load dynamic balance and fluctuation suppression control system according to claim 1, characterized in that, The control algorithm is based on the rotor motion equations of a synchronous generator, which include: ;in For angular frequency deviation, The virtual inertial time constant. This is a reference value for mechanical power. Electromagnetic power, is the damping coefficient.

5. The power grid load dynamic balance and fluctuation suppression control system according to claim 1, characterized in that, It also includes a blockchain evidence storage module, which comprises a resource registration unit, a contribution record unit, and a smart contract unit; the resource registration unit is connected to the contribution record unit, and the contribution record unit is connected to the smart contract unit; the resource registration unit is used to register the basic parameters and performance indicators of the adjustable resources within the plant; the contribution record unit is used to record the actual power support, response time, and duration provided by the adjustable resources within the plant; The smart contract unit is used to execute automatic economic compensation based on the recorded contribution data.

6. The power grid load dynamic balance and fluctuation suppression control system according to claim 1, characterized in that, The status awareness module includes a data aggregation unit and a data quality assessment unit. The data aggregation unit is used to synchronize the time and unify the format of the collected plant power system data to obtain processed data. The data quality assessment unit is used to assess the reliability of the processed data and trigger an anomaly handling mechanism when the data quality is lower than a preset threshold.

7. A method for dynamic load balancing and fluctuation suppression control in a power grid, characterized in that, Includes the following steps: S1: The cloud-based collaborative optimization layer performs plant-wide load forecasting and auxiliary equipment power demand forecasting through the forecasting module, generating forecasting results on a time scale of 5 to 60 minutes; S2: The cloud-based collaborative optimization layer obtains the optimization results by performing distributed robust optimization calculations based on the prediction results through the optimization calculation module. The calculation process takes into account the uncertainty boundary of the prediction results. The strategy generation module generates in-plant adjustment strategies based on the optimization calculation results. S3: The cloud-based collaborative optimization layer generates in-plant adjustment strategies through the strategy generation module and sends them to the edge collaborative control layer through the first communication network. S4: The edge collaborative control layer collects plant power system data within the area through the state awareness module; S5: The edge collaborative control layer performs rapid optimization calculations based on local status data and received in-plant regulation strategies through the instruction decomposition module, generating specific power instructions for energy storage systems, backup auxiliary equipment, and interruptible loads; S6: The edge collaborative control layer sends power commands to the resource execution layer through the second communication network; S7: The plant's adjustable resources in the resource execution layer autonomously adjust their active power output based on received power commands or through control algorithms.

8. The method for dynamic load balancing and fluctuation suppression control of a power grid according to claim 7, characterized in that, In S1, the prediction uses a spatiotemporal graph neural network model. The input of the model includes historical load data, weather forecast data, and date type, and the output is a probability distribution prediction of load and auxiliary power demand.

9. The method for dynamic load balancing and fluctuation suppression control of a power grid according to claim 7, characterized in that, In S2, the objective function of the distributed robust optimization computation is: ,in This is the traditional power generation cost coefficient. For traditional power generation, This is the cost coefficient for adjustable resource allocation within the factory. Power for adjustable resource allocation within the factory. The fluctuation penalty coefficient, This represents the power fluctuation.

10. The method for dynamic load balancing and fluctuation suppression control of a power grid according to claim 7, characterized in that, In S5, the fast optimization calculation includes the following sub-steps: S41: Transient stability verification: Simulate the power angle swing curve of the system after disturbance, and calculate the maximum power angle difference. It was deemed qualified at that time; S42: Voltage stability verification: The nose point of the PV curve is obtained through continuous power flow calculation. When the power margin between the operating point and the nose point is >15%, it is considered qualified. S43: Equipment Capacity Verification: Verify whether the power command in the verification strategy is satisfied. ,in The maximum allowable power of the equipment. This is a power command; if satisfied... If the requirement is met, the instruction passes verification at the device capacity level; otherwise, it fails verification. If the command fails to pass verification, it will be flagged as a device overload risk.

Images