A method for optimizing the start and stop of units in real-time response to frequency regulation market demand

Abstract

The present application provides a method for optimizing the start and stop of units in real-time response to frequency modulation market demand, including: collecting market demand, load power data, and electricity price data in real time through mobile applications, sensors, and smart meters, establishing a demand forecasting model and predicting future trends; training multiple optimization models, predicting electricity cost and benefit indicators according to different electricity consumption scenarios, and performing multi-objective optimization; obtaining unit health status monitoring data, and evaluating the best start and stop scheme under the multi-objective constraints of power grid response, market participation, and reliability assurance; using the start and stop units and friendly power grids as physical execution units, and building a closed-loop test platform based on multi-source system simulation in a digital twin environment; and designing a fast start system to ensure safe operation of the power grid while allowing rapid response by limiting the maximum start and stop rate and power slope indicators.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for optimizing the start and stop of units in real time in response to frequency modulation market demand. Background Art

[0002] With the continuous expansion of the scale of power grids and the increase in electricity demand, power grid operations face many challenges. Traditional power grid management methods often fail to meet the complex needs of power systems, resulting in a series of problems in the economy, flexibility, convenience and safety of power grid operation. First, due to the complexity and uncertainty of the power system, traditional methods often fail to accurately predict the future trend of the power grid. This leads to uncertainty in power grid planning and operation, and cannot effectively respond to future power demand. Secondly, the multi-faceted performance indicators of the power grid often cannot be effectively optimized. Traditional methods can often only optimize a specific indicator, while ignoring the impact of other performance indicators. This leads to unbalanced and inefficient power grid operation. In addition, traditional methods also find it difficult to generate the best start-stop plan for the power grid under multiple objective constraints and give operational recommendations. This leads to the inflexibility and inconvenience of power grid operation, and cannot meet the optimal operation mode under different needs. Finally, the safety of the power grid also needs to be solved. Traditional methods often fail to accurately assess the safety risks of the power grid and provide corresponding measures and suggestions. This brings certain safety hazards to the operation of the power grid. Summary of the invention

[0003] The present invention provides a method for optimizing the start and stop of units in real time in response to the frequency modulation market demand, which mainly includes:

[0004] Through mobile applications, sensors and smart meters, market demand, load power data and electricity price data are collected in real time to establish demand forecasting models and predict future trends; multiple optimization models are trained to predict electricity cost and benefit indicators according to different electricity consumption scenarios, and multi-objective optimization is performed; unit health status monitoring data is obtained to evaluate the best start-stop scheme under the multi-objective constraints of grid response, market participation and reliability assurance; a closed-loop test platform based on multi-source system simulation in a digital twin environment is built with start-stop units and friendly power grids as physical execution units; a fast start system is designed to ensure the maximum start-stop rate and power slope index. It ensures safe operation of the power grid while allowing rapid response; by optimizing the load rate distribution of the units, the operating time of the units below the minimum load rate threshold is reduced, the revenue is increased on the basis of meeting the market response requirements, and the load rate distribution plan is cyclically optimized; a distributed energy management system is built, and the start-stop data and computing resources are shared by combining blockchain and smart contracts, a trust mechanism is established, the innovation and fairness of the power trading model are improved, and personalized services are provided; an online monitoring and health assessment system is set up for power grid equipment, and the impact of start-stop on the life of components is predicted through the start-stop impact assessment model, so as to provide recommended solutions for daily operation and maintenance operations.

[0005] Furthermore, the mobile application, sensors and smart meters are used to collect market demand, load data and electricity price data in real time, establish a demand forecasting model and predict future trends, including:

[0006] Through the smart meter connected to the power grid company, the market demand, load data, and electricity price data are collected in real time; if the data source is heterogeneous, it is converted into a standard format; it is determined whether the data has been verified, and if not, outlier processing is performed; based on the multivariate linear regression algorithm

[0007] y=β 0 +β 1 x t1 +β 2 x t2 +…+β kx tk +ε t

[0008] Establish a demand forecasting model; input the processed data into the demand forecasting model, calculate and output the future trend curve; if the demand forecasting model outputs an abnormality in the future trend curve, check the status of the sensor equipment; determine whether the sensor equipment needs maintenance or replacement, and if necessary, issue a work order; collect information on the number of users using the mobile application and establish a user portrait; if the number of users using the mobile application in a certain period of time exceeds the preset threshold, it is determined that there is a demand peak in this period; by analyzing the error of the demand forecasting model, determine whether the model needs to be updated; if the demand forecasting model needs to be updated, obtain new data samples and retrain the model with the new data; if the business attribute value changes by more than the preset threshold, rebuild the demand forecasting model; use the new demand forecasting model in combination with historical data and real-time data to predict the peak load period; based on the predicted peak load period, if there is a supply and demand gap, issue a warning notification and take emergency measures.

[0009] Furthermore, the training of multiple optimization models, predicting electricity cost and benefit indicators according to different electricity usage scenarios, and performing multi-objective optimization include:

[0010] Using historical data as samples and linear regression algorithms, multiple optimization models for predicting electricity costs and benefits are established; if some sample data are missing or abnormal, data completion and smoothing are performed; different scenarios are set for key influencing factors, including market demand, load power, weather and electricity prices, to form electricity scenario sample data; if the distribution of electricity scenario sample data is uneven, oversampling or undersampling is performed; multiple optimization models are trained with electricity scenario sample data, labeled as cost and benefit; using the trained optimization models, electricity costs and benefits within a planning cycle under a given scenario are predicted; the best optimization model is selected through confidence evaluation to obtain the predicted electricity cost and benefit data According to the predicted electricity cost and benefit data, the genetic algorithm is used to iteratively optimize the resource scheduling plan to obtain multiple optimal solutions; multi-objective planning is carried out with the goal of maximizing electricity benefits and minimizing electricity costs; it is determined whether the optimization result meets the system constraints. If not, the relevant parameters are adjusted and the multi-objective optimization is carried out again until the optimal solution that meets the analysis requirements is obtained; new market electricity prices and weather data are obtained to predict the electricity cost and benefit data for the next planning period; if the predicted result is greater than the preset threshold, the model needs to be retrained and optimized; among the multiple optimal solutions obtained by optimization, the resource scheduling plan with the best benefit-risk ratio is selected, and the update is sent to the scheduling entity for execution.

[0011] Furthermore, the acquisition of unit health status monitoring data and evaluation of the optimal start-stop scheme under the multi-objective constraints of power grid response, market participation and reliability assurance include:

[0012] Obtain the health status monitoring data of the unit; establish a fault prediction model based on the recurrent neural network algorithm, use the historical and real-time health status monitoring data of the unit as training samples to predict the possibility of unit failure; if the possibility of failure exceeds the preset threshold, it is judged that the unit needs priority maintenance; according to the predicted load demand curve, calculate the minimum start-stop capacity to ensure the grid frequency response requirements; generate multiple random start-stop schemes under the multi-objective constraints of grid response, market participation and reliability; for each start-stop scheme, predict the load curve change of the grid after the response; if the load response slope of a certain scheme exceeds the standard, it is judged that the start-stop scheme does not meet the grid constraints; combine the market electricity price and start-stop cost to evaluate the economic feasibility of each scheme; if the rate of return of a certain scheme is lower than the requirement, it is judged that the scheme does not meet the economic constraints; through the hierarchical analysis method, construct a hierarchical model of start-stop schemes, set the three major goals of grid response, economic benefits, and environmental impact, and determine the corresponding weight scores; select the scheme with the highest score as the recommendation result; if the recommended scheme is very different from the current scheme, generate a transition scheme to ensure a smooth transition.

[0013] Furthermore, the start-stop unit and the friendly power grid are used as physical execution units to build a closed-loop test platform based on multi-source system simulation in a digital twin environment, including:

[0014] Acquire real-time data of physical systems including start-stop units and friendly power grids; construct virtual counterparts of physical systems through digital twin technology; build a closed-loop test platform based on multi-source system simulation in the digital twin environment; set up scheduling algorithm modules including power load forecasting and resource optimization in the closed-loop test platform; acquire historical multi-source heterogeneous data, conduct correlation analysis, determine input data for different simulation test scenarios, and configure data-driven modules; use the closed-loop test platform to generate simulated input data to trigger the physical response of the start-stop units and power grid control system; determine whether the physical response meets expectations; if the physical response does not meet expectations, locate the fault module; check model parameters and data quality, and correct the digital twin system according to the inspection results; calculate the deviation between the simulated input data of the digital twin system and the data of the physical system; if the deviation exceeds the preset threshold, incrementally adjust the relevant model until the accuracy requirement is met; in the digital twin environment after the accuracy requirement is met, the start-stop units and the power grid coordinately control the closed-loop test platform; for different control strategies, use the closed-loop test platform for coordinated control of the start-stop units and the power grid to perform cyclic simulation tests, provide debugging basis, and guide parameter optimization.

[0015] Furthermore, the fast start system is designed to ensure safe operation of the power grid while allowing for rapid response by limiting the maximum start and stop rate and power slope index, including:

[0016] Query the relevant business standards issued by the power grid company, including the power slope limit and frequency response requirements for the start and stop units; set adaptability indicators including the start and stop rate and power slope; use the gray box modeling method to build a digital twin environment for the rapid start system according to different types of power stations; build a virtual model of the rapid start system based on real-time data and physical laws to simulate the behavior of the rapid start system; set limit constraints and configure reference monitoring points; obtain the real-time status data of the unit to determine whether start and stop are allowed; if the status is abnormal or the load is too high, wait until the conditions are met; use the virtual model to predict the start and stop response curve, check the maximum slope and speed Check whether the rate meets the limit requirements; if not, modify the control parameters; perform simulation verification multiple times in the digital twin system until the start-stop response meets the grid constraints; generate the final control strategy and send it to the physical system; set up online monitoring for the start-stop units, and record the key parameters of the start-stop process in real time; if the monitoring data exceeds the limit, stop immediately and adjust the controller; analyze the long-term start-stop process data of the unit to determine whether it affects the safe and stable operation of the grid; when there are analysis results that affect the safe and stable operation of the grid, add constraints to the strategy; send back execution data to adjust the simulation system, perform online updates, and continuously optimize and verify to form a closed-loop test.

[0017] Furthermore, the optimization of the load rate distribution of the units reduces the operating time of the units below the minimum load rate threshold, improves the revenue on the basis of meeting the market response requirements, and cyclically optimizes the load rate distribution plan, including:

[0018] Obtain historical operation and maintenance data, analyze the efficiency and cost of the unit at different load rates; determine the highest hourly efficiency and lowest cost at different load rates; predict the regional load demand in the next planning period based on the highest hourly efficiency, and build multi-scenario samples in combination with electricity price changes; optimize the unit load rate allocation plan based on the samples; determine whether the optimization results meet the corresponding market output requirements; if the maximum output requirement or the regulation gap exceeds the limit, relax the constraints and re-plan; evaluate the energy saving and emission reduction and additional electricity bill income of the optimal plan; check whether the power limit and start-stop number constraints of a single unit are met; issue the load rate optimization results to guide the unit to obtain the best technical and economic benefits; count the load distribution data of each unit after execution; if the unit's operating time below the minimum load rate threshold does not decrease, update and adjust the samples; based on the updated status, cyclically optimize the load rate allocation plan to reduce the difference with the global optimal.

[0019] Furthermore, the distributed energy management system is built, combining blockchain and smart contracts to share start-stop data and computing resources, establish a trust mechanism, improve the innovation and fairness of the power trading model, and provide personalized services, including:

[0020] The databases of each power station are connected through the network, and on this basis, a software system for managing energy resource data is built to form a distributed energy management system, which stores the data of each distributed energy resource in the power station; obtains the real-time status data of each distributed energy resource in the power station, deploys smart metering equipment, and uploads data to the blockchain; shares data and computing resources in the blockchain and establishes a trust mechanism; writes smart contract code based on the market electricity price and load demand in the shared data to trigger the unit's start and stop computing power request; after obtaining the start and stop request, executes the preset algorithm module, completes the calculation, and returns the optimization result; builds a start and stop optimization environment at the edge computing node of the optimization result, registers as a computing power provider, and listens for trigger events; and stores and uploads the calculation process and results to the blockchain to ensure the transparency of the process; the participating parties use the results to automatically settle electricity transactions; based on historical transaction data, uses association rules to analyze personalized service preferences; pushes personalized services to promote innovation; analyzes the ratio of computing power supply to requests, predicts gaps, and expands the management module.

[0021] Furthermore, the online monitoring and health assessment system is set up for the power grid equipment, and the impact of start-stop on the life of components is predicted through the start-stop impact assessment model, so as to provide suggestions for daily operation and maintenance operations, including:

[0022] An online monitoring and health assessment system is set up for power grid equipment to collect the operating parameters of vibration, temperature, and pressure sensors installed on the power grid equipment; thresholds for the operating parameters of the power grid equipment are set, and the current operating parameters are compared with the thresholds; if the operating parameters exceed the thresholds, the health status of the corresponding power grid equipment is assessed to be unhealthy, and an alarm is issued; a start-stop impact assessment model is established based on the recurrent neural network algorithm, and the weights and parameters of the start-stop impact assessment model are set using a random initialization method; historical fault data and start-stop operation data are obtained, and data normalization is performed; the normalized historical fault data and start-stop operation data are input into the start-stop impact assessment model for model training, and the health status data of the equipment after the start-stop operation is used as the output result; the trained start-stop The impact assessment model predicts the impact of start-stop operations in the future on the equipment health status data. If the prediction results show that continuous start-stop will shorten the life or increase the accident rate, intervention is required. Pre-set the response operation methods for each device under different health states and risks, including equipment start-stop, operating time adjustment, spare parts replacement and maintenance cycle adjustment. According to the current prediction results, the equipment response operation method is derived to form a recommended plan for daily operation and maintenance operations. The current fault data and start-stop operation data are fed back to the start-stop impact assessment model to expand the data set for model training and continuously optimize the equipment health assessment and risk prediction capabilities of the start-stop impact assessment model. The response operation methods of the equipment under different health states and risks are updated based on the results of the start-stop impact assessment model.

[0023] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0024] The present invention discloses a method for optimizing the start and stop of units in real-time response to the market demand for frequency modulation. This method combines the current advanced mobile application network and smart grid equipment to collect various data of power grid operation in real time. On the one hand, it can be used to predict the future trend of the power grid, predict and optimize various performance indicators of the power grid, and on the other hand, generate the best daily operation and maintenance proposals for the power grid under multiple target constraints. In summary, the present invention can improve the economy, flexibility, convenience and safety of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The present invention is a flow chart of a method for optimizing the start and stop of units in real time response to frequency modulation market demand.

[0026] Figure 2 It is a schematic diagram of a method for optimizing the start and stop of units in real time response to frequency modulation market demand according to the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0028] In this embodiment, a method for optimizing the start and stop of units in real time in response to frequency modulation market demand may specifically include:

[0029] S101. Through mobile applications, sensors and smart meters, market demand, load power data and electricity price data are collected in real time to establish a demand forecasting model and predict future trends.

[0030] Through the smart meter connected to the power grid company, the market demand, load data, and electricity price data are collected in real time. If the data source is heterogeneous, it is converted to a standard format. Determine whether the data has been verified. If not, perform outlier processing. Based on the multivariate linear regression algorithm

[0031] y=β 0 +β 1 x t1 +β 2 x t2 +…+β kx tk +ε t

[0032] Establish a demand forecasting model. Input the processed data into the demand forecasting model, calculate and output the future trend curve. If the demand forecasting model outputs an abnormal future trend curve, check the status of the sensor device. Determine whether the sensor device needs maintenance or replacement, and issue a work order if necessary. Collect information on the number of users using the mobile application and establish a user profile. If the number of users using the mobile application in a certain period of time exceeds the preset threshold, it is determined that there is a peak demand in this period. By analyzing the error of the demand forecasting model, determine whether the model needs to be updated. If the demand forecasting model needs to be updated, obtain the newly added data samples and retrain the model with the new data. If the business attribute value changes by more than the preset threshold, rebuild the demand forecasting model. Use the new demand forecasting model to combine historical data and real-time data to predict the peak load period. According to the predicted peak load period, if there is a gap between supply and demand, issue a warning notification and take emergency measures.

[0033] For example, load power data is collected once a minute through a smart meter connected to a power grid company. Assuming that the data sources are heterogeneous, some of the data is in kWh and the other part is in MWh. All data needs to be converted into a standard format, i.e., kWh. For example, the first piece of data collected is 5MWh, which needs to be converted into 1500kWh. Next, determine whether the data has been verified. If the collected load power data is negative or exceeds the rated load range of the system, it is considered an abnormal value. The processed data will be input into the demand forecasting model to calculate and output the future trend curve. The preset model predicts the load power in the next hour. The prediction results show that the load power in the next hour shows a gradual increase. When the future trend curve output by the demand forecasting model is abnormal, check the status of the sensor device. It is found that the temperature sensor of the sensor device is faulty, and determine whether the sensor device needs maintenance or replacement. If necessary, issue a work order. At the same time, the number of users using the mobile application is collected to establish a user profile. The preset user threshold is 800 people, and this user threshold is used to determine whether there is a demand peak during this period. Assume that the collected data shows that 1,000 people use the mobile application at the same time during a specific time period. Since the actual number of people is 1,000, it can be determined that there is a peak demand during this period. By analyzing the error of the prediction model, it can be determined whether the model needs to be updated. When the error between the calculation model prediction and the actual load power is 5%, which exceeds the preset error threshold, the model needs to be updated. Collect the load power data of the past week as a new data sample and retrain the model with the new data. When the electricity price is adjusted and the electricity price is greater than the preset threshold, the load prediction model based on the new electricity price is rebuilt. The new load prediction model is combined with historical data and real-time data, and the model is trained with historical data for the load power data of the past week, and the load power of the next hour is predicted in combination with real-time data. When the predicted load power shows a peak load period, indicating that there is a supply and demand gap in the next hour, a warning notification is issued to the relevant departments to prepare for the power demand during the peak load period and take emergency measures.

[0034] S102. Train multiple optimization models, predict electricity cost and benefit indicators according to different electricity consumption scenarios, and perform multi-objective optimization.

[0035] Taking historical data as samples, a linear regression algorithm is used to establish multiple optimization models for predicting electricity costs and benefits. If some sample data are missing or abnormal, data completion and smoothing are performed. Different scenarios are set for key influencing factors, including market demand, load power, weather and electricity prices, to form electricity scenario sample data. If the distribution of electricity scenario sample data is uneven, oversampling or undersampling is performed. Multiple optimization models are trained with electricity scenario sample data, labeled as cost and benefit. Using the trained optimization model, the electricity cost and benefit within a planning cycle under a given scenario are predicted. The best optimization model is selected through confidence evaluation to obtain the predicted electricity cost and benefit data. Based on the predicted electricity cost and benefit data, a genetic algorithm is used to iteratively optimize the resource scheduling scheme to obtain multiple optimal solutions. Multi-objective planning is performed with the goal of maximizing electricity benefits and minimizing electricity costs. It is determined whether the optimization result meets the system constraints. If not, the relevant parameters are adjusted and the multi-objective optimization is performed again until the optimal solution that meets the analysis requirements is obtained. New market electricity prices and weather data are obtained to predict the electricity cost and benefit data for the next planning cycle. If the prediction result is greater than the preset threshold, the model needs to be retrained and optimized. Among the multiple optimal solutions obtained by optimization, the resource scheduling solution with the best benefit-risk ratio is selected and updated and sent to the scheduling entity for execution.

[0036] For example, the cost and income related data of the past three years are normalized, and after outliers are deleted, they are used as training samples. The linear regression algorithm is used to establish the operation and maintenance cost prediction model, the electricity expenditure prediction model and the income prediction model respectively. The most recent month is selected as the test set, and the MAPE evaluation error is calculated, which is within 5%, meeting the requirements. Three scenarios of high, medium and low electricity prices are set, and corresponding input samples are generated according to each scenario to predict the costs and benefits under different scenarios. The input samples are represented as: [demand, load, weather, electricity price], and the labels are cost and benefit. Genetic algorithm is used for multi-objective optimization, and the objective function is to minimize cost and maximize benefit. The constraints are resource margin, etc. The population size is set to 100, and 500 generations of iterative optimization are performed to finally obtain a set of Pareto optimal solutions. The solution with the highest risk-benefit ratio is selected as the output, and the corresponding resource scheduling scheme is: the output power is [60MW, 80MW, 90MW] respectively. It is judged that the scheme meets the constraints such as unit power limit. The scheduling plan is updated, and the Techno-economic analysis is carried out after calculation and implementation. The operation and maintenance cost is reduced by 8%, ensuring the maximization of economic benefits. If the MAPE error of the forecast model is greater than 10% for three consecutive months, re-acquire the new sample data for the past year. Check and find that the data distribution has changed, re-select the algorithm and parameters for the new samples, and train the updated forecast model. Add new meteorological data input as a new feature, and perform multi-objective planning again. After adjustment, the model performance is improved, and the MAPE is maintained at around 5%. Ensure the interpretability and effectiveness of cost and benefit forecasts and resource optimization solutions.

[0037] S103. Obtain unit health status monitoring data and evaluate the optimal start-stop plan under the multi-objective constraints of grid response, market participation and reliability assurance.

[0038] Obtain the health status monitoring data of the unit. Establish a fault prediction model based on the recurrent neural network algorithm, and use the historical and real-time health status monitoring data of the unit as training samples to predict the possibility of unit failure. If the probability of failure exceeds the preset threshold, it is judged that the unit needs to be repaired first. According to the predicted load demand curve, calculate the minimum start-stop capacity to ensure the frequency response requirements of the power grid. Generate multiple random start-stop schemes under the multi-objective constraints of power grid response, market participation, and reliability. For each start-stop scheme, predict the load curve change of the power grid after the response. If the load response slope of a certain scheme exceeds the standard, it is judged that the start-stop scheme does not meet the power grid constraints. Combine the market electricity price and start-stop cost to evaluate the economic feasibility of each scheme. If the rate of return of a certain scheme is lower than the requirement, it is judged that the scheme does not meet the economic constraints. Through the hierarchical analysis method, construct a hierarchical model of start-stop schemes, set the three major goals of power grid response, economic benefits, and environmental impact, and determine the corresponding weight scores. Select the scheme with the highest score as the recommendation result. If the recommended scheme is very different from the current scheme, generate a transition scheme to ensure a smooth transition.

[0039] For example, a fault prediction model is established based on the recurrent neural network algorithm, and the historical and real-time health status monitoring data of the unit are used as training samples, including temperature, pressure, response time, frequency, and speed, to predict the probability of unit failure. The unit temperature data collected from the sensor is stored in the time series database, and the temperature range is 100-500 degrees Celsius. Read the temperature data of the last 2 hours. If there is a continuous data loss of 10 minutes, an alarm is triggered. Analyze the temperature change curve. When the predicted unit failure probability within 1 hour is greater than 20%, it is determined that it needs to be shut down for maintenance immediately. According to the load forecast results, the total power consumption in the region will vary between 2500-3000MW in the next 6 hours. Referring to the frequency response capacity requirements of the power grid company, the minimum start-stop capacity to ensure 48-52Hz frequency regulation is calculated to be 50MW. Set the start-stop capacity range to 50-100MW, the total installed capacity to 300MW, and randomly generate 20 start-stop combination schemes through Monte Carlo. Take a start-stop scheme as an example, corresponding to the start-stop of 3 60MW units. A grid transmission model is established to predict that the maximum load response slope under this scheme is 2.5MW / minute, which meets the limit requirement of less than 3MW / minute. The current market electricity price is 0.3 yuan / kWh. Referring to the historical start-stop cost accounting model, the start-stop cost of three units is about 10,000 yuan. It is predicted that the start-stop will increase the power generation income by 15,000 yuan, and the corresponding rate of return is 50%, which is higher than the 10% economic requirement. A hierarchical model of start-stop schemes is constructed, and the three major goals of grid response, economic benefits, and environmental impact are set, and the corresponding weight scores are determined to be [0.3, 0.5, 0.2]. The scores of each start-stop scheme are evaluated, and the scheme with a score of 0.5 is selected. Since the recommended scheme is quite different from the current scheme, the current scheme is to start one unit with a capacity of 70MW, so a transition scheme is generated. The transition scheme is to start an additional unit with a capacity of 30MW first, and then stop the current unit, so that the total capacity is 80MW.

[0040] S104. Using the start-stop units and friendly power grids as physical execution units, a closed-loop test platform based on multi-source system simulation in a digital twin environment is built.

[0041] Acquire real-time data of physical systems including start-stop units and friendly power grids. Build virtual counterparts of physical systems through digital twin technology. In the digital twin environment, build a closed-loop test platform based on multi-source system simulation. Set up scheduling algorithm modules including power load forecasting and resource optimization in the closed-loop test platform. Acquire historical multi-source heterogeneous data, perform correlation analysis, determine the input data for different scenarios of simulation tests, and configure data-driven modules. Generate simulated input data using the closed-loop test platform to trigger the physical response of the start-stop units and power grid control system. Determine whether the physical response meets expectations. If the physical response does not meet expectations, locate the faulty module. Check model parameters and data quality, and correct the digital twin system based on the inspection results. Calculate the deviation between the simulated input data of the digital twin system and the data of the physical system. If the deviation exceeds the preset threshold, incrementally adjust the relevant model until the accuracy requirement is met. In the digital twin environment after the accuracy requirement is met, the start-stop units and the power grid are collaboratively controlled to control the closed-loop test platform. For different control strategies, use the closed-loop test platform for collaborative control of the start-stop units and the power grid to perform cyclic simulation tests, provide debugging basis, and guide parameter optimization.

[0042] For example, obtain the real-time active power, reactive power data, temperature, speed and other parameters of a start-stop unit, with a sampling frequency of 50Hz. Apply digital twin technology to build a three-dimensional model of the unit and set relevant physical characteristic parameters. Configure similar performance parameters and status data interfaces in the model. Use the cyclic iterative update between the platform's real-time data input and model data output to realize the digital twin system of the unit. In this environment, build a power station test platform containing multiple unit digital twin systems. Set up the grid connection model, configure the power load forecasting and unit start-stop optimization algorithm modules. Obtain historical daily load curve data, use the LSTM model to predict the load demand for the next week, and use the output curve as one of the test inputs. Simulate a high load scenario, and the corresponding predicted power peak will occur on Friday, exceeding 80% of the total installed capacity. For this input, trigger the test platform, and the unit digital twin system simulates the adjustment of output power according to the start-stop optimization results. Detect whether the total output power response of the platform relay power station meets the requirements of the plan. If the load response curve fluctuates beyond the limit, the test is judged to be abnormal and the model parameter settings are checked. Compare the actual power data of the unit with the power output of the digital twin. If the average deviation exceeds 5%, increase the random noise component in the model to make the digital twin system response closer to reality. In this way, the test platform is iteratively optimized to provide a reliable simulation environment for the start-stop control of the unit.

[0043] S105. Design a fast start system to ensure safe operation of the power grid while allowing rapid response by limiting the maximum start and stop rate and power slope indicators.

[0044] Query the relevant business standards issued by the power grid company, including the power slope limit and frequency response requirements for the start-stop units. Set adaptability indicators including the start-stop rate and power slope. According to different types of power stations, use the gray box modeling method to build a digital twin environment for the rapid start system. Construct a virtual model of the rapid start system based on real-time data and physical laws to simulate the behavior of the rapid start system. Set limit constraints and configure reference monitoring points. Obtain the real-time status data of the unit to determine whether start-stop is allowed. If the status is abnormal or the load is too high, wait until the conditions are met. Use the virtual model to predict the start-stop response curve and check whether the maximum slope and rate meet the limit requirements. If not, modify the control parameters. Perform multiple simulations in the digital twin system until the start-stop response meets the grid constraints. Generate the final control strategy and send it to the physical system. Set up online monitoring for the start-stop units and record the key parameters in the start-stop process in real time. If the monitoring data exceeds the limit, stop immediately and adjust the controller. Analyze the long-term start-stop process data of the unit to determine whether it affects the safe and stable operation of the power grid. When there are analysis results that affect the safe and stable operation of the power grid, add constraints to the strategy. The returned execution data is used to adjust the simulation system, perform online updates, and continuously optimize and verify to form a closed-loop test.

[0045] For example, the standards issued by the power grid company that were queried require that during the start-up and shutdown of the unit, the power change slope shall not exceed 10MW / min, and the adjustment rate shall not exceed 2MW / min. Set the corresponding limit values ​​and configure the monitoring points, and build a digital twin fast start system for a certain type of unit through the gray box modeling method. Obtain the real-time active power, speed and other data of the unit, and determine that the start and stop are allowed when the speed is within the normal range of 1890-1910rpm. Test the response curves under various conditions and determine that when the valve opening increase rate is controlled at 10% / min, the power slope can be controlled within 10MW / min. The operation results meet the requirements, and the corresponding control strategy is formed and sent to the physical system for execution. Online monitoring of power and speed is set for the start-up and shutdown units. When the power slope is detected to suddenly increase to 12MW / min, the abnormality is recorded and the shutdown command is immediately issued. Analysis shows that this phenomenon mainly occurs when the unit starts after a long period of low load. The problem is solved by limiting the number of rapid starts and stops after low-load operation. After collecting data on the start-stop process for one year, according to the analysis of the grid operation data, no adverse effects of unit start-up on system stability were detected. The digital twin rapid start system was continuously optimized using actual operation data to make the simulation results consistent with the actual situation, forming an accurate closed-loop test platform.

[0046] S106. By optimizing the load rate distribution of the units, the operating time of the units below the minimum load rate threshold is reduced, the revenue is increased on the basis of meeting the market response requirements, and the load rate distribution plan is cyclically optimized.

[0047] Obtain historical operation and maintenance data and analyze the efficiency and cost of the unit at different load rates. Determine the highest hourly efficiency and lowest cost at different load rates. Predict the regional load demand for the next planning period based on the highest hourly efficiency, and build multi-scenario samples in combination with electricity price changes. Optimize the unit load rate allocation plan based on the sample. Determine whether the optimization result meets the corresponding market output requirements. If the maximum output requirement or the adjustment gap exceeds the limit, relax the constraints and re-plan. Evaluate the energy saving and emission reduction and additional electricity bill income of the optimal solution. Check whether the power limit and start-stop number constraints of a single unit are met. Send the load rate optimization results to guide the unit to obtain the best technical and economic benefits. Statistical load distribution data of each unit after execution. If the unit's operating time below the minimum load rate threshold does not decrease, update and adjust the sample. Based on the updated status, cyclically optimize the load rate allocation plan to reduce the difference with the global optimal.

[0048] For example, from the analysis of historical operation data, the average efficiency of the unit can reach 38% when the load is 60%-90%, and the efficiency is significantly reduced when the load is lower than 30% or higher than 95%. It is predicted that the total demand load in the region will remain at 300-400MW in the next week, and three scenarios of high, medium and low are set with reference to weather temperature and electricity price level. For different scenarios, the biological inspiration algorithm with a population size of 50 is set for 300 generations of iteration to optimize the load rate distribution of the six units. The objective function is to maximize the 60%-90% high efficiency zone and reduce the low load ratio. The constraints include maximum output limit, etc. The optimization results show that under the expected high load scenario, the average efficiency can be increased to 36%, and the low efficiency operation ratio is controlled at 15%, meeting the maximum output requirements of the corresponding frequency regulation support service. Check that the load of each unit is within the technical allowable range and the number of starts and stops also meets the standards. The evaluation can increase the electricity bill income by about 10,000 yuan per week and reduce CO2 emissions by 20 tons. The plan was issued for execution, but statistics showed that the operating time of Unit 3 below the minimum load rate threshold was still higher than the duration threshold requirement. Therefore, its weight in the Fitness function was adjusted to strengthen the optimization target constraint strength. After 5 iterations, an improved optimization plan was obtained.

[0049] S107. Build a distributed energy management system, combine blockchain and smart contracts to share start and stop data and computing resources, establish a trust mechanism, improve the innovation and fairness of electricity trading models, and provide personalized services.

[0050] Connect the databases of each power station through the network, and build a software system for managing energy resource data on this basis to form a distributed energy management system and store the data of each distributed energy resource in the power station. Obtain the real-time status data of each distributed energy resource in the power station, deploy smart metering equipment, and upload data to the blockchain. Share data and computing resources in the blockchain to establish a trust mechanism. Write smart contract code based on the market electricity price and load demand in the shared data to trigger the unit start and stop computing power request. After obtaining the start and stop request, execute the preset algorithm module, complete the calculation, and return the optimization result. Build a start and stop optimization environment at the edge computing node of the optimization result, register as a computing power provider, and listen for trigger events. And upload the calculation process and results to the blockchain to ensure the transparency of the process. The participating parties use the results to automatically settle power transactions. Based on historical transaction data, use association rules to analyze personalized service preferences. Push personalized services to promote innovation. Analyze the ratio of computing power supply and request, predict gaps, and expand the capacity management module.

[0051] For example, a distributed energy management system is built based on Hadoop. Distributed energy nodes such as photovoltaics and energy storage in the power station upload power generation, SOC and other data to the blockchain platform in real time through the LoRa network. Blockchain encryption, timestamp and other technologies are used to ensure the trustworthiness and traceability of the data sharing process. When it is predicted that the supplier needs to be started and stopped during the peak, the smart contract code is triggered, and the IDential outputs the start-stop calculation request event, requiring a 50MW response capacity, the fastest response time is 5min, and the calculation fee is settled by volume. The blockchain edge computing node has built a unit start-stop optimization platform, registered as a qualified computing provider, and monitored events in real time. After receiving the request, it automatically reads the relevant load data of the blockchain database, runs the reserved start-stop algorithm module, completes the calculation within the limited time, and returns the optimized start-stop solution of 60MW for 3 units to meet the requirements. At the same time, the calculation steps and result hash summaries are recorded in the transaction details to ensure that the calculation process is transparent and auditable. Finally, all parties use the results to automatically complete the transaction settlement of electricity and currency. After counting multiple transactions, it was found that 80% of users would choose to request CO2 emission offset services. Therefore, we push new transaction models to further expand the platform's service scope and enhance user experience.

[0052] S108. Set up an online monitoring and health assessment system for power grid equipment, and use the start-stop impact assessment model to predict the impact of start-stop on component life, so as to provide recommended solutions for daily operation and maintenance operations.

[0053] An online monitoring and health assessment system is set up for power grid equipment to collect the operating parameters of vibration, temperature, and pressure sensors installed on power grid equipment. Thresholds for power grid equipment operating parameters are set, and the current operating parameters are compared with the thresholds. If the operating parameters exceed the thresholds, the health status of the corresponding power grid equipment is assessed as unhealthy, and an alarm is issued. A start-stop impact assessment model is established based on the recurrent neural network algorithm, and the weights and parameters of the start-stop impact assessment model are set using a random initialization method. Historical fault data and start-stop operation data are obtained, and data normalization is performed. The normalized historical fault data and start-stop operation data are input into the start-stop impact assessment model for model training, and the health status data of the equipment after the start-stop operation is used as the output result. The trained start-stop impact assessment model is used to predict the impact of start-stop operations on the health status data of the equipment in the future period. If the prediction results show that continuous start-stop operations will shorten the life or increase the accident rate, intervention is required. Pre-set the response operation methods for each device under different health states and risks, including equipment start-stop, operation time adjustment, spare parts replacement, and maintenance cycle adjustment. The response operation methods of the equipment are obtained based on the current prediction results, and a recommended plan for daily operation and maintenance operations is formed. Feed the current fault data and start-stop operation data back to the start-stop impact assessment model to expand the data set for model training and continuously optimize the equipment health assessment and risk prediction capabilities of the start-stop impact assessment model. The results of the comprehensive start-stop impact assessment model are used to update the equipment's response methods under different health states and risks.

[0054] For example, temperature and vibration sensors are arranged at the turbine bearing position, with a sampling frequency of 100Hz and a vibration collection range of 0.1-60Hz. Real-time data collection is input into the health assessment module, and a threshold is set. When the temperature exceeds 300 degrees Celsius or the vibration amplitude exceeds 5 microns, it is judged as an unhealthy state and an alarm is issued. The historical fault data and start-stop operation data of the steam turbine in the past two years are obtained, and the start-stop impact assessment model of the steam turbine is established based on the recurrent neural network algorithm LSTM. The weights and parameters of the start-stop impact assessment model are set using a random initialization method. The historical fault data and start-stop operation data of the steam turbine are obtained, and the data are normalized using MinMaxScaler. The normalized historical fault data and start-stop operation data are input into the start-stop impact assessment model for model training, and the health status data of the equipment after the start-stop operation is used as the output result. The start-stop impact assessment model is used to predict the impact of start-stop on the health status of the equipment in the next three months once a week, and an impact score is given, ranging from 0 to 1. The higher the score, the more serious the loss. Statistics show that after three consecutive starts and stops, the fatigue index of components increases to 0.75. To avoid the risk of failure, it is recommended to rest for at least 12 hours after each start and stop. After updating the response method, the number of turbine-related failures in the past six months has decreased by 23% compared with before.

[0055] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for optimizing the start and stop of units in real-time response to frequency modulation market demand, characterized in that: The method comprises: Through mobile applications, sensors and smart meters, market demand, load power data and electricity price data are collected in real time to establish demand forecasting models and predict future trends; multiple optimization models are trained to predict electricity cost and benefit indicators according to different electricity consumption scenarios, and multi-objective optimization is performed; unit health status monitoring data is obtained to evaluate the best start-stop scheme under the multi-objective constraints of grid response, market participation and reliability assurance; a closed-loop test platform based on multi-source system simulation in a digital twin environment is built with start-stop units and friendly power grids as physical execution units; a fast start system is designed to ensure the maximum start-stop rate and power slope index. Ensure safe operation of the power grid while allowing rapid response; optimize the load rate distribution of the units, reduce the operating time of the units below the minimum load rate threshold, increase revenue on the basis of meeting market response requirements, and cyclically optimize the load rate distribution plan; build a distributed energy management system, combine blockchain and smart contracts to share start-stop data and computing resources, establish a trust mechanism, improve the innovation and fairness of the power trading model, and provide personalized services; set up an online monitoring and health assessment system for power grid equipment, and use the start-stop impact assessment model to predict the impact of start-stop on the life of components, so as to provide recommended solutions for daily operation and maintenance operations; Among them, the training of multiple optimization models, predicting electricity cost and benefit indicators according to different electricity consumption scenarios, and performing multi-objective optimization include: using historical data as samples, using linear regression algorithm to establish multiple optimization models for predicting electricity cost and benefit; if some sample data are missing or abnormal, data completion and smoothing are performed; different scenarios are set for key influencing factors, including market demand, load power, weather and electricity price, to form electricity consumption scenario sample data; if the distribution of electricity consumption scenario sample data is uneven, oversampling or undersampling is performed; multiple optimization models are trained with electricity consumption scenario sample data, labeled as cost and benefit; using the trained optimization model to predict the electricity cost and benefit within a planning cycle under a given scenario; through confidence Evaluate and select the best optimization model to obtain predicted electricity cost-benefit data; based on the predicted electricity cost-benefit data, use genetic algorithms to iteratively optimize resource scheduling solutions to obtain multiple optimal solutions; perform multi-objective planning with the goal of maximizing electricity benefits and minimizing electricity costs; determine whether the optimization results meet the system constraints. If not, adjust the relevant parameters and re-perform multi-objective optimization until the optimal solution that meets the analysis requirements is obtained; obtain new market electricity prices and weather data to predict the electricity cost and benefit data for the next planning cycle; if the predicted result is greater than the preset threshold, retrain the model and optimize; among the multiple optimal solutions obtained by optimization, select the resource scheduling solution with the best benefit-risk ratio, and update and send it to the scheduling entity for execution; Among them, the construction of a distributed energy management system, combining blockchain and smart contracts to share start-stop data and computing resources, establish a trust mechanism, improve the innovation and fairness of the power trading model and provide personalized services, including: connecting the databases of each power station through the network, and building a software system for managing energy resource data on this basis to form a distributed energy management system, storing the distributed energy resource data in the power station; obtaining the real-time status data of each distributed energy resource in the power station, deploying smart metering equipment, and uploading data to the blockchain; sharing data and computing resources in the blockchain to establish a trust mechanism; writing smart contract code according to the market electricity price and load demand in the shared data to trigger the unit start-stop computing power request; after obtaining the start-stop request, executing the preset algorithm module, completing the calculation, and returning the optimization result; building a start-stop optimization environment at the edge computing node of the optimization result, registering as a computing power provider, and listening to the trigger event; and storing and uploading the calculation process and results to the blockchain to ensure the transparency of the process; the participating parties use the results to automatically settle power transactions; based on historical transaction data, using association rules to analyze personalized service preferences; pushing personalized services to promote innovation; analyzing the ratio of computing power supply to request, predicting gaps, and expanding the capacity management module; Among them, the online monitoring and health assessment system is set for the power grid equipment, and the impact of start-stop on the life of components is predicted through the start-stop impact assessment model, so as to provide suggestions for daily operation and maintenance operations, including: setting an online monitoring and health assessment system for the power grid equipment, collecting the operating parameters of vibration, temperature and pressure sensors installed on the power grid equipment; setting a threshold for the operating parameters of the power grid equipment, and comparing the current operating parameters with the threshold; if the operating parameters exceed the threshold, the health status of the corresponding power grid equipment is assessed as unhealthy, and an alarm is issued; establishing a start-stop impact assessment model based on the recurrent neural network algorithm, and using a random initialization method to set the weights and parameters of the start-stop impact assessment model; obtaining historical fault data and start-stop operation data, and performing data normalization processing; inputting the normalized historical fault data and start-stop operation data into the start-stop impact assessment model for Model training uses the health status data of the equipment after start-stop operations as the output result; use the trained start-stop impact assessment model to predict the impact of start-stop operations on the equipment health status data in the future period; if the prediction results show that continuous start-stop will shorten the life or increase the accident rate, intervention is required; pre-set the response operation methods for each device under different health states and risks, including equipment start-stop, operation time adjustment, spare parts replacement and maintenance cycle adjustment; derive the equipment response operation methods based on the current prediction results, and form a recommended plan for daily operation and maintenance operations; feed back the current fault data and start-stop operation data to the start-stop impact assessment model, expand the data set for model training, and continuously optimize the equipment health assessment and risk prediction capabilities of the start-stop impact assessment model; update the equipment response operation methods under different health states and risks based on the results of the start-stop impact assessment model.

2. The method according to claim 1, wherein: The method uses mobile applications, sensors and smart meters to collect market demand, load data, and electricity price data in real time, establish a demand forecasting model and predict future trends, including: Through smart meters connected to the power grid company, real-time data on market demand, load power and electricity prices are collected; If the data sources are heterogeneous, they are converted to a standard format; determine whether the data has been verified, and if not, perform outlier processing; based on the multivariate linear regression algorithm y=β 0 +β 1 x t1 +β 2 x t2 +…+β kx tk +ε t Establish a demand forecasting model; input the processed data into the demand forecasting model, calculate and output the future trend curve; if the demand forecasting model outputs an abnormal future trend curve, check the status of the sensor equipment; determine whether the sensor equipment needs maintenance or replacement, and if necessary, issue a work order; collect information on the number of users using the mobile application and establish a user profile; if the number of users using the mobile application in a certain period of time exceeds the preset threshold, it is determined that there is a demand peak in this period; by analyzing the error of the demand forecasting model, determine whether the model needs to be updated; If the demand forecasting model needs to be updated, new data samples are obtained and the model is retrained using the new data. If the business attribute value changes by more than the preset threshold, the demand forecasting model is rebuilt. The new demand forecasting model is used in combination with historical data and real-time data to predict the peak load period. Based on the predicted peak load period, if there is a supply-demand gap, a warning notification is issued and emergency measures are taken.

3. The method according to claim 1, wherein: The acquisition of unit health status monitoring data and evaluation of the optimal start-stop plan under the constraints of grid response, market participation and reliability guarantee include: Obtain the unit's health status monitoring data; establish a fault prediction model based on the recurrent neural network algorithm, using the unit's historical and real-time health status monitoring data as training samples to predict the possibility of unit failure; If the probability of failure exceeds the preset threshold, the unit is judged to need priority maintenance; based on the predicted load demand curve, the minimum start-stop capacity to ensure the grid frequency response requirements is calculated; multiple random start-stop plans are generated under the multi-objective constraints of grid response, market participation, and reliability; for each start-stop plan, the load curve change of the grid after the response is predicted; if the load response slope of a plan exceeds the standard, the start-stop plan is judged to be inconsistent with the grid constraints; combined with the market electricity price and start-stop cost, the economic feasibility of each plan is evaluated; If the rate of return of a certain plan is lower than the requirement, it is judged that the plan does not meet the economic constraints; through the hierarchical analysis method, a hierarchical model of the start-stop plan is constructed, the three major goals of power grid response, economic benefits and environmental impact are set, and the corresponding weight scores are determined; the plan with the highest score is selected as the recommendation result; If the recommended solution is significantly different from the current solution, a transition solution is generated to ensure a smooth transition.

4. The method according to claim 1, wherein: The start-stop unit and the friendly power grid are used as physical execution units to build a closed-loop test platform based on multi-source system simulation in a digital twin environment, including: Obtain real-time data on physical systems including start-stop units and friendly power grids; build virtual counterparts of physical systems through digital twin technology; In the digital twin environment, a closed-loop test platform based on multi-source system simulation is built; a scheduling algorithm module including power load forecasting and resource optimization is set up in the closed-loop test platform; Obtain historical multi-source heterogeneous data, conduct correlation analysis, determine the input data for different simulation test scenarios, and configure the data-driven module; use the closed-loop test platform to generate simulation input data to trigger the physical response of the start-stop unit and the power grid control system; determine whether the physical response meets expectations; if the physical response does not meet expectations, locate the faulty module; check model parameters and data quality, and correct the digital twin system based on the inspection results; calculate the deviation between the simulation input data of the digital twin system and the data of the physical system; If the deviation exceeds the preset threshold, the relevant model is adjusted incrementally until the accuracy requirement is met. In the digital twin environment after the accuracy requirement is met, the closed-loop test platform is controlled by the start-stop unit and the power grid in a coordinated manner. For different control strategies, the closed-loop test platform for coordinated control of the start-stop unit and the power grid is used to perform cyclic simulation tests to provide a basis for debugging and guide parameter optimization.

5. The method according to claim 1, wherein: The fast start system designed ensures safe operation of the power grid while allowing fast response by limiting the maximum start and stop rate and power slope index, including: Query the relevant business standards issued by the power grid company, including the power slope limit and frequency response requirements for starting and stopping units; set adaptability indicators including start-stop rate and power slope; use the gray box modeling method to build a digital twin environment for the rapid start system according to different types of power stations; build a virtual model of the rapid start system based on real-time data and physical laws to simulate the behavior of the rapid start system; Set limit constraints and configure reference monitoring points; obtain real-time status data of the unit to determine whether to allow start and stop; if the status is abnormal or the load is too high, wait until the conditions are met; Use the virtual model to predict the start-stop response curve and check whether the maximum slope and rate meet the limit requirements; if not, modify the control parameters; perform multiple simulations in the digital twin system until the start-stop response meets the grid constraints; Generate the final control strategy and send it to the physical system; set up online monitoring for the start and stop units, and record the key parameters of the start and stop process in real time; if the monitoring data exceeds the limit, stop immediately and adjust the controller; analyze the long-term start and stop process data of the unit to determine whether it affects the safe and stable operation of the power grid; when there are analysis results that affect the safe and stable operation of the power grid, add constraints to the strategy; send back execution data to adjust the simulation system, perform online updates, and continuously optimize and verify to form a closed-loop test.

6. The method according to claim 1, wherein: The above mentioned method optimizes the load rate distribution of the units, reduces the operation time of the units below the minimum load rate threshold, improves the revenue on the basis of meeting the market response requirements, and cyclically optimizes the load rate distribution plan, including: Obtain historical operation and maintenance data, analyze the efficiency and cost of the unit at different load rates; determine the highest efficiency and lowest cost at different load rates; Predict regional load demand for the next planning period based on the highest efficiency, and build multi-scenario samples in combination with electricity price changes; Optimize the unit load rate allocation plan based on samples; determine whether the optimization results meet the corresponding market output requirements; If the maximum output requirement or the regulation gap exceeds the limit, relax the constraints and re-plan; evaluate the energy saving and emission reduction and additional electricity bill income of the optimal solution; check whether the power limit and start-stop number constraints of a single unit are met; send the load rate optimization results to guide the unit to obtain the best technical and economic benefits; count the load distribution data of each unit after execution; if the operating time of the unit below the minimum load rate threshold does not decrease, update and adjust the sample; based on the updated status, cyclically optimize the load rate distribution plan to reduce the difference with the global optimum.

Images