User power utilization optimization method for output collaboration of new energy power supply cluster

Abstract

The invention provides a new energy power supply cluster output collaborative user power consumption optimization method, and relates to the technical field of user power consumption optimization, and the method comprises the steps: obtaining a power grid physical structure and real-time operation state data, constructing a safety check model, and obtaining a new energy output scheme set; evaluating the new energy output scheme set by adopting a fuzzy comprehensive evaluation method, and screening an optimal new energy output scheme; introducing a power supply start-stop state variable into the security check model, and obtaining a new energy output and power supply combination optimization scheme considering power supply start-stop; and constructing a multi-time-scale source network load storage collaborative optimization scheduling model to obtain an optimization scheduling curve of each link. The big data platform is deployed to collect the operation data, coordination and unification of efficient consumption of new energy and safe and stable operation of the power grid are achieved, and the overall operation efficiency and economical efficiency of the system are improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for optimizing power consumption of users with coordinated output of a new energy power supply cluster. Background Art

[0002] When building a safety verification model that considers the physical structure of the power grid and real-time operating status data, we are faced with the technical challenge of how to balance the feasibility and economy of different new energy output schemes under multiple constraints such as line flow, voltage amplitude, and equipment capacity. Due to the complexity and nonlinear characteristics of the power grid, it is difficult for traditional optimization algorithms to take into account multiple constraints and optimization goals at the same time, resulting in the possibility of safety hazards or insufficient economy in the obtained new energy output scheme. In addition, when considering the start-stop characteristics of different types of power sources, how to reasonably introduce the start-stop state of the power source as an optimization variable into the safety verification model, and dynamically adjust the power source combination during the optimization process to avoid frequent start-stop or dispatching instructions that exceed the adjustment capacity range is also a technical challenge that needs to be solved urgently. At the same time, in the actual implementation process of the optimized dispatching scheme, it is also necessary to coordinate the information interaction and control instruction issuance between the new energy power station, the user-side intelligent control equipment, and the power grid dispatching center. How to design a set of efficient and reliable communication protocols and control strategies to achieve accurate coordination of all links of source, network, load and storage, and improve the level of new energy consumption and user power consumption experience is also a complex technical problem. Summary of the invention

[0003] The present invention provides a method for optimizing user electricity consumption by coordinating the output of a new energy power supply cluster, which mainly includes:

[0004] The physical structure and real-time operation status data of the power grid are obtained, a safety verification model is constructed, and a set of new energy output schemes is obtained; the fuzzy comprehensive evaluation method is used to evaluate the set of new energy output schemes and select the optimal new energy output scheme; the power supply start-stop state variables are introduced into the safety verification model to obtain the new energy output and power supply combination optimization scheme considering the start-stop of the power supply; a multi-time scale source-grid-load-storage collaborative optimization scheduling model is constructed to obtain the optimized scheduling curve of each link; a communication mechanism based on the MQTT protocol is designed to realize information interaction between each link to ensure real-time data interaction and coordinated control between each link; according to the optimized new energy output scheme and the source-grid-load-storage collaborative scheduling curve, the active power, reactive power and voltage regulation instructions of each power station are calculated to realize smooth control of new energy output and voltage-reactive optimization; a big data platform is deployed to continuously optimize the new energy power prediction model and load demand response model using an online machine learning algorithm, and a reinforcement learning algorithm is used to autonomously learn and adjust the source-grid-load-storage collaborative optimization strategy to maximize the new energy absorption rate and minimize the system operation cost.

[0005] Furthermore, the method of acquiring the physical structure and real-time operation status data of the power grid, constructing a safety verification model, and obtaining a set of new energy output schemes includes: acquiring constraint condition parameters such as line flow, voltage amplitude, and equipment capacity according to the physical structure and real-time operation status data of the power grid; using a penalty function method to convert the constraint conditions into a part of the objective function and constructing a safety verification model; and using an improved particle swarm optimization algorithm to solve the objective function to obtain a set of new energy output schemes that meet the safety constraints.

[0006] Furthermore, the fuzzy comprehensive evaluation method is used to evaluate the set of new energy output schemes and screen the optimal new energy output scheme, including: obtaining evaluation indicators such as power grid safety margin, new energy consumption rate, and system economy according to the set of new energy output schemes; constructing a multi-level fuzzy evaluation matrix to calculate the comprehensive score of each scheme; and screening out the optimal new energy output scheme according to the comprehensive score ranking results.

[0007] Furthermore, the power start and stop state variables are introduced into the safety verification model to obtain an optimization solution for new energy output and power combination considering power start and stop, including: setting the minimum duration and maximum start and stop number parameters of the power start and stop to limit frequent power start and stop; using a mixed integer programming algorithm to solve the optimization model containing the power start and stop state variables to obtain an optimization solution for new energy output and power combination considering power start and stop.

[0008] Furthermore, the multi-time-scale source-grid-load-storage collaborative optimization scheduling model is constructed to obtain the optimized scheduling curve of each link, including: setting new energy power prediction, load demand response, and energy storage charging and discharging plan decision variables; using a distributed model predictive control method to achieve hierarchical distributed solution, and through iterative calculation, the scheduling curves of each link are made consistent to obtain the optimized scheduling curves of each link.

[0009] Furthermore, the design realizes information interaction between various links based on the communication mechanism of the MQTT protocol, ensuring real-time data interaction and coordinated control between various links, including: using the Kafka message queue to cache and sort scheduling instructions by timestamp; deploying the Zookeeper distributed coordination service to realize state synchronization and failover of each node.

[0010] Furthermore, according to the optimized new energy output plan and the source-grid-load-storage coordinated dispatch curve, the active power, reactive power and voltage regulation instructions of each power station are calculated to achieve smooth control of new energy output and voltage-reactive optimization, including: using a model predictive control algorithm to calculate the active power, reactive power and voltage regulation instructions of each power station in real time; sending control instructions to each new energy power station and intelligent control equipment through the power communication network, and using intelligent converters and reactive power compensation devices to respond to dispatch instructions.

[0011] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0012] The present invention discloses a method for optimizing user power consumption with coordinated output of a new energy power cluster. The method first constructs a safety verification model that considers multiple constraints, and uses an improved particle swarm algorithm to solve and obtain a set of new energy output solutions. Then, a fuzzy comprehensive evaluation method is used to evaluate and screen the solutions, and the power supply start-stop characteristics are introduced for further optimization. Then, a multi-time scale source-grid-load-storage coordinated optimization scheduling model is constructed, and a distributed model predictive control method is used to achieve hierarchical distributed solution. In this process, a communication mechanism based on the MQTT protocol is designed to realize information interaction in each link. Finally, the model predictive control algorithm is used to calculate and issue control instructions in real time to achieve smooth control of new energy output and voltage and reactive power optimization. The present invention also deploys a big data platform to collect operating data, and uses online machine learning and reinforcement learning algorithms to continuously optimize prediction models and decision-making strategies. Through the organic combination of the above technical solutions, the present invention realizes the coordinated unification of efficient consumption of new energy and safe and stable operation of the power grid, and improves the overall operating efficiency and economy of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 The present invention is a flow chart of a method for optimizing user electricity consumption by coordinating the output of a new energy power supply cluster. DETAILED DESCRIPTION

[0014] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0015] like Figure 1 In this embodiment, a method for optimizing user power consumption by coordinating the output of a new energy power supply cluster may specifically include:

[0016] Step S101, based on the physical structure of the power grid and real-time operating status data, a safety verification model is constructed that takes into account multiple constraints such as line flow, voltage amplitude and equipment capacity. The constraint conditions are converted into part of the objective function by introducing a penalty function, and the objective function is solved using an improved particle swarm optimization algorithm. The algorithm improves the convergence speed and global search capability by introducing adaptive inertia weights and learning factors, and obtains a set of new energy output solutions that meet safety constraints.

[0017] According to the physical structure of the power grid and the real-time operation status data, multiple constraint parameters such as line flow, voltage amplitude and equipment capacity are obtained. For multiple constraints, the penalty function method is used to transform the constraints into part of the objective function, and a safety verification model considering the line flow, voltage amplitude and equipment capacity constraints is constructed. For the constructed safety verification model, the improved particle swarm optimization algorithm is used to solve the objective function. In the particle swarm optimization algorithm, adaptive inertia weight and learning factor are introduced. By dynamically adjusting the inertia weight and learning factor, the convergence speed and global search ability of the algorithm are improved. Specifically, the inertia weight is set to decrease linearly with the number of iterations, and the learning factor changes nonlinearly with the particle fitness to balance the local search and global search capabilities of the algorithm. Using the improved particle swarm optimization algorithm, the objective function of the safety verification model is iteratively optimized and solved to obtain a set of new energy output solutions that meet the constraints of line flow, voltage amplitude and equipment capacity. For the set of renewable energy output solutions obtained through optimization, the power flow calculation program is used to determine whether each solution meets the constraints of safe operation of the power grid, including line power flow limits, bus voltage amplitude limits, etc. If the constraints are met, the solution is added to the feasible solution set. After obtaining the feasible solution set, the solution with the largest total renewable energy output and the smallest network loss is selected as the optimal renewable energy output dispatching solution. The renewable energy output instructions in the optimal dispatching solution are issued to each renewable energy station, and the optimization results are fed back to the power grid dispatching control system to guide the real-time safe and economic dispatching of the power grid, so as to achieve efficient consumption of renewable energy and safe and stable operation of the power grid.

[0018] Specifically, the goal of power grid security verification is to ensure that the power system can operate safely and stably under various operating conditions. In order to achieve this goal, it is necessary to analyze the physical structure and real-time operating status data of the power grid to obtain constraint parameters such as line flow, voltage amplitude and equipment capacity. For example, the maximum transmission capacity of a transmission line is 100MW, the capacity of a transformer is 50MVA, the voltage amplitude of a bus must be maintained between 95pu and 05pu, etc. These constraint parameters are the basis for security verification. The penalty function method is a common method to convert constraints into part of the objective function. For example, if the flow of a line exceeds its limit, a penalty term is added to the objective function, and the size of the penalty term is proportional to the degree of excess. In this way, during the optimization process, the algorithm will automatically avoid violations of constraints. Assuming that the objective function is to minimize network loss, if a certain scheme causes the line flow to exceed the limit, the network loss of the scheme will be added with a penalty value, so that its total objective function value increases, thereby reducing the probability of the scheme being selected. The particle swarm optimization algorithm is an intelligent optimization algorithm that simulates the foraging behavior of bird flocks. Each particle represents a possible solution. By continuously iterating and updating the position and velocity of the particle, the optimal solution is finally found. The introduction of adaptive inertia weight and learning factor can improve the performance of the algorithm. For example, in the early stage of iteration, the inertia weight is large, which is conducive to global search; in the later stage of iteration, the inertia weight is small, which is conducive to local search. Assume that the initial inertia weight is 9, and it decreases linearly to 4 as the number of iterations increases. The learning factor controls the degree to which the particle learns from the individual optimal solution and the global optimal solution. It can be nonlinearly adjusted according to the particle fitness to balance the local search and global search capabilities. Assume that the initial value of the learning factor is 2, and it decreases nonlinearly to 5 as the particle fitness increases. Through the improved particle swarm optimization algorithm, the objective function of the safety verification model can be iteratively optimized and solved to obtain a set of new energy output solutions that meet the constraints of line flow, voltage amplitude and equipment capacity. For example, multiple solutions may be obtained, each of which specifies the output value of each new energy station. After obtaining the set of new energy output solutions, it is necessary to use the power flow calculation program to verify each solution to determine whether it meets the constraints of safe operation of the power grid. The power flow calculation program can simulate the operating state of the power system and calculate parameters such as line power flow and bus voltage. For example, if the new energy output of a scheme causes the power flow of a certain line to exceed the limit, the scheme is considered infeasible. After obtaining the feasible solution set, the optimal scheme needs to be selected from it. The selection criteria can be the maximum total output of new energy and the minimum network loss. For example, there are two feasible schemes. The total output of new energy in Scheme A is 100MW and the network loss is 5MW; the total output of new energy in Scheme B is 90MW and the network loss is 4MW. If the priority is given to the consumption of new energy, then choose Scheme A; if the priority is to reduce network losses, then choose Scheme B.Finally, the new energy output command in the optimal dispatching scheme is sent to each new energy station, and then fed back to the power grid dispatching control system to guide the real-time safe and economical dispatching of the power grid, so as to achieve efficient consumption of new energy and safe and stable operation of the power grid. For example, according to the optimal scheme, the dispatching control system sends an output command to a wind farm, requiring its output to be 30MW. At the same time, the optimization results are displayed on the dispatcher's monitoring interface so that the dispatcher can understand the operation status of the power grid. This can effectively improve the utilization rate of new energy while ensuring the safe and stable operation of the power grid.

[0019] Step S102: For the set of renewable energy output solutions obtained in the previous step, a fuzzy comprehensive evaluation method is used to quantitatively evaluate the feasibility and economy of each solution. A multi-level fuzzy evaluation matrix is ​​constructed, and the evaluation indicators include power grid safety margin, renewable energy consumption rate, system economy, etc., and the weight of each indicator is determined in combination with expert scoring. The comprehensive score of each solution is calculated, and the optimal renewable energy output solution is selected according to the ranking results.

[0020] According to the set of new energy output schemes, multiple evaluation indicators including grid safety margin, new energy consumption rate, system economy, etc. are obtained. For each evaluation indicator, the importance weight of each indicator is obtained by expert scoring. According to the obtained evaluation indicators, a multi-level fuzzy evaluation matrix is ​​constructed. The elements in the fuzzy evaluation matrix represent the membership of each new energy output scheme on each evaluation indicator. For each new energy output scheme, its performance value on each evaluation indicator is converted into a membership value through the membership function and substituted into the corresponding position in the fuzzy evaluation matrix. The fuzzy comprehensive evaluation method is used to calculate the comprehensive score of each new energy output scheme in combination with the evaluation indicator weight. The calculation formula of fuzzy comprehensive evaluation is: B = A·R, where B is the comprehensive scoring matrix, A is the indicator weight matrix, and R is the fuzzy evaluation matrix. Arrange the elements in the comprehensive scoring matrix B in order from large to small to obtain the ranking result of the advantages and disadvantages of the new energy output scheme. Obtain the new energy output scheme with the highest comprehensive score in the sorting result and determine it as the optimal scheme.

[0021] Specifically, the goal of power grid security verification is to ensure that the power system can operate safely and stably under various operating conditions. After obtaining a set of renewable energy output solutions that meet safety constraints, it is necessary to further evaluate the pros and cons of these solutions and finally select the optimal solution. The criteria for evaluating the pros and cons of the solutions are multiple evaluation indicators, such as power grid safety margin, renewable energy consumption rate, and system economy. Power grid safety margin refers to the margin from the stability limit of the system when a disturbance occurs. The larger the margin, the stronger the system's ability to resist disturbances and the higher the safety. For example, assuming that the transmission limit of line 1 is 100MW, and the flow of solution A on line 1 is 80MW, then the safety margin of solution A on line 1 is 20MW. The renewable energy consumption rate refers to the proportion of the actual power consumption of renewable energy to the total power generation of renewable energy. The higher the consumption rate, the higher the utilization of renewable energy. For example, if the total installed capacity of wind farms in a certain area is 200MW, and the wind power output of solution A is 180MW, then the wind power consumption rate of solution A is 90%. System economy is usually measured by the total operating cost of the system or network loss. The lower the operating cost or the smaller the network loss, the better the economy of the system. For example, if the network loss of Scheme A is 5MW and the network loss of Scheme B is 4MW, then the economy of Scheme B is better than that of Scheme A. Different evaluation indicators have different importance to the final decision. In order to quantify this importance, expert scoring is usually used to assign a weight to each indicator. The larger the weight, the more important the indicator. For example, through expert scoring, the weights of the power grid safety margin, the new energy consumption rate and the system economy are determined to be 4, 3 and 3 respectively. In order to comprehensively consider multiple evaluation indicators, the fuzzy comprehensive evaluation method is adopted. First, a fuzzy evaluation matrix needs to be constructed. The elements of this matrix represent the membership of each scheme on each indicator. The membership degree indicates the performance of the scheme on the indicator, and the value range is 0 to 1. The larger the value, the better the performance. For example, assuming that the safety margin of Scheme A on line 1 is 20MW and the safety margin on line 2 is 10MW, the membership of Scheme A on the power grid safety margin indicator may be 8 (the specific value needs to be calculated by the membership function). The membership function is a function that converts the actual value of an indicator into a membership function. Different indicators can use different membership functions. For example, for the grid safety margin, a linear membership function can be used; for the new energy consumption rate, an S-type membership function can be used. Assuming that the membership of scheme A in the three indicators of grid safety margin, new energy consumption rate and system economy is 8, 9 and 7 respectively, the row of the fuzzy evaluation matrix corresponding to scheme A is [8,9,7]. Substituting the membership values ​​of all schemes into the fuzzy evaluation matrix, and then combining the indicator weights, the comprehensive score of each scheme can be calculated. The calculation formula for the comprehensive score is B = A·R, where B is the comprehensive score matrix, A is the indicator weight matrix, and R is the fuzzy evaluation matrix.For example, assuming that the indicator weight matrix A is [4,3,3], and one row of the fuzzy evaluation matrix corresponding to plan A is [8,9,7], then the comprehensive score of plan A is 4*8+3*9+3*7=79. Finally, the comprehensive scores of all plans are arranged in order from large to small to obtain the ranking results of the plans. The plan with the highest comprehensive score is the optimal plan. For example, assuming that the comprehensive score of plan A is 79 and the comprehensive score of plan B is 85, then plan B is better than plan A, and plan B is the optimal plan. After selecting the optimal plan, the new energy output command corresponding to the plan is issued to each new energy station to guide the real-time safe and economic dispatch of the power grid, realize the efficient consumption of new energy and the safe and stable operation of the power grid.

[0022] Step S103, introduce the power supply start-stop state variable into the safety verification model, and use the power supply start-stop characteristics as optimization constraints. Set the minimum duration and maximum start-stop number parameters of the power supply start-stop to limit frequent power supply start-stop. Use the mixed integer programming algorithm to solve the optimization model containing the power supply start-stop state variable, obtain the new energy output and power supply combination optimization plan considering the power supply start-stop, and further optimize the optimal plan screened in the second step.

[0023] According to the power supply start-stop state variable, the start-stop state information of the power supply is obtained. If the power supply is in the start state, it is determined whether its duration exceeds the minimum duration threshold. If it does not exceed the threshold, the constraint condition is included in the optimization model. If the power supply is in the stop state, the number of stops is counted to determine whether it exceeds the maximum number limit. If it does not exceed the limit, it is also included as a constraint condition in the optimization model. All constraints are integrated into the optimization model, including power supply start-stop state constraints, new energy consumption constraints, etc. The CPLEX solver is used to solve the optimization model through a mixed integer linear programming algorithm to obtain the optimal power supply combination scheme under the constraints and determine the output of each power supply. For the solved power supply combination scheme, the output of the new energy power supply is analyzed to determine whether it meets the new energy consumption requirements set in the optimization model. If it does not meet the requirements, the relevant parameters in the optimization model, such as the proportion of new energy output constraints, are adjusted, and then the solution is repeated. For the multiple power supply combination schemes that meet the constraints, a weighted scoring method is used to conduct a comprehensive evaluation from the aspects of power supply start-stop times, new energy output ratio, and load satisfaction rate. The scheme with the highest total score is selected as the optimal power supply combination scheme. According to the optimal solution, the actual start-stop status and output scheduling of each power source are controlled, and the optimization results of the power source start-stop status variables are applied to the actual operation control of the power source. During the operation of the power source, the actual start-stop status, duration, number of stops and other information of the power source are monitored every 15 minutes. If the number of starts and stops of a single power source exceeds 3 times within 1 hour, it is judged as frequent start and stop, triggering the re-solution of the optimization model. The actual operation effect of the optimization plan is evaluated monthly, and key operation indicators such as the amount of new energy consumption, the number of power source starts and stops, and the number of hours of unit utilization are counted. According to the evaluation results, the parameter settings in the optimization model are adjusted in a timely manner, such as adjusting the minimum duration threshold, the maximum number of starts and stops, etc., to continuously improve the optimization effect.

[0024] Specifically, the power supply start-stop state variable is used to describe the operating state of the power supply, for example, "1" represents start and "0" represents stop. By obtaining the values ​​of these variables, the real-time state of each power supply can be understood. For example, if the state variable of coal-fired unit A is "1", it means that the unit is running; if the state variable of wind turbine unit B is "0", it means that the unit is currently in a shutdown state. For the running power supply, its continuous running time needs to be paid attention to. For example, the minimum continuous running time threshold of coal-fired unit C is 2 hours. If unit C has just started, its continuous running time has not reached the threshold. At this time, it is necessary to add constraints in the optimization model to restrict it from shutting down within the next 2 hours. This is done to avoid frequent start-stop of the unit, reduce equipment loss, and extend service life. For the power supply in the stopped state, its stop times need to be counted. For example, the maximum allowed shutdown times of coal-fired unit D is 3 times a day. If unit D has stopped twice today, then constraints need to be added in the optimization model to restrict it from stopping again today. This is done to avoid exceeding the allowed shutdown times of the unit and ensure the reliability and stability of the equipment. Integrate all constraints such as power supply start-stop state constraints and new energy consumption constraints into the optimization model. For example, suppose that the power system in a certain area includes coal-fired units, gas-fired units, wind turbines, and photovoltaic units. Various constraints need to be considered, such as: the minimum continuous operation time and maximum shutdown times of coal-fired units; the ramp rate limit of gas-fired units; the output volatility and randomness of wind turbines and photovoltaic units; and the requirement to ensure that the consumption rate of new energy reaches a certain proportion. Use the CPLEX solver to solve the optimization model using a mixed integer linear programming algorithm. For example, the objective function can be set to minimize the total operating cost of the system, and considering the above constraints, the optimal power supply combination scheme can be obtained through the solver to determine the output of each power supply. Check the output of new energy in the solution results. For example, suppose that the optimization model requires that the new energy consumption rate is not less than 80%. If the solution shows that the new energy consumption rate is only 70%, it is necessary to adjust the model parameters, such as increasing the proportion of new energy output constraints, and then re-solve until the requirements are met. Comprehensively evaluate multiple power supply combination schemes that meet the constraints. Assume that three solutions A, B and C that meet the constraints are obtained. Weighted scoring can be performed based on the number of power supply starts and stops, the proportion of renewable energy output, and the load satisfaction rate. For example, solution A has the highest proportion of renewable energy output, but the number of starts and stops is also high; solution B has the least number of starts and stops, but the load satisfaction rate is slightly lower; and solution C has relatively balanced indicators. Through weighted scoring, the solution with the highest total score is finally selected as the optimal solution. The actual operation of the power supply is controlled according to the optimal solution. For example, if solution A is selected as the optimal solution, the actual output of the coal-fired unit, gas-fired unit, wind turbine unit, and photovoltaic generator unit is controlled according to the output instructions of each power supply in solution A. The actual operating status of the power supply is monitored regularly.For example, the actual start and stop status, duration, and number of stops of the power supply are monitored every 15 minutes. If the coal-fired unit E starts and stops more than 3 times within 1 hour, it is judged as frequent start and stop, and the optimization model needs to be re-solved to adjust the output plan. The actual operation effect of the optimization plan is evaluated every month. For example, key indicators such as the amount of new energy consumption, the number of power supply starts and stops, and the number of hours of unit utilization are counted. If it is found that the amount of new energy consumption is lower than expected, the parameter settings in the optimization model can be adjusted, such as lowering the minimum duration threshold or increasing the maximum number of starts and stops to increase the new energy consumption rate.

[0025] Step S104, construct a multi-time-scale collaborative optimization scheduling model for each link of source, grid, load and storage. Set the decision variables of new energy power prediction, load demand response, and energy storage charging and discharging plan, and use the distributed model predictive control method to achieve hierarchical distributed solution. The multiple time scales include three scales: day-ahead, intraday, and real-time, corresponding to the prediction periods of 24 hours, 1 hour, and 5 minutes, respectively. Through iterative calculation, the scheduling curves of each link are made consistent to obtain the optimized scheduling curves of each link.

[0026] The historical operation data and real-time monitoring data of each link of source, grid, load and storage are obtained. The corresponding data preprocessing methods are used for different time scales to clean and normalize the data, and feature selection and feature extraction are performed to construct appropriate feature vectors. According to the preprocessed data, the long short-term memory neural network (LSTM) algorithm is used to establish a multi-time scale new energy power prediction model and load demand prediction model. The deep Q network (DQN) algorithm in reinforcement learning is used to establish an energy storage charging and discharging optimization model, with the optimization goal of maximizing new energy consumption and system economy. The prediction models and optimization models are integrated into the collaborative optimization scheduling model, and the decision variables of each link of source, grid, load and storage are set, including new energy output plan, load response plan, energy storage charging and discharging plan, etc. The random optimization method based on the scenario tree is used to solve the scheduling model and generate multiple possible optimization scheduling schemes. The Monte Carlo simulation method is used to evaluate each scheduling scheme, and the optimal scheduling scheme is selected as the optimization scheduling curve of each link at different time scales according to indicators such as new energy consumption rate and system operation cost. According to the optimization scheduling results, source-grid-load-storage collaborative control instructions are formed, including new energy power limit instructions, demand response instructions, and energy storage charging and discharging instructions. The control instructions are issued to the control equipment in each link through the energy management system to coordinate the optimized operation of the source-grid-load-storage to improve the new energy consumption capacity and system operation economy. In the actual operation process, the operating status of the system is continuously monitored, and the real-time operation data of each link is obtained and compared with the optimized scheduling curve. When the actual operation deviates from the optimized scheduling curve by more than a certain threshold, a new round of collaborative optimization scheduling is triggered to ensure real-time optimization control of the system.

[0027] Specifically, the synergistic interaction of source, grid, load and storage is to maximize the absorption of new energy, while taking into account the safe and stable operation of the power grid and the electricity demand of users. Obtaining historical and real-time data of each link is the basis of collaborative optimization. For example, the photovoltaic power generation data, power load data, energy storage equipment charging and discharging data, and power grid topology of a certain area in the past year are obtained. Real-time monitoring data includes current photovoltaic output, load demand, energy storage status, etc. Data of different time scales require different preprocessing methods. For example, for short-term predictions (such as the next few hours), minute-level or hour-level data can be used; for long-term predictions (such as the next few days), daily-level data can be used. Data preprocessing includes operations such as cleaning abnormal data, filling missing values, and normalization. For example, the photovoltaic output data is normalized to between 0 and 1 to facilitate model training. Feature selection and extraction are to extract useful information for prediction and optimization from the original data. For example, extract features such as sunshine intensity, temperature, and humidity in photovoltaic output data, and features such as periodicity and trend in load data to form feature vectors. Long short-term memory neural network (LSTM) is suitable for processing time series data and can capture long-term dependencies in data. LSTM can be used to establish new energy power prediction models and load demand prediction models respectively. For example, a LSTM model can be trained using photovoltaic power generation data from the past year to predict photovoltaic output in the next few hours. Similarly, another LSTM model can be trained using historical load data to predict electricity demand in the next few hours. Deep Q network (DQN) is a reinforcement learning algorithm that can be used to optimize energy storage charging and discharging strategies. With the goal of maximizing new energy consumption and system economy, the DQN model can learn how to control the charging and discharging of energy storage devices under different circumstances. For example, when photovoltaic output is high and load is low, excess energy is stored in energy storage devices; when photovoltaic output is low and load is high, energy in energy storage devices is released to supplement power supply. The prediction model and optimization model are integrated into the collaborative optimization scheduling model to achieve overall optimization of source, grid, load and storage. The decision variables of the scheduling model include new energy output plan, load response plan, energy storage charging and discharging plan, etc. For example, according to the predicted photovoltaic output and load demand, a storage charging and discharging plan is formulated to maximize the use of photovoltaic energy and reduce dependence on traditional thermal power. The random optimization method based on the scenario tree can handle the uncertainty of the prediction. For example, for the photovoltaic output in the next 24 hours, multiple possible scenarios can be generated, each of which represents a possible photovoltaic output situation. Then optimize the scheduling for each scenario to obtain multiple possible scheduling schemes. The Monte Carlo simulation method can evaluate multiple scheduling schemes. For example, multiple simulations are performed for each scheduling scheme to calculate the average and variance of indicators such as the new energy consumption rate and the system operation cost. Based on these indicators, the optimal scheduling scheme is selected. The optimized scheduling results form the source-grid-load-storage collaborative control instructions.For example, according to the optimal scheduling scheme, a new energy power limit instruction is generated to control the output of the photovoltaic power station; a demand response instruction is generated to guide users to adjust their electricity consumption behavior; and an energy storage charging and discharging instruction is generated to control the charging and discharging of the energy storage device. The energy management system sends the control instructions to the control devices in each link. For example, the power limit instruction is sent to the inverter of the photovoltaic power station to control the photovoltaic output; the demand response instruction is sent to the user's smart terminal to guide the user to use electricity at off-peak times; and the charging and discharging instruction is sent to the control system of the energy storage device. The system operation status is continuously monitored, and the actual operation data is compared with the optimized scheduling curve. For example, data such as photovoltaic output, load demand, and energy storage status are collected every 15 minutes and compared with the optimized scheduling curve. When the actual operation deviates from the optimized scheduling curve by more than a certain threshold, for example, the photovoltaic output is lower than expected, a new round of collaborative optimization scheduling is triggered, and the control instructions are regenerated to ensure real-time optimization control of the system. This can respond to various uncertain factors in a timely manner, such as weather changes and load fluctuations, to ensure the stable operation of the system and the efficient use of new energy.

[0028] Step S105, in the collaborative optimization scheduling process, a communication mechanism based on the MQTT protocol is designed to realize information interaction between various links. The Kafka message queue is used to cache and sort the scheduling instructions by timestamp to ensure the real-time and consistency of the scheduling instructions. The Zookeeper distributed coordination service is deployed to realize the state synchronization and failover of each node. This communication mechanism is combined with the distributed model predictive control method in the fourth step to ensure real-time data interaction and coordinated control between various links.

[0029] The real-time status information of each link is published to the Kafka message queue through the MQTT protocol. While the message is being published, Zookeeper monitors the status of the nodes in each link. If a node fails, Zookeeper transfers the task information of the failed node to other available nodes and notifies other nodes to take over the task through the MQTT protocol. Consume the status information from the Kafka message queue, sort the messages using timestamps to ensure the timing of the scheduling instructions. The sorted scheduling instructions contain information such as target values ​​and constraints, and are stored in JSON format and distributed to each link through the MQTT protocol. After receiving the scheduling instructions, each link uses the pre-configured genetic algorithm to perform optimized scheduling, and publishes the scheduling results to Kafka in JSON format through the MQTT protocol. The distributed ADMM algorithm updates the model parameters based on the real-time scheduling results in Kafka, and dynamically adjusts the parameters in the optimized scheduling strategy, such as the population size, crossover probability, and mutation probability of the genetic algorithm. The updated scheduling strategy parameters are distributed to each link in JSON format through the MQTT protocol. Each link re-executes the genetic algorithm for optimized scheduling based on the updated scheduling strategy parameters, forming a closed-loop feedback and continuously optimizing the collaborative scheduling effect.

[0030] Specifically, the MQTT protocol is a lightweight message transmission protocol suitable for bandwidth-sensitive scenarios such as the Internet of Things. It adopts a publish / subscribe mode to achieve decoupling and asynchronous communication between devices. In this solution, the status information of each link of the source, grid, load and storage, such as photovoltaic output, load demand, energy storage SOC, etc., can be published to the Kafka message queue through the MQTT protocol. For example, a photovoltaic inverter can publish the current output data in the form of an MQTT message to a topic named "pv_output" every 5 minutes. Kafka is a distributed stream processing platform with the characteristics of high throughput, low latency, and scalability. It can be used as a message queue to store and forward status information of each link. Zookeeper is a distributed coordination service that can be used to monitor the status of nodes in each link. If a node fails, such as a photovoltaic inverter disconnection, Zookeeper can transfer the task information of the node, such as data collection and reporting tasks, to other available nodes. At the same time, Zookeeper can also notify other nodes to take over tasks through the MQTT protocol to ensure the stable operation of the system. For example, if an energy storage unit fails, Zookeeper will transfer control to the backup energy storage unit and notify the energy management system through MQTT. When consuming status information from the Kafka message queue, you need to use timestamps to sort the messages to ensure the timing of the scheduling instructions. This is because the execution order of the scheduling instructions is very important. For example, the energy storage charging instruction must be issued first, and then the load adjustment instruction. The sorted scheduling instructions contain information such as target values, constraints, etc., and are stored in JSON format. For example, a scheduling instruction can be: {"device":"ESS1","action":"charge",

[0031] "target_soc":90, "start_time": "2024-10-2710:00:00", "end_time": "2024-10-2712:00:00"}, indicating that the energy storage unit ESS1 is required to charge to 90% SOC between 10 and 12 o'clock. These instructions are distributed to each link through the MQTT protocol. After receiving the scheduling instructions, each link uses a pre-configured genetic algorithm to perform optimal scheduling. Genetic algorithm is a global optimization algorithm that can be used to solve complex nonlinear optimization problems. In this solution, each link can use genetic algorithms to optimize its own operating strategy based on the received scheduling instructions. For example, the energy storage unit can optimize the charging and discharging strategy, and the load can optimize the power consumption strategy. The scheduling results are published to Kafka in JSON format through the MQTT protocol. For example, after the energy storage unit executes the genetic algorithm, the optimized charging and discharging plan can be published to Kafka in JSON format. The distributed ADMM (Alternating Direction Method of Multipliers) algorithm is a distributed optimization algorithm that can be used to solve large-scale optimization problems. In this solution, the distributed ADMM algorithm updates the model parameters according to the real-time scheduling results in Kafka, and dynamically adjusts the parameters in the optimization scheduling strategy, such as the population size, crossover probability, and mutation probability of the genetic algorithm. This allows the genetic algorithm to better adapt to the real-time changing system state and improve the optimization effect. The updated scheduling strategy parameters are distributed to each link in JSON format through the MQTT protocol. For example, the ADMM algorithm may adjust the population size of the genetic algorithm to 200 and the crossover probability to 8 according to the current photovoltaic output to speed up the convergence of the algorithm. Each link re-executes the genetic algorithm for optimization scheduling based on the updated scheduling strategy parameters, forming a closed-loop feedback and continuously optimizing the collaborative scheduling effect. For example, after receiving the updated genetic algorithm parameters, the energy storage unit will re-run the genetic algorithm, generate a new charging and discharging plan, and publish it to Kafka. This closed-loop feedback mechanism allows the system to continuously adapt to new situations and improve the overall performance of the system.

[0032] Step S106: Based on the optimized new energy output scheme and the source-grid-load-storage coordinated dispatch curve, the model predictive control algorithm is used to calculate the active power, reactive power and voltage regulation instructions of each power station in real time. The control instructions are sent to each new energy power station and intelligent control equipment through the power communication network. Intelligent converters and reactive power compensation devices are used to respond to dispatch instructions to achieve smooth control of new energy output and voltage reactive power optimization.

[0033] (1) Establish mathematical models of new energy power stations, power grids and loads, including generator models, converter models, line models, load models, etc.

[0034] (2) Based on the current system status and optimization goals, predict the system state changes in the future.

[0035] (3) Add various constraints to the prediction model, such as voltage amplitude constraints, line flow constraints, generator output constraints, etc.

[0036] (4) Solve the optimization problem and obtain the optimal active power, reactive power and voltage regulation instructions for each power station in the future.

[0037] (5) The optimization result at the current moment is sent as a control instruction to the new energy power station and intelligent control equipment.

[0038] (6) Based on the real-time feedback information from the system, carry out rolling optimization and continuously revise the optimization results.

[0039] The active power, reactive power and voltage regulation instructions calculated by the model predictive control algorithm are sent to the corresponding new energy power station and intelligent control equipment through the power communication network, including: (1) The intelligent converter on the new energy power station side adjusts the output of the new energy generator set according to the received active power instruction to achieve smooth control of the new energy output. (2) The reactive power compensation device on the new energy power station side adjusts the reactive power compensation capacity according to the received reactive power instruction to optimize the reactive power of the new energy grid connection point.

[0040] (3) The intelligent converter and reactive power compensation device on the new energy power station side coordinately adjust the voltage level of the new energy grid connection point according to the received voltage regulation instructions to achieve voltage optimization control. The real-time operating condition data of the new energy power station, including active power, reactive power, voltage, etc., are fed back to the source-grid-load-storage coordinated dispatching system through the power communication network. The source-grid-load-storage coordinated dispatching system uses machine learning algorithms such as support vector machines or neural networks based on the real-time data fed back to predict the new energy output and load demand, and generates an optimized new energy output plan and coordinated dispatching curve to guide the source-grid-load-storage coordinated optimization dispatching in the next period of time, forming a closed-loop optimization control.

[0041] Specifically, the mathematical models of renewable energy power stations, power grids and loads are the basis for coordinated dispatching. The generator set model describes the output characteristics of the generator. For example, a quadratic function can be used to represent the relationship between the fuel cost and output of a thermal power generator set. The output corresponding to the lowest fuel cost point is the economic output point of the unit. The inverter model describes the power conversion process of renewable energy sources such as photovoltaics and wind power. For example, an efficiency curve can be used to represent the relationship between the energy conversion efficiency and input power of a photovoltaic inverter, and the power point with the highest efficiency can be found. The line model describes the electrical characteristics of the transmission line. For example, a π-type equivalent circuit can be used to represent the impedance and capacitance parameters of the transmission line, and used to calculate line losses and power flow distribution. The load model describes the user's electricity consumption characteristics. For example, a time series can be used to represent the load demand in a certain area at different times of the day, which is used to predict future load change trends. Predicting the state changes of the system in the future is crucial to scheduling decisions. For example, assuming that the predicted photovoltaic output value of a certain area in the next 24 hours is:

[0042] [10,15,20,25,30,35,40,45,50,55,60,65,60,55,50,45,40,35,30,25,20,15,10,5]MW. The load forecast value is: [40,45,50,55,60,65,70,75,80,85,90,95,90,85,80,75,70,65,60,55,50,45,40,35]MW. Based on these forecast values, the supply and demand gap in the next 24 hours can be estimated, so as to formulate the corresponding dispatch strategy. Various constraints need to be considered in the forecast model to ensure the safe and stable operation of the system. For example, the voltage amplitude constraint requires that the voltage amplitude of all nodes in the system must be within the allowable range, such as between 95pu and 05pu. The line flow constraint requires that the flow of all transmission lines must be less than its maximum transmission capacity. The output constraint of the generator set requires that the output of all generator sets must be within its allowed range. These constraints can ensure that the system operates in a safe and stable state. By solving the optimization problem, the optimal output instructions of each power station in the future can be obtained. For example, based on the predicted photovoltaic output, load demand and various constraints, the optimal output value of each thermal power generator set in the next 24 hours can be calculated, for example:

[0043] [10,10,5,0,0,0,0,0,0,0,0,0,5,10,15,20,25,30,35,40,40,40,40,40]MW. This shows that during the period of high photovoltaic output, the thermal power generating set can reduce its output, thereby saving fuel costs. The optimization results at the current moment are sent to each power station. For example, the target output value of the thermal power generating set in the current hour is sent to the control system of the power plant through the power communication network. The control system adjusts the excitation current and the speed regulating valve opening of the generating set according to the instructions to achieve precise control of the output of the generating set. According to the real-time feedback information of the system, rolling optimization is carried out to continuously correct the optimization results. For example, if the actual photovoltaic output is lower than the predicted value, the output of the thermal power generating set needs to be increased to make up for the supply and demand gap. By continuously collecting real-time data from the system and feeding it back to the optimization model, dynamic adjustment of the scheduling plan can be achieved, thereby improving the operating efficiency and reliability of the system. The control instructions calculated by the model predictive control algorithm are sent to the corresponding new energy power station and intelligent control equipment through the power communication network. For example, if the smart converter of a photovoltaic power station receives an active power instruction of 30MW, it will adjust the output power of the photovoltaic array according to the instruction. If the reactive power compensation device receives a reactive power instruction of 5MVar, it will adjust the switching of the compensation capacitor to provide the required reactive power. If the voltage regulation instruction is 0pu, the smart converter and the reactive compensation device will work together to maintain the grid connection point voltage at 0pu. The real-time operation data of the new energy power station is fed back to the coordinated dispatching system through the power communication network. For example, the photovoltaic power station uploads data such as active power, reactive power, and voltage to the dispatching system every 5 minutes. These data are used for model prediction and optimized dispatching to form a closed-loop control. The dispatching system uses machine learning algorithms to predict the output of new energy and load demand. For example, it uses support vector machines to predict the photovoltaic output in the next 24 hours, and generates optimized new energy output plans and coordinated dispatching curves based on the prediction results. This can improve the prediction accuracy, thereby better guiding the coordinated dispatching, forming a closed-loop optimization control, and continuously improving the overall operation efficiency of the system.

[0044] Step S107, deploy a big data platform to collect real-time data on new energy power, load demand, and energy storage status. Use online machine learning algorithms to continuously optimize new energy power prediction models and load demand response models. Compare actual operating data with the dispatch plan, and use a policy gradient-based reinforcement learning algorithm to autonomously learn and adjust the source-grid-load-storage collaborative optimization strategy. The learning goal is set to maximize the new energy absorption rate and minimize the system operating cost. By continuously iterating and optimizing decision-making strategies, the new energy absorption capacity and user-side response quality are improved to achieve global optimization and adaptive control.

[0045] Based on the real-time data of new energy power, load demand and energy storage status collected by the big data platform, a source-grid-load-storage collaborative optimization model is constructed, taking into account multiple optimization objectives such as new energy consumption and operating costs. Using historical operation data, through algorithms such as multivariate linear regression, a new energy power prediction model and a load demand response model are established, and the model parameters are continuously optimized to improve the accuracy of prediction and response. On the basis of the source-grid-load-storage collaborative optimization model, a deep reinforcement learning algorithm (such as DQN, DDPG, etc.) is used to take the state of the source-grid-load-storage system as the environmental state, the optimization decision as the intelligent action, and the new energy consumption rate and operating cost as the reward function. Through continuous interaction with the environment, the optimization strategy is learned and adjusted to generate the optimal source-grid-load-storage collaborative scheduling strategy. During the scheduling execution process, the operating status and key parameters of each control unit are continuously monitored, the actual operating data is compared with the scheduling plan, and the deviation value is calculated. A reasonable deviation threshold is set. When the deviation exceeds the threshold, the reinforcement learning algorithm is triggered to retrain, adjust the optimization strategy, correct the scheduling plan, and achieve adaptive control. According to the source-grid-load-storage coordination strategy optimized by the reinforcement learning algorithm, a new dispatching plan is generated, and the dispatching instructions are sent to each control unit to coordinate the operation of new energy, power grid, load and energy storage. If serious deviations or abnormal conditions are found during the execution process, the emergency control plan is activated and the safe operation mode is switched to ensure the stable operation of the power grid. The execution effect of the source-grid-load-storage coordinated dispatch is evaluated, and key performance indicators such as the new energy consumption rate and operating cost are calculated. The evaluation results are compared with the dispatching strategy and optimization objectives to calculate the optimization effect and improvement space of the strategy. The evaluation results are used as new sample data and fed back to the load forecasting model and reinforcement learning algorithm. Through continuous training and learning, the performance of the source-grid-load-storage coordinated optimization is continuously improved to achieve closed-loop iterative optimization of the strategy.

[0046] Specifically, the big data platform collects operating data such as new energy power, load demand and energy storage status in real time, which is the basis for realizing the coordinated optimization of source, grid, load and storage. For example, the real-time power data of photovoltaic power stations, wind farms, energy storage systems and user loads in a certain area, as well as the power flow, voltage and other data of the power grid, will be collected on the big data platform. After cleaning and processing, these data are used to build a coordinated optimization model, as well as train prediction models and reinforcement learning algorithms. The source, grid, load and storage coordinated optimization model needs to comprehensively consider multiple optimization goals, such as maximizing new energy consumption, minimizing operating costs, and grid stability. In order to achieve these goals, the model needs to consider various constraints, such as the volatility of new energy output, changes in load demand, the charging and discharging characteristics of the energy storage system, and the transmission capacity of the power grid. Using historical operating data, a new energy power prediction model and a load demand response model can be constructed. For example, a multivariate linear regression algorithm can be used to predict the new energy output and load demand in the future based on meteorological data such as light intensity, wind speed, temperature, etc. in the past period of time, as well as historical load data. By continuously updating historical data and adjusting model parameters, the accuracy of the prediction can be improved. For example, the initial model may only consider the impact of meteorological factors on the output of renewable energy, but later it was found that the user's electricity consumption behavior also has certain regularity, so the user behavior data was also added to the model to improve the prediction accuracy. Deep reinforcement learning algorithms can be used to optimize the source-grid-load-storage collaborative scheduling strategy. For example, the DQN algorithm can be used to take the state of the source-grid-load-storage system (such as renewable energy output, load demand, energy storage state, etc.) as the environmental state, the scheduling decision (such as energy storage charging and discharging power, renewable energy output adjustment, etc.) as the agent action, and the renewable energy consumption rate and operating cost as the reward function. By continuously interacting with the environment, the agent can learn the optimal scheduling strategy. For example, the agent may learn to store excess energy in the energy storage system during periods of high photovoltaic output, and release the energy of the energy storage system during peak load periods to meet load demand and increase the renewable energy consumption rate. During the scheduling execution process, it is necessary to continuously monitor the operating status and key parameters of each control unit, and compare the actual operating data with the scheduling plan. For example, if the actual photovoltaic output is lower than the predicted value, or the load demand is higher than the predicted value, the scheduling plan needs to be adjusted, such as increasing the output of thermal generators or reducing the discharge power of the energy storage system. When the deviation exceeds the preset threshold, it is necessary to trigger the reinforcement learning algorithm to retrain to adapt to the new operating environment. For example, if the actual photovoltaic output is lower than the predicted value for several consecutive days, it may be necessary to adjust the parameters of the photovoltaic output prediction model or retrain the reinforcement learning algorithm. After the scheduling instructions are sent to each control unit, it is necessary to coordinate the operation of new energy, power grid, load and energy storage.For example, the energy storage system needs to perform charging and discharging operations according to the dispatching instructions, the new energy power station needs to adjust the output according to the dispatching instructions, and the load-side users need to respond to the demand according to the dispatching instructions. If serious deviations or abnormal conditions are found during the execution process, such as the grid voltage or frequency exceeds the safety range, it is necessary to start the emergency control plan, such as cutting off part of the load or starting the backup power supply. The execution effect of the source-grid-load-storage coordinated dispatch can be evaluated to calculate key performance indicators such as the new energy consumption rate and operating cost. For example, the consumption ratio of new energy power generation in one day and the total operating cost of the system can be calculated. By comparing the evaluation results with the dispatching strategy and optimization objectives, the effectiveness and optimization space of the dispatching strategy can be judged. The evaluation results are used as new sample data and fed back to the load forecasting model and reinforcement learning algorithm. Through continuous training and learning, the performance of the source-grid-load-storage coordinated optimization is continuously improved to achieve closed-loop iterative optimization of the strategy. For example, if the evaluation results show that the new energy consumption rate is lower than the expected target, it is necessary to analyze the reasons and improve the dispatching strategy or prediction model, such as improving the algorithm or parameters of the prediction model, or adjusting the reward function of the reinforcement learning algorithm.

[0047] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

Claims

1. A method for optimizing user electricity consumption by coordinating the output of a new energy power supply cluster, characterized in that: include: Obtain the physical structure and real-time operation status data of the power grid, build a safety verification model, and obtain a set of new energy output solutions; The fuzzy comprehensive evaluation method is used to evaluate the set of new energy output schemes and select the optimal new energy output scheme; Introducing power supply start-stop state variables into the safety verification model to obtain a new energy output and power supply combination optimization solution considering power supply start-stop; Construct a multi-time-scale source-grid-load-storage coordinated optimization dispatching model to obtain the optimized dispatching curve of each link; Design a communication mechanism based on the MQTT protocol to realize information interaction between various links and ensure real-time data interaction and coordinated control between various links; According to the optimized renewable energy output plan and source-grid-load-storage coordinated dispatch curve, the active power, reactive power and voltage regulation instructions of each power station are calculated to achieve smooth control of renewable energy output and voltage-reactive power optimization; Deploy a big data platform, use online machine learning algorithms to continuously optimize the new energy power prediction model and load demand response model, and adopt reinforcement learning algorithms to autonomously learn and adjust the source-grid-load-storage collaborative optimization strategy to maximize the new energy absorption rate and minimize the system operating costs.

2. The user power optimization method of the new energy power supply cluster output coordination according to claim 1 is characterized in that: The method of obtaining the physical structure and real-time operation status data of the power grid, building a safety verification model, and obtaining a set of new energy output solutions includes: Obtain line flow, voltage amplitude and equipment capacity constraints based on the physical structure and real-time operation status data of the power grid; The penalty function method is used to transform the constraint conditions into a part of the objective function and construct a safety verification model; The objective function is solved using an improved particle swarm optimization algorithm to obtain a set of new energy output solutions that meet safety constraints.

3. The user power optimization method of the new energy power supply cluster output coordination as claimed in claim 1, characterized in that: The fuzzy comprehensive evaluation method is used to evaluate the set of new energy output solutions and select the optimal new energy output solution, including: According to the set of new energy output plans, obtaining evaluation indicators of power grid safety margin, new energy consumption rate, and system economy; Construct a multi-level fuzzy evaluation matrix and calculate the comprehensive score of each solution; The optimal new energy output plan is selected according to the comprehensive scoring and sorting results.

4. The user power optimization method of the new energy power supply cluster output coordination according to claim 1 is characterized in that: The power supply start-stop state variable is introduced into the safety verification model to obtain a new energy output and power supply combination optimization solution considering power supply start-stop, including: Set the minimum duration and maximum number of power on / off times to limit frequent power on / off. The mixed integer programming algorithm is used to solve the optimization model including power supply start-stop state variables, and the optimization scheme of new energy output and power supply combination considering power supply start-stop is obtained.

5. The user power optimization method of the new energy power supply cluster output coordination according to claim 1 is characterized in that: The multi-time-scale source-grid-load-storage coordinated optimization scheduling model is constructed to obtain the optimization scheduling curve of each link, including: Set decision variables for new energy power forecasting, load demand response, and energy storage charging and discharging plan; The distributed model predictive control method is used to realize hierarchical distributed solution. The scheduling curves of each link are made consistent through iterative calculation to obtain the optimized scheduling curves of each link.

6. The user power optimization method of the new energy power supply cluster output coordination according to claim 1 is characterized in that: The design uses the MQTT protocol as the communication mechanism to achieve information interaction between various links, ensuring real-time data interaction and coordinated control between various links, including: Use Kafka message queue to cache and sort scheduling instructions by timestamp; Deploy Zookeeper distributed coordination service to achieve state synchronization and failover of each node.

7. The user power optimization method of the new energy power supply cluster output coordination according to claim 1 is characterized in that: The active power, reactive power and voltage regulation instructions of each power station are calculated according to the optimized new energy output scheme and the source-grid-load-storage coordinated dispatch curve to achieve smooth control of new energy output and voltage-reactive power optimization, including: The model predictive control algorithm is used to calculate the active power, reactive power and voltage regulation instructions of each power station in real time; Control instructions are sent to each renewable energy power station and intelligent control equipment through the power communication network, and intelligent converters and reactive power compensation devices are used to respond to dispatch instructions.

8. The user power optimization method of the new energy power supply cluster output coordination as claimed in claim 7, characterized in that: The model predictive control algorithm is used to calculate the active power, reactive power and voltage regulation instructions of each power station in real time, including: Establish mathematical models of new energy power stations, power grids and loads, including generator models, converter models, line models and load models; Based on the current system status and optimization goals, predict the system status changes in the future; Add various constraints to the prediction model, such as voltage amplitude constraints, line flow constraints, and generator output constraints; Solve the optimization problem and obtain the optimal active power, reactive power and voltage regulation instructions for each power station in the future; Send the optimization results at the current moment as control instructions to new energy power stations and intelligent control equipment; Based on the real-time feedback information from the system, rolling optimization is carried out and the optimization results are continuously corrected.

9. The user power optimization method of the new energy power supply cluster output coordination as claimed in claim 7, characterized in that: The control instructions are sent to each new energy power station and intelligent control device through the power communication network, including: The intelligent converter on the renewable energy power station side adjusts the output of the renewable energy generator set according to the received active power command; The reactive power compensation device on the new energy power station side adjusts the reactive power compensation capacity according to the reactive power command received; The smart converter and reactive power compensation device on the renewable energy power station side coordinately adjust the voltage level of the renewable energy grid connection point according to the received voltage regulation instructions; The real-time operating data of the new energy power station, including active power, reactive power and voltage, is fed back to the source-grid-load-storage coordinated dispatching system through the power communication network.

10. The user power optimization method of the new energy power supply cluster output coordination according to claim 7, characterized in that: The use of intelligent converters and reactive power compensation devices to respond to dispatch instructions refers to the source-grid-load-storage coordinated dispatch system using machine learning algorithms to predict renewable energy output and load demand based on real-time data feedback, generate optimized renewable energy output plans and coordinated dispatch curves, which are used to guide the source-grid-load-storage coordinated optimization dispatch in the next period of time, forming a closed-loop optimization control.

Images