Resource scheduling system and method based on smart power grid

Abstract

The invention discloses a resource scheduling system and method based on a smart power grid. The system comprises a power grid cloud edge collaborative data acquisition module, a double-layer time scale random scheduling modeling module, a random saddle point approximate optimization calculation module, a high-dimensional target Pareto scheduling analysis module, a resource scheduling instruction generation module and a scheduling effect feedback adjustment module. The method comprises the steps of collecting multi-dimensional data of a power grid through cloud edge cooperation, constructing a double-layer time scale random scheduling model, calculating a scheduling decision variable through a random saddle point approximate optimization algorithm, screening an optimal scheme through a high-dimensional target Pareto scheduling model, converting the optimal scheme into a scheduling instruction, and issuing and executing the scheduling instruction. And parameters of each module are adjusted according to actual operation data feedback. According to the method, scheduling cooperation of different time scales is realized, power grid scheduling economy, reliability and environmental protection are considered, resource scheduling precision and efficiency are improved, and a complex operation scene of a smart power grid is effectively adapted.

Description

Technical Field

[0001] This invention relates to the field of power grid resource scheduling technology, and in particular to a resource scheduling system and method based on smart grids. Background Technology

[0002] As the scale of smart grids continues to expand, the large-scale integration of distributed power sources, energy storage devices, and diverse loads has led to a surge in multi-source heterogeneous data and dynamic fluctuations in operating status, placing higher demands on the real-time performance, economy, and reliability of resource scheduling. Currently, the power grid needs to simultaneously handle scheduling demands at different time scales, from hourly to minute-level, while also considering power generation cost control, power supply reliability, and environmental protection goals. Traditional methods relying on single scheduling models and centralized data processing are no longer suitable for complex scenarios. Therefore, leveraging a cloud-edge collaborative architecture to achieve efficient multi-dimensional data collection and transmission, combined with multi-objective optimization models to improve the scientific nature of scheduling decisions, has become an important development direction in the field of smart grid resource scheduling. There is an urgent need to build a comprehensive scheduling system that integrates multi-scale scheduling, multi-algorithm optimization, and multi-objective analysis.

[0003] Existing technologies for smart grid resource scheduling have two significant drawbacks: First, existing scheduling systems mostly use single-time-scale modeling, failing to achieve deep coupling and coordination between hourly and minute-level scheduling. This makes it difficult for scheduling schemes to simultaneously adapt to the grid's long-term planning and short-term dynamic adjustment needs, easily leading to a disconnect between scheduling instructions and the actual operating state of the grid, and failing to effectively cope with rapid fluctuations in distributed power output and load demand. Second, existing technologies lack systematic integration of cloud-edge collaborative architecture and multi-objective optimization models. Dynamic adjustment mechanisms are not established during data acquisition and transmission, and the stability verification of solutions is not fully considered during multi-objective scheduling analysis. This makes it difficult to achieve an optimal balance between economy, reliability, and environmental protection in scheduling decisions, while also resulting in insufficient data interaction efficiency and the actual executability of scheduling schemes. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides a resource scheduling system and method based on smart grids.

[0005] The technical solution adopted in this invention is a resource scheduling system based on a smart grid, comprising: a grid-cloud-edge collaborative data acquisition module that collects distributed power output data, load demand data, transmission line impedance data, and energy storage device charging and discharging status data in the smart grid through a data interaction channel between edge nodes and a cloud server, and transmits the collected multi-dimensional data to a two-layer time-scale stochastic scheduling modeling module; the two-layer time-scale stochastic scheduling modeling module constructs a stochastic scheduling model including intraday hourly scheduling sub-scales and intraday minute-scale scheduling sub-scales based on the received multi-dimensional data, and transmits the model parameters to a stochastic saddle point approximation optimization calculation module; the stochastic saddle point approximation optimization calculation module uses a stochastic saddle point approximation optimization algorithm to optimize the received model parameters, and outputs the optimized scheduling decision variables to a high-dimensional objective Pareto scheduling analysis module; the high-dimensional objective Pareto scheduling analysis module is based on... The received scheduling decision variables, combined with the economic, reliability, and environmental objectives of smart grid resource scheduling, are used to construct a high-dimensional objective Pareto scheduling model and perform non-dominated ranking analysis. The Pareto optimal scheduling scheme is then output to the resource scheduling instruction generation module. The resource scheduling instruction generation module converts the Pareto optimal scheduling scheme into specific scheduling instructions for distributed power sources, load nodes, energy storage devices, and transmission lines, and transmits them to the smart grid execution unit. The scheduling effect feedback and adjustment module collects the actual operating data of the smart grid execution unit and compares it with the expected data corresponding to the scheduling instructions output by the resource scheduling instruction generation module. The deviation data is then fed back to the grid-cloud-edge collaborative data acquisition module, the two-layer time-scale stochastic scheduling modeling module, the stochastic saddle point approximate optimization calculation module, the high-dimensional objective Pareto scheduling analysis module, and the resource scheduling instruction generation module for dynamic adjustment of the parameters of each module.

[0006] Furthermore, the expression for the two-layer time-scale stochastic scheduling model constructed by the two-layer time-scale stochastic scheduling modeling module is as follows: The first hourly scheduling subscale within the day Hourly scheduling decision variables This represents the total number of time periods for hourly scheduling within a day. This indicates the first minute-level scheduling subscale within the day. Minute-level scheduling decision variables For the first Total number of time periods scheduled per minute within an hour; Indicates the first The probability of such scenarios, Total number of scenes; The cost function representing the hourly scheduling subscale includes the distributed generation cost coefficient and the transmission loss cost coefficient. The cost function representing the minute-level scheduling subscale includes the energy storage device charging and discharging cost coefficient and the load reduction penalty cost coefficient; Indicates the first Hour Time weighting coefficient for minutes; The coordination coefficient represents the two-level time-scale scheduling decision variable; Indicates the first Scenario, First Under hourly conditions The scheduling decision variable for each minute.

[0007] Furthermore, the random saddle point approximation optimization algorithm used by the random saddle point approximation optimization calculation module is expressed as follows: ,in, This represents the minimization variable in the optimization problem, corresponding to the power output allocation variable in smart grid resource scheduling. for The feasible domain; This represents the maximization variable in the optimization problem, corresponding to the load demand allocation variable in smart grid resource scheduling. for The feasible domain; Represents a random variable The mathematical expectation, These are random parameters in a smart grid, including random fluctuations in distributed generation output and random fluctuations in load demand. for The probability distribution; The objective function for random saddle point optimization includes the matching coefficient between power output and load demand, and the transmission line capacity constraint coefficient. They represent the minimization variables respectively. With maximizing variables The regularization coefficient; They represent the first During the next iteration and The value of .

[0008] Furthermore, the expression for the high-dimensional objective Pareto scheduling model constructed by the high-dimensional objective Pareto scheduling analysis module is as follows: ,in, The decision variables representing high-dimensional objective Pareto scheduling include distributed power generation output, energy storage device charging and discharging power, and load transfer amount; This represents the economic objective function. Cost per unit of electricity generated, for Power output at all times The charging and discharging cost of energy storage units, for Energy storage charging and discharging power at all times, To reduce costs per unit load, for Load reduction amount at any time; Represents the reliability objective function. for Energy storage and discharge power at all times for Constant load demand, It is a very small positive number; This represents the environmental protection objective function. Emissions per unit of electricity generated The equivalent emissions reduction per unit load; These are the minimum and maximum constraints for the power output, respectively. These are the minimum and maximum constraints for load demand, respectively. The apparent power of the transmission line. This is a constraint on the maximum apparent power of transmission lines.

[0009] Furthermore, the data acquisition and transmission model expression of the power grid cloud-edge collaborative data acquisition module is as follows: ,in, express Aggregated data received in real time from the cloud; Indicates the total number of edge nodes; express Time of the first The data weight coefficient of each edge node is related to the communication bandwidth and data integrity of the edge node; express Time of the first Raw data collected by each edge node, including distributed power output and load demand data; express Time of the first The data noise coefficient of each edge node; express Time of the first Noise data for each edge node; This represents matrix multiplication. express Time of the first A data correction matrix for each edge node is used to eliminate systematic errors during the data acquisition process.

[0010] Furthermore, the instruction conversion model expression of the resource scheduling instruction generation module is as follows: ,in, express At any time for the first Scheduling instructions for each execution unit; This represents the total number of smart grid execution units, including distributed power sources, energy storage devices, and load nodes. Indicates the first The instruction translation coefficient matrix of each execution unit is related to the type and rated parameters of the execution unit; express At any time for the first Pareto optimal scheduling parameters for each execution unit; Indicates the first Feedback adjustment coefficient matrix for each execution unit; express Time of the first Feedback data from each execution unit, including actual operating power, voltage, and current data; This represents the Hadamard product operation; express Time of the first The state coefficient of each execution unit is set to 1 when the execution unit operates normally and 0.5 when the execution unit operates at a reduced rate.

[0011] Furthermore, the high-dimensional objective Pareto scheduling analysis module includes a Pareto solution generation unit, a non-dominated sorting calculation unit, an optimal solution screening unit, and a solution stability verification unit. The Pareto solution generation unit receives scheduling decision variables, combines them with economic, reliability, and environmental protection objective parameters, and generates an initial Pareto solution set through multi-objective optimization iteration under constraints. The solution set is then substituted into the objective function to calculate and exclude solutions with hyperparameter boundaries. The non-dominated sorting calculation unit determines the dominance relationship of each solution in the initial solution set, compares the objective function values ​​to divide the non-dominated levels, calculates the congestion distance, and considers the distance differences between adjacent solutions in each objective dimension. The optimal solution screening unit selects candidate optimal solutions based on the sorting results and congestion distance, and, combined with transmission line load rate and remaining energy storage capacity verification, eliminates solutions that do not meet real-time constraints. The solution stability verification unit substitutes the optimal solution into the model, changes the power output and load demand fluctuations, observes the change in the objective function value, and returns to regenerate the solution set if the value exceeds a threshold.

[0012] Furthermore, the power grid cloud-edge collaborative data acquisition module includes an edge data acquisition unit, a data preprocessing unit, a cloud data receiving unit, and a data interaction control unit. The edge data acquisition unit collects power output, load demand, line current and voltage, and energy storage charging and discharging status data at preset intervals through sensors and smart meters. The data preprocessing unit uses the statistical deviation method to identify abnormal values ​​in the original data, and converts the heterogeneous data into a unified format through adjacent data interpolation and replacement. The cloud data receiving unit receives standardized data through an encrypted link, calculates checksums and compares the data sent from the transmitting end, and requests retransmission if there is a discrepancy. The data interaction control unit dynamically adjusts the data transmission frequency and priority according to the communication status of the edge nodes and the cloud load. When the bandwidth is insufficient, the frequency of non-calibrated data is reduced, and when the load is high, calibrated data is received first.

[0013] Furthermore, the dual-layer time-scale stochastic scheduling modeling module includes an hourly model building unit, a minute-level model building unit, a scale coordination unit, and a model parameter update unit. The hourly model building unit determines the hourly scheduling segments, load and power output prediction values ​​based on historical and real-time data, and constructs a model that includes power output and line capacity constraints with the goal of minimizing operating costs. The minute-level model building unit divides each hour into minute-level time periods, collects high-frequency fluctuation data of load and power output, and constructs a model that aims to track the hourly plan while considering energy storage response capabilities. The scale coordination unit uses the hourly model output as the boundary conditions of the minute-level model and sets the energy storage charging and discharging plan adjustment amount for model coupling coordination. The model parameter update unit uses feedback deviation data to statistically analyze and continuously correct the load prediction error and power output fluctuation coefficient in the model.

[0014] A resource scheduling method based on a smart grid, applied to a resource scheduling system based on a smart grid, includes the following steps: First, a grid-cloud-edge collaborative data acquisition module collects power output, load demand, line impedance, and energy storage status data through edge nodes, transmits them to a cloud server for aggregation and format unification, forming a standardized data source; Second, a two-layer time-scale stochastic scheduling modeling module calls the data source, extracts hourly and minute-level calibration parameters, constructs a stochastic scheduling model containing two sub-models, embeds the probability distribution functions of power output and load demand fluctuations, and determines the objective function and constraints; Third, a stochastic saddle point approximation optimization calculation module receives the model parameters, initializes the iteration variables and... The algorithm iterates and optimizes the scheduling scheme according to its steps, calculating the objective function value and constraint satisfaction at each step, and outputting the scheduling decision variables after convergence or reaching the required number of iterations. The fourth step involves a high-dimensional objective Pareto scheduling analysis module that substitutes the decision variables into the model, calculates the objective function value, uses non-dominated sorting to divide the scheme into levels and calculates congestion distance, and selects the optimal scheduling scheme. The fifth step involves a resource scheduling instruction generation module that decomposes the optimal scheme into control parameters, converts them into scheduling instructions according to the execution unit protocol, and transmits them to each execution unit via the communication network. The sixth step involves a scheduling effect feedback and adjustment module that collects actual operating data from the execution units, compares it with expected data to calculate the deviation, and transmits it to each module according to its classification. Each module then adjusts its parameters or logic based on the deviation to perform dynamic optimization.

[0015] Beneficial Effects: This invention proposes a resource scheduling system and method based on a smart grid. The grid-cloud-edge collaborative data acquisition module can dynamically adjust the data transmission frequency and priority. Combined with preprocessing and verification mechanisms, it solves the problems of low data interaction efficiency and lack of dynamic adjustment mechanisms in existing technologies, ensuring efficient and accurate acquisition and transmission of multi-dimensional data. The dual-layer time-scale stochastic scheduling modeling module constructs hourly and minute-level sub-models and achieves deep coupling through a scale coordination unit, overcoming the shortcomings of single-time-scale modeling in adapting to long-cycle planning and short-term dynamic adjustment, making the scheduling scheme more in line with the actual operating state of the power grid. The stochastic saddle point approximate optimization calculation module, in conjunction with the high-dimensional objective Pareto scheduling analysis module, can generate Pareto optimal solutions with stability verification, taking into account economy, reliability, and environmental protection, solving the problems of multi-objective balance difference and lack of solution stability verification in existing technologies. At the same time, the scheduling effect feedback adjustment module can dynamically correct the parameters of each module, further improving scheduling accuracy and executability, and achieving overall high efficiency, accuracy, and optimization of smart grid resource scheduling, meeting the scheduling needs in complex scenarios. Attached Figure Description

[0016] Figure 1 This is a diagram showing the system module composition of the present invention; Figure 2 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1 As shown, a resource scheduling system based on a smart grid includes: The power grid cloud-edge collaborative data acquisition module collects distributed power output data, load demand data, transmission line impedance data, and energy storage device charging and discharging status data in the smart grid through the data interaction channel between edge nodes and cloud servers, and transmits the collected multi-dimensional data to the two-layer time-scale random scheduling modeling module. Specifically, the implementation process of the power grid cloud-edge collaborative data acquisition module involves deploying edge nodes and cloud servers. The number of edge nodes is determined based on the power grid coverage area, typically one edge node is configured for every 50 square kilometers. Each edge node is equipped with a 32-bit microprocessor and a gigabit Ethernet communication module, supporting the acquisition and transmission of 1000 data points per second. The smart grid data collected by this module includes distributed power generation output data (collected at a frequency of 1 minute / time, with data accuracy controlled within ±0.5kW), load demand data (collected categorized by residential, industrial, and commercial load types, with residential load acquisition accuracy ±10kW and industrial load ±1kW), and transmission line impedance data (collected at 5-minute intervals). The data acquisition process includes measurement accuracy of ±0.01Ω and data on the charging and discharging status of energy storage devices (including charging and discharging power and remaining capacity, with power acquisition accuracy of ±0.1kW and remaining capacity accuracy of ±2%). During the acquisition process, the edge nodes first cache the raw data locally with a cache capacity of 10GB, and then transmit the data to the cloud server through an encrypted communication link (using the AES-256 encryption algorithm). After receiving the data, the cloud server performs integrity verification, and the data that passes the verification is stored in a database with a capacity of 100TB. This module provides a real-time, accurate, and comprehensive data source for subsequent scheduling modeling, avoiding deviations in the scheduling model due to data loss or errors, and ensuring the accuracy of scheduling decisions.

[0019] The two-layer time-scale stochastic scheduling modeling module constructs a stochastic scheduling model, including intraday hourly scheduling sub-scales and intraday minute-scale scheduling sub-scales, based on the received multi-dimensional data, and transmits the model parameters to the stochastic saddle point approximate optimization calculation module. Specifically, the dual-time-scale stochastic scheduling modeling module is implemented based on multi-dimensional data transmitted by the power grid cloud-edge collaborative data acquisition module. First, it determines the time division of the intraday hourly and minute-level scheduling sub-scales. The hourly scheduling sub-scale divides one day into 24 time periods, each lasting one hour; the minute-level scheduling sub-scale divides each hour into 60 time periods, each lasting one minute. During the modeling process, the following inputs are required: distributed power generation output prediction (prediction period matched to the corresponding sub-scale, hourly prediction error controlled within ±5%, minute-level ±2%), load demand prediction (hourly prediction error ±3%, minute-level ±1%), transmission line capacity parameters (set according to line type, e.g., 40MVA for 110kV lines), and energy storage capacity. The module is equipped with parameters (charge and discharge power limit set at 500kW, capacity limit at 1000kWh, and charge and discharge efficiency at 0.9). It constructs a stochastic scheduling model containing an objective function and constraints. The objective function focuses on minimizing operating costs (cost components include generation cost and transmission loss cost; the benchmark value for generation cost is set at 0.5 yuan / kWh, and the benchmark value for transmission loss cost is 0.1 yuan / kWh). Constraints cover upper and lower limits of power output, load supply and demand balance, and line capacity limitations. This module adapts to the scheduling needs of the power grid across different time dimensions through a dual-timescale design. Hourly-level design ensures the rationality of long-term planning, minute-level design addresses short-term fluctuations, and stochastic modeling enhances adaptability to grid uncertainties, providing a scientific model foundation for optimization calculations.

[0020] The random saddle point approximation optimization calculation module uses the random saddle point approximation optimization algorithm to optimize the received model parameters and outputs the optimized scheduling decision variables to the high-dimensional objective Pareto scheduling analysis module. Specifically, the stochastic saddle point approximation optimization calculation module first receives model parameters output by the two-layer time-scale stochastic scheduling modeling module, including objective function coefficients, constraint boundary values, and random parameter probability distribution data (e.g., the probability distribution of distributed power generation output fluctuations adopts a normal distribution, with the mean being the predicted value and the standard deviation being 5% of the predicted value). The module initializes the optimization iteration parameters, sets the number of iterations to 1000, and sets the convergence threshold to 10^-6. In each iteration, 100 scenario samples are first generated based on the random parameter probability distribution, and then saddle point approximation calculation is performed on each scenario sample. During the calculation, the gradient descent method is used to update the minimum variation. The module iterates through two variables: a quantity (corresponding to power output allocation, with an initial update step size of 0.01 that decreases linearly with the number of iterations) and a maximization variable (corresponding to load demand allocation, with an initial update step size of 0.005). It also checks whether the variables meet constraints (e.g., power output does not exceed the upper limit, load allocation is not lower than the lower limit). If not, the variables are corrected. After iteration to convergence or reaching the maximum number of iterations, the optimized scheduling decision variables (including power output, load allocation, and energy storage charging / discharging power for each time period) are output. This module improves the rationality of scheduling decision variables through optimization algorithms, reduces operating costs, and addresses random parameter fluctuations to ensure the feasibility of the scheduling scheme.

[0021] The high-dimensional objective Pareto scheduling analysis module, based on the received scheduling decision variables and combined with the economic, reliability, and environmental protection objective parameters of smart grid resource scheduling, constructs a high-dimensional objective Pareto scheduling model and performs non-dominated sorting analysis, outputting the Pareto optimal scheduling scheme to the resource scheduling instruction generation module. Specifically, the high-dimensional objective Pareto scheduling analysis module first receives the scheduling decision variables output by the stochastic saddle point approximate optimization calculation module, and then determines the high-dimensional objective parameters for smart grid resource scheduling, including economic objectives (based on average daily operating costs, with a target value controlled within 100,000 yuan), reliability objectives (based on power supply reliability rate, with a target value ≥ 99.95%), and environmental objectives (based on average daily carbon emissions, with a target value ≤ 50 tons). When constructing the high-dimensional objective Pareto scheduling model, the module first substitutes the scheduling decision variables into each objective function to calculate the objective values, and then uses a non-dominated ranking method to rank the objective values ​​corresponding to all decision variables. During the ranking process, a dominance judgment criterion is set (if the decision variable... If all objective values ​​of variable A are better than those of decision variable B, then A dominates B. The congestion distance of each decision variable is calculated (reflecting the distribution density of decision variables in the objective space; a larger distance value indicates a more uniform distribution). The set of decision variables with the highest non-dominant level is selected through sorting, i.e., the Pareto optimal solution set. Then, the feasibility of the solution set is verified by combining real-time power grid operating parameters (such as transmission line load rate, with an upper limit set at 80% and a lower limit set at 20% for remaining energy storage capacity). Solutions that do not conform to the real-time state are eliminated, and finally, the Pareto optimal scheduling scheme is output. This module achieves a balance between multiple objectives, avoiding the deterioration of other objectives due to the optimization of a single objective, and providing an optimal solution that takes into account multiple needs for subsequent instruction generation.

[0022] The resource scheduling instruction generation module converts the Pareto optimal scheduling scheme into specific scheduling instructions for distributed power sources, load nodes, energy storage devices and transmission lines, and transmits them to the smart grid execution unit. Specifically, when the resource scheduling instruction generation module is implemented, it first receives the Pareto optimal scheduling scheme output by the high-dimensional objective Pareto scheduling analysis module. The scheme includes distributed power output, load node power consumption, energy storage device charging and discharging, and transmission line operation parameters for each time period. The module decomposes the scheme according to the type of execution unit. The decomposed parameters for distributed power execution units (such as photovoltaic power plants and wind farms) include the output adjustment range (step size set to 10kW) and adjustment time (accurate to the second). The decomposed parameters for load node execution units include the power consumption limit value (residential load limit accuracy ±50W, industrial load ±500kW) and the limit start time. The decomposed parameters for energy storage device execution units include the charging and discharging mode (charging, discharging, standby) and power setting. The setpoint (accuracy ±0.05kW) and the decomposed parameters of the transmission line execution unit include power flow control values ​​(accuracy ±0.1MW). After decomposition, the module converts the parameters into executable scheduling instructions according to the communication protocol of each execution unit (e.g., Modbus protocol for distributed power sources and DL / T645 protocol for load nodes). The instruction format adopts binary encoding and the length is set to 64 bytes. After conversion, the instructions are transmitted to each execution unit of the smart grid through a dedicated communication network (transmission rate ≥1Mbps, delay ≤100ms) to ensure that the execution units receive them in a timely manner. This module transforms the abstract scheduling scheme into specific executable instructions, builds a bridge between scheduling decisions and actual execution, and ensures the implementation of the scheduling scheme.

[0023] The scheduling effect feedback and adjustment module collects the actual operating data of the smart grid execution unit and compares it with the expected data corresponding to the scheduling instructions output by the resource scheduling instruction generation module. The deviation data is then fed back to the power grid cloud-edge collaborative data acquisition module, the two-layer time-scale random scheduling modeling module, the random saddle point approximate optimization calculation module, the high-dimensional objective Pareto scheduling analysis module, and the resource scheduling instruction generation module for dynamic adjustment of the parameters of each module.

[0024] Specifically, when implementing the scheduling effect feedback and adjustment module, data acquisition equipment (such as smart meters and power sensors) is first deployed to collect the actual operating data of the smart grid execution unit. The collected parameters include the actual output of distributed power sources (collection frequency 1 second / time, accuracy ±0.2kW), the actual power consumption of load nodes (1 second / time, accuracy ±5W / ±100kW), the actual charging and discharging power and remaining capacity of energy storage devices (1 second / time, accuracy ±0.05kW / ±1%), and the actual current and voltage of transmission lines (0.5 seconds / time, current accuracy ±0.1A, voltage accuracy ±0.5V). The collected data is then compared with the expected data corresponding to the scheduling instructions output by the resource scheduling instruction generation module to calculate the deviation (deviation rate). The calculation formula is |actual value - expected value| / expected value × 100%, with a deviation rate threshold set at 5%. If the deviation rate exceeds the threshold, the deviation data is categorized and fed back according to module type. Deviation data fed back to the power grid cloud-edge collaborative data acquisition module is used to adjust the acquisition frequency (the acquisition frequency is increased to 0.5 minutes / time when the deviation exceeds the limit). Deviation data fed back to the dual-layer time-scale random scheduling modeling module is used to correct model parameters (such as adjusting the prediction error coefficient). Deviation data fed back to other modules is used to optimize the calculation logic, adjust the analysis standards, and correct the instruction conversion rules, respectively. This module forms a scheduling closed loop, which promptly detects and adjusts scheduling deviations through real-time feedback, continuously improving the adaptability and accuracy of the scheduling system and ensuring the long-term stable operation of the smart grid.

[0025] Preferably, the expression for the two-layer time-scale stochastic scheduling model constructed by the two-layer time-scale stochastic scheduling modeling module is as follows: The first hourly scheduling subscale within the day Hourly scheduling decision variables This represents the total number of time periods for hourly scheduling within a day. This indicates the first minute-level scheduling subscale within the day. Minute-level scheduling decision variables For the first Total number of time periods scheduled per minute within an hour; Indicates the first The probability of such scenarios, Total number of scenes; The cost function representing the hourly scheduling subscale includes the distributed generation cost coefficient and the transmission loss cost coefficient. The cost function representing the minute-level scheduling subscale includes the energy storage device charging and discharging cost coefficient and the load reduction penalty cost coefficient; Indicates the first Hour Time weighting coefficient for minutes; The coordination coefficient represents the two-level time-scale scheduling decision variable; Indicates the first Scenario, First Under hourly conditions The scheduling decision variable for each minute.

[0026] Specifically, the model constructed using the dual-timescale stochastic scheduling modeling module should clearly define the time periods for hourly and minute-level scheduling sub-scales within a day. At the hourly level, one day is divided into 24 time periods, each lasting one hour; at the minute level, each hour is divided into 60 time periods, each lasting one minute. The total number of scenarios and the probability of each scenario are also determined. Typically, the number of scenarios is set to 50-100, and the probability of a single scenario is determined based on historical data statistics, ranging from 0.01 to 0.05. The cost function in the model requires specific parameters. The power generation cost coefficient is set according to the power source type: 0.5-0.6 yuan / kWh for thermal power and 0.3-0.4 yuan / kWh for photovoltaic power. The transmission loss cost coefficient is set to 0.08-0.12 yuan / kWh, the energy storage charging and discharging cost coefficient is 0.2-0.3 yuan / kWh, and the load shedding penalty cost coefficient is 1.5-2 yuan / kWh. The time weighting coefficients are allocated based on the importance of minute-level time periods within hour-level time periods. Peak electricity consumption periods have a weight of 0.02-0.03, average periods 0.015-0.02, and valley periods 0.01-0.015. A coordination coefficient of 0.8-1.2 is set to balance the differences in decision variables between the two scales. By quantifying model parameters, a scheduling model adapted to different time scales of the power grid is accurately constructed, improving the model's fit to actual operating scenarios and providing an accurate model foundation for subsequent optimization calculations. This avoids scheduling schemes deviating from actual needs due to parameter ambiguity.

[0027] Preferably, the random saddle point approximation optimization algorithm used by the random saddle point approximation optimization calculation module is expressed as follows: ,in, This represents the minimization variable in the optimization problem, corresponding to the power output allocation variable in smart grid resource scheduling. for The feasible domain; This represents the maximization variable in the optimization problem, corresponding to the load demand allocation variable in smart grid resource scheduling. for The feasible domain; Represents a random variable The mathematical expectation, These are random parameters in a smart grid, including random fluctuations in distributed generation output and random fluctuations in load demand. for The probability distribution; The objective function for random saddle point optimization includes the matching coefficient between power output and load demand, and the transmission line capacity constraint coefficient. They represent the minimization variables respectively. With maximizing variables The regularization coefficient; They represent the first During the next iteration and The value of .

[0028] Specifically, the algorithm used in the random saddle point approximation optimization calculation module determines the feasible regions of the minimization and maximization variables during implementation. The feasible region of the minimization variable (power output allocation) is set according to the rated power of the power source, such as 0-100MW for a 100MW photovoltaic power station. The feasible region of the maximization variable (load demand allocation) is set according to the maximum electricity capacity of users, such as 0-10kW for residential users and 0-1000kW for industrial users. The probability distribution of the random parameters (distributed power output and load demand fluctuations) needs to be fitted based on historical data, usually using a normal distribution. The mean of the fluctuation is set to 0, and the standard deviation is adjusted according to the power source type and load type. The standard deviation of photovoltaic power output fluctuation is 5%-8% of the rated power, and the standard deviation of residential load fluctuation is 3%-5% of the maximum electricity capacity. The matching coefficient in the objective function is set to 0.6-0.8 to measure the degree of matching between power output and load demand, and the line capacity constraint coefficient is set to 0.9-1.1 to ensure that the decision variables do not exceed the line capacity limit. Regularization coefficients α and β are set to 0.05-0.1 and 0.03-0.05, respectively. The initial values ​​for iteration are set according to the optimal solution of the previous cycle, and the step size is adjusted every 100 iterations during the iteration process. By clearly defining the algorithm parameters and feasible region, the convergence and accuracy of the optimization calculation are ensured, effectively addressing random fluctuations in the power grid, outputting scheduling decision variables that meet the constraints, and providing reliable input data for high-dimensional objective analysis.

[0029] Preferably, the expression for the high-dimensional objective Pareto scheduling model constructed by the high-dimensional objective Pareto scheduling analysis module is as follows: ,in, The decision variables representing high-dimensional objective Pareto scheduling include distributed power generation output, energy storage device charging and discharging power, and load transfer amount; This represents the economic objective function. Cost per unit of electricity generated, for Power output at all times The charging and discharging cost of energy storage units, for Energy storage charging and discharging power at all times, To reduce costs per unit load, for Load reduction amount at any time; Represents the reliability objective function. for Energy storage and discharge power at all times for Constant load demand, It is a very small positive number; This represents the environmental protection objective function. Emissions per unit of electricity generated The equivalent emissions reduction per unit load; These are the minimum and maximum constraints for the power output, respectively. These are the minimum and maximum constraints for load demand, respectively. The apparent power of the transmission line. This is a constraint on the maximum apparent power of transmission lines.

[0030] Specifically, the algorithm used in the random saddle point approximation optimization calculation module determines the feasible regions of the minimization and maximization variables during implementation. The feasible region of the minimization variable (power output allocation) is set according to the rated power of the power source, such as 0-100MW for a 100MW photovoltaic power station. The feasible region of the maximization variable (load demand allocation) is set according to the maximum electricity capacity of users, such as 0-10kW for residential users and 0-1000kW for industrial users. The probability distribution of the random parameters (distributed power output and load demand fluctuations) needs to be fitted based on historical data, usually using a normal distribution. The mean of the fluctuation is set to 0, and the standard deviation is adjusted according to the power source type and load type. The standard deviation of photovoltaic power output fluctuation is 5%-8% of the rated power, and the standard deviation of residential load fluctuation is 3%-5% of the maximum electricity capacity. The matching coefficient in the objective function is set to 0.6-0.8 to measure the degree of matching between power output and load demand, and the line capacity constraint coefficient is set to 0.9-1.1 to ensure that the decision variables do not exceed the line capacity limit. Regularization coefficients α and β are set to 0.05-0.1 and 0.03-0.05, respectively. The initial values ​​for iteration are set according to the optimal solution of the previous cycle, and the step size is adjusted every 100 iterations during the iteration process. By clearly defining the algorithm parameters and feasible region, the convergence and accuracy of the optimization calculation are ensured, effectively addressing random fluctuations in the power grid, outputting scheduling decision variables that meet the constraints, and providing reliable input data for high-dimensional objective analysis.

[0031] Preferably, the data acquisition and transmission model expression of the power grid cloud-edge collaborative data acquisition module is as follows: ,in, express Aggregated data received in real time from the cloud; Indicates the total number of edge nodes; express Time of the first The data weight coefficient of each edge node is related to the communication bandwidth and data integrity of the edge node; express Time of the first Raw data collected by each edge node, including distributed power output and load demand data; express Time of the first The data noise coefficient of each edge node; express Time of the first Noise data for each edge node; This represents matrix multiplication. express Time of the first A data correction matrix for each edge node is used to eliminate systematic errors during the data acquisition process.

[0032] Specifically, the data acquisition and transmission model of the power grid cloud-edge collaborative data acquisition module determines the number of edge nodes and data weighting coefficients during implementation. Edge nodes are configured according to the power grid node density, with one node per 30-50 square kilometers. The weighting coefficients are determined based on the node's communication bandwidth and data integrity score: nodes with bandwidth ≥100Mbps and data integrity ≥99.5% have a weight of 0.15-0.2; those with bandwidth 50-100Mbps and integrity 99%-99.5% have a weight of 0.1-0.15; and those with bandwidth <50Mbps or integrity <99% have a weight of 0.05-0.1. The noise figure is set according to the acquisition environment: 0.02-0.03 for outdoor nodes and 0.01-0.02 for indoor nodes. Noise data is randomly generated at 1%-3% of the acquired data. The data correction matrix is ​​3×3 in dimension, with diagonal elements set to 0.98-1.02 to correct system errors and off-diagonal elements set to 0-0.01 to reduce cross-interference. The timestamp precision of the collected data is set to milliseconds to ensure data time synchronization. Data is transmitted in blocks, each 1024 bytes in size, with an interval of 10-20 milliseconds between blocks. By clearly defining the parameters for data collection and transmission, the accuracy, integrity, and synchronization of the data are improved, providing a high-quality data source for subsequent modeling and computation, and avoiding scheduling decision deviations due to data issues.

[0033] Preferably, the instruction conversion model expression of the resource scheduling instruction generation module is: ,in, express At any time for the first Scheduling instructions for each execution unit; This represents the total number of smart grid execution units, including distributed power sources, energy storage devices, and load nodes. Indicates the first The instruction translation coefficient matrix of each execution unit is related to the type and rated parameters of the execution unit; express At any time for the first Pareto optimal scheduling parameters for each execution unit; Indicates the first Feedback adjustment coefficient matrix for each execution unit; express Time of the first Feedback data from each execution unit, including actual operating power, voltage, and current data; This represents the Hadamard product operation; express Time of the first The state coefficient of each execution unit is set to 1 when the execution unit operates normally and 0.5 when the execution unit operates at a reduced rate.

[0034] Specifically, the instruction conversion model of the resource scheduling instruction generation module determines the number of execution units and the conversion coefficient matrix during implementation. Execution units are categorized by type: distributed power sources, energy storage devices, load nodes, and transmission lines are each divided into independent unit groups. The number of units in each group is set according to the grid scale: 5-10 units per group for small grids and 20-30 units per group for large grids. The conversion coefficient matrix is ​​2×2 dimension, with conversion coefficients set at 0.95-1.05 for distributed power sources, 0.9-1.1 for energy storage devices, 0.85-0.95 for load nodes, and 0.98-1.02 for transmission lines. The feedback adjustment coefficient matrix has the same dimension as the conversion coefficient matrix, and its coefficient value is 0.1-0.2 times the conversion coefficient. Pareto optimal scheduling parameters are updated periodically, every 5-10 minutes. Feedback data is collected every 1-2 seconds, including key parameters such as voltage, current, and power. The state coefficient is adjusted according to the equipment operating status: 1 for normal operation, 0.5 for minor faults during derating operation, and 0 for severe faults during shutdown. The instruction encoding adopts a fixed format: the first 8 bytes are the unit identifier, the middle 40 bytes are the parameter data, and the last 16 bytes are the checksum. By clearly defining the instruction conversion parameters and format, it is ensured that the scheduling instructions can be accurately adapted to each execution unit, improving the accuracy and efficiency of instruction execution, building a reliable bridge between the scheduling plan and actual execution, and ensuring the effective implementation of scheduling decisions.

[0035] Preferably, the high-dimensional objective Pareto scheduling analysis module includes a Pareto solution generation unit, a non-dominated sorting calculation unit, an optimal solution screening unit, and a solution stability verification unit. The Pareto solution generation unit receives scheduling decision variables, combines them with economic, reliability, and environmental protection objective parameters, and generates an initial Pareto solution set through multi-objective optimization iteration under constraints. The solution set is then substituted into the objective function to calculate and exclude solutions with hyperparameter boundaries. The non-dominated sorting calculation unit determines the dominance relationship of each solution in the initial solution set, compares the objective function values ​​to divide the non-dominated levels, calculates the congestion distance, and considers the distance differences between adjacent solutions in each objective dimension. The optimal solution screening unit selects candidate optimal solutions based on the sorting results and congestion distance, and, in conjunction with the transmission line load rate and remaining energy storage capacity verification, eliminates solutions that do not meet real-time constraints. The solution stability verification unit substitutes the optimal solution into the model, changes the power output and load demand fluctuations, observes the change amplitude of the objective function value, and returns to regenerate the solution set if the value exceeds the threshold.

[0036] Specifically, the high-dimensional objective Pareto scheduling analysis module comprises four units: Pareto solution generation, non-dominated sorting calculation, optimal solution selection, and solution stability verification. During implementation, the Pareto solution generation unit first receives the scheduling decision variables output by the stochastic saddle point approximate optimization calculation module. Combined with objective parameters of economic efficiency (daily operating cost controlled at 80,000-120,000 RMB), reliability (power supply reliability ≥99.95%), and environmental friendliness (daily carbon emissions ≤45-55 tons), it generates an initial Pareto solution set through multi-objective optimization iteration under constraints. The number of iterations is set to 200-300. In each iteration, the decision variables are substituted into the objective function calculation, and solutions exceeding the upper and lower limits of the objective parameters by 10% are excluded. The non-dominated sorting calculation unit judges the dominance relationship of each solution in the initial solution set, compares the magnitudes of different solutions in the three objective function values, divides the solutions into 3-5 non-dominated levels, and calculates the congestion distance of each solution. During the calculation, the distance difference between adjacent solutions is controlled within 5%-8% to ensure that the sorting results reflect the uniformity of solution distribution. The optimal solution screening unit selects the top 5-8 solutions with the highest non-dominated level and the largest congestion distance as candidate optimal solutions. Feasibility is then verified by considering the transmission line load rate (upper limit 75%-85%) and the remaining capacity of energy storage devices (lower limit 15%-25%), eliminating solutions that do not meet real-time constraints. The solution stability verification unit substitutes the selected optimal solutions into a high-dimensional objective Pareto scheduling model, changing the output fluctuation of distributed power sources (fluctuation amplitude 5%-10%) and the load demand fluctuation (fluctuation amplitude 3%-7%), observing the change in the objective function value. If the change amplitude is ≤3%, it is determined as the final solution; otherwise, it returns to the Pareto solution generation unit to regenerate the solution set. Through multi-unit collaboration, the generated scheduling scheme is ensured to consider multiple objectives and possess stability, avoiding scheme deviations caused by single-unit processing, and improving the scientific and practical nature of scheduling decisions.

[0037] Preferably, the power grid cloud-edge collaborative data acquisition module includes an edge data acquisition unit, a data preprocessing unit, a cloud data receiving unit, and a data interaction control unit. The edge data acquisition unit collects power output, load demand, line current and voltage, and energy storage charging and discharging status data at preset intervals through sensors and smart meters. The data preprocessing unit identifies abnormal values ​​in the original data using the statistical deviation method, and converts the heterogeneous data into a unified format through adjacent data interpolation and replacement. The cloud data receiving unit receives standardized data through an encrypted link, calculates checksums and compares the data sent from the transmitting end, and requests retransmission if there is a discrepancy. The data interaction control unit dynamically adjusts the data transmission frequency and priority according to the communication status of the edge nodes and the cloud load. When the bandwidth is insufficient, the frequency of non-calibrated data is reduced, and when the load is high, calibrated data is received first.

[0038] Specifically, the power grid cloud-edge collaborative data acquisition module comprises four units: edge data acquisition, data preprocessing, cloud data reception, and data interaction control. During implementation, the edge data acquisition unit collects data every 1-2 minutes via sensors and smart meters deployed at key nodes of the smart grid. This data covers distributed power generation output (accuracy ±0.3-0.5kW), load demand (residential load accuracy ±8-12W, industrial load ±0.8-1.2kW), transmission line current and voltage (current accuracy ±0.1-0.2A, voltage accuracy ±0.3-0.6V), and the charging and discharging status of energy storage devices (power accuracy ±0.08-0.12kW, remaining capacity accuracy ±1.5%-2.5%). The data preprocessing unit uses a statistical bias method based on the 3σ principle to identify outliers, corrects them through linear interpolation of data from 3-5 adjacent time points, and converts heterogeneous data of different formats into JSON format, with a data compression rate controlled at 30%-40%. The cloud-based data receiving unit receives data via a communication link employing the AES-256 encryption algorithm. Upon receiving data, it calculates a 32-bit CRC checksum and compares it with the sending end. The checksum pass rate must be ≥99.8%. If inconsistent, a retransmission mechanism is triggered, with a maximum of 3 retransmissions. The data interaction control unit monitors the edge node communication bandwidth (threshold set at 50-100Mbps) and the cloud server CPU load (threshold set at 60%-80%) in real time. When bandwidth is insufficient, the frequency of non-critical data transmission is reduced to 5-10 minutes / time. When the load is too high, critical data such as power output and load demand are prioritized. Through multi-unit collaboration, the real-time performance, accuracy, and transmission stability of data acquisition are ensured, providing a high-quality data source for subsequent scheduling modeling and calculation, and preventing data problems from affecting scheduling decisions.

[0039] Preferably, the dual-layer time-scale stochastic scheduling modeling module includes an hourly model building unit, a minute-level model building unit, a scale coordination unit, and a model parameter update unit. The hourly model building unit determines the hourly scheduling segments, load and power output prediction values ​​based on historical and real-time data, and builds a model with power output and line capacity constraints, aiming at minimizing operating costs. The minute-level model building unit divides each hour into minute-level time periods, collects high-frequency fluctuation data of load and power output, and builds a model that aims to track the hourly plan while considering energy storage response capabilities. The scale coordination unit uses the hourly model output as the boundary conditions of the minute-level model and sets the energy storage charging and discharging plan adjustment amount for model coupling coordination. The model parameter update unit uses feedback deviation data to statistically analyze and continuously correct the load prediction error and power output fluctuation coefficient in the model.

[0040] Specifically, the dual-layer time-scale stochastic scheduling modeling module comprises four units: hourly model construction, minute-level model construction, scale coordination, and model parameter updating. During implementation, the hourly model construction unit, based on 30-60 days of historical and real-time data provided by the power grid-cloud-edge collaborative data acquisition module, divides a day into 24-hour periods. It determines the load forecast (error controlled within ±4%-6%) and power output forecast (thermal power error ±2%-3%, renewable energy error ±8%-12%) for each period, constructing a model with the objective of minimizing daily operating costs. The model incorporates constraints such as upper and lower limits of power output (20%-100% of rated power for thermal power, 0-100% for renewable energy) and transmission line capacity (35-45 MVA for 110kV lines). The minute-level model building unit divides each hour into 60 minute-level time periods, collecting minute-level rapid load fluctuation data (fluctuation frequency 1-2 times / minute) and distributed power output high-frequency fluctuation data (fluctuation frequency 3-5 times / minute) to build a model aimed at tracking the hourly scheduling plan. The model considers the charging and discharging response time of energy storage equipment (≤10-15 seconds) and the real-time load adjustment range (residential load ≤±5%, industrial load ≤±10%). The scale coordination unit uses the time-period output plan from the hourly model as the boundary condition for the minute-level model, setting the adjustment amount for the energy storage equipment charging and discharging plan (each adjustment ≤5%-8% of rated power), and performs inter-scale parameter coordination every 15-20 minutes. The model parameter update unit adjusts the deviation data fed back by the scheduling effect feedback module (deviation rate threshold set at 4%-6%), and every 24 hours, based on the statistical analysis results of the deviation data, uses a rolling window method to correct the load forecast error coefficient (adjustment range ±0.02-0.05) and the power output fluctuation coefficient (adjustment range ±0.03-0.06). By constructing a scheduling model that adapts to different time scales through multi-unit collaboration, coordination between scales and dynamic parameter updates are achieved, improving the model's adaptability to the power grid's operating state and providing a scientific and accurate model foundation for optimization calculations.

[0041] The dual-time-scale stochastic scheduling model in this invention is a modeling system adapted to the scheduling needs of smart grids in different time dimensions. It includes two scheduling sub-scales: hourly and minute-level within a day, and is used to integrate multi-source data from the power grid to generate a scientific scheduling model foundation. Its implementation relies on a dual-layer time-scale stochastic scheduling modeling module. First, it acquires historical and real-time data for the past 30-60 days through a grid-cloud-edge collaborative data acquisition module, dividing a day into 24-hour time periods and each hour into 60 minute-level time periods. The hourly model combines load forecasts (error ±4%-6%) and power output forecasts (thermal power error ±2%-3%, new energy ±8%-12%), aiming to minimize the average daily operating cost, and incorporates constraints such as upper and lower limits of power output and transmission line capacity (35-45MVA for 110kV lines). The minute-level model collects high-frequency fluctuation data of load and power output, considering the energy storage charging and discharging response time (≤10-15 seconds) and load adjustment amplitude, aiming to track the hourly plan. At the same time, the scale coordination unit coordinates the parameters of the two sub-models every 15-20 minutes, and the model parameter update unit corrects the prediction error and fluctuation coefficient every 24 hours based on the deviation data (deviation rate threshold 4%-6%). The purpose of this model is to balance the rationality of long-term power grid planning with the needs of short-term dynamic adjustment, avoiding the limitations of single-scale modeling; to improve the adaptability of the scheduling model to the power grid operating status, to provide accurate and dynamic model support for subsequent optimization calculations, and to ensure that the scheduling scheme fits the actual operating scenario of the power grid.

[0042] The stochastic saddle point approximation optimization algorithm in this invention is the core algorithm for optimizing the parameters of a two-layer time-scale stochastic scheduling model. It outputs scheduling decision variables that meet the constraints through collaborative optimization of minimization and maximization variables. Its implementation relies on the stochastic saddle point approximation optimization calculation module. First, it receives the objective function coefficients, constraint boundary values, and random parameter probability distribution data (e.g., distributed power output fluctuations follow a normal distribution with a standard deviation of 5%-8% of rated power) output from the two-layer time-scale stochastic scheduling model. It initializes the iteration parameters, setting the iteration count to 1000 and the convergence threshold to 10^-6. Each iteration generates 100 scenario samples. The gradient descent method is used to update the minimization variable (power output allocation, initial step size 0.01) and the maximization variable (load demand allocation, initial step size 0.005). Simultaneously, it verifies whether the variables meet constraints such as upper and lower limits of power output and load supply-demand balance; if not, it corrects them. After iteration to convergence or reaching the maximum number of iterations, it outputs the optimized scheduling decision variables. The algorithm aims to improve the rationality of scheduling decision variables, reduce operating costs while coping with random fluctuations in the power grid, and ensure that variables meet power grid constraints. It also addresses the problem that traditional optimization algorithms struggle to balance randomness and constraints, providing reliable input data for high-dimensional objective Pareto scheduling analysis and promoting the transformation of scheduling decisions from "experience-based" to "precise."

[0043] The high-dimensional objective Pareto scheduling model in this invention is a multi-objective analysis model that takes into account the economy, reliability, and environmental protection of smart grids. It uses non-dominated sorting to select the Pareto optimal scheduling scheme and provides a basis for generating resource scheduling instructions. The implementation relies on a high-dimensional objective Pareto scheduling analysis module. It first receives scheduling decision variables from the random saddle point approximate optimization calculation module to determine target parameters (economic efficiency: daily operating cost of 80,000-120,000 RMB; reliability: power supply reliability ≥99.95%; environmental friendliness: daily carbon emissions ≤45-55 tons). The Pareto solution generation unit generates an initial solution set through 200-300 multi-objective optimization iterations, excluding solutions with hyperparameter limits of 10%. The non-dominated sorting calculation unit compares the objective function values ​​of the solutions, divides them into 3-5 non-dominated levels, and calculates the congestion distance (difference between adjacent solutions of 5%-8%). The optimal solution screening unit selects the top 5-8 candidate solutions, combining transmission line load rate (upper limit 75%-85%) and remaining energy storage capacity (lower limit 15%-25%) for verification, eliminating solutions that do not meet the constraints. The solution stability verification unit changes power output (fluctuation 5%-10%) and load demand (fluctuation 3%-7%); if the objective function value changes by ≤3%, the final solution is determined. The model aims to achieve multi-objective balance, avoiding the deterioration of other objectives due to the optimization of a single objective; it breaks through the limitations of traditional single-objective scheduling, allowing scheduling schemes to simultaneously meet the needs of economical grid operation, reliable power supply, and green environmental protection, thus aligning with the sustainable development direction of smart grids.

[0044] The power grid cloud-edge collaborative analysis platform in this invention is an architecture supporting multi-dimensional data acquisition, transmission, and interaction in smart grids. Through collaboration between edge nodes and the cloud, it provides high-quality data sources for various scheduling modules. Its implementation relies on the power grid cloud-edge collaborative data acquisition module. The edge data acquisition unit collects data at a frequency of 1-2 minutes per acquisition, using sensors to collect data on distributed power output (accuracy ±0.3-0.5kW), load demand (residential ±8-12W, industrial ±0.8-1.2kW), line current and voltage (current ±0.1-0.2A, voltage ±0.3-0.6V), and energy storage status (power ±0.08-0.12kW, capacity ±1.5%-2.5%). The data preprocessing unit identifies outliers using the 3σ principle, and after processing by 3-5 adjacent... The platform performs real-time data interpolation and correction, converting heterogeneous data into JSON format (compression rate 30%-40%). The cloud data receiving unit receives data via an AES-256 encrypted link, calculates a 32-bit CRC checksum (pass rate ≥99.8%), and retransmits data if inconsistent (maximum 3 times). The data interaction control unit monitors edge node bandwidth (threshold 50-100Mbps) and cloud CPU load (threshold 60%-80%). When bandwidth is insufficient, it reduces the frequency of non-critical data to 5-10 minutes / time; when the load is too high, it prioritizes receiving critical data. The platform's role is to ensure data real-time performance, accuracy, and transmission stability; it solves the problems of low efficiency and high latency in traditional centralized data processing, providing a solid data foundation for scheduling modeling, optimization calculations, and decision analysis. It serves as the "data hub" for the efficient operation of the smart grid resource scheduling system.

[0045] like Figure 2As shown, a resource scheduling method based on a smart grid is applied to a resource scheduling system based on a smart grid. The method includes the following steps: First, a grid-cloud-edge collaborative data acquisition module collects power output, load demand, line impedance, and energy storage status data through edge nodes, transmits them to a cloud server for aggregation and format unification, forming a standardized data source; Second, a two-layer time-scale stochastic scheduling modeling module calls the data source, extracts hourly and minute-level calibration parameters, constructs a stochastic scheduling model containing two sub-models, embeds the probability distribution functions of power output and load demand fluctuations, and determines the objective function and constraints; Third, a stochastic saddle point approximation optimization calculation module receives the model parameters and initializes the iteration variables. The algorithm iterates and optimizes the scheduling scheme by calculating the objective function value and constraint satisfaction at each iteration, outputting the scheduling decision variables after convergence or reaching the required number of iterations. The fourth step involves a high-dimensional objective Pareto scheduling analysis module that substitutes the decision variables into the model, calculates the objective function value, uses non-dominated sorting to divide the scheme into levels and calculates congestion distance, and selects the optimal scheduling scheme. The fifth step involves a resource scheduling instruction generation module that decomposes the optimal scheme into control parameters, converts them into scheduling instructions according to the execution unit protocol, and transmits them to each execution unit via the communication network. The sixth step involves a scheduling effect feedback and adjustment module that collects actual operating data from the execution units, compares it with expected data to calculate the deviation, and transmits it to each module according to the deviation. Each module then adjusts its parameters or logic based on the deviation to perform dynamic optimization.

[0046] A resource scheduling system and method based on smart grids forms a complete scheduling closed loop through the collaboration of six modules. The grid-cloud-edge collaborative data acquisition module ensures efficient and accurate transmission of multi-dimensional data; the dual-timescale stochastic scheduling modeling module adapts to scheduling needs at different time scales; the stochastic saddle point approximation optimization calculation module improves the optimization accuracy of decision variables; the high-dimensional objective Pareto scheduling analysis module achieves multi-objective balance; the resource scheduling instruction generation module ensures instruction execution; and the scheduling effect feedback and adjustment module dynamically optimizes parameters. The cooperation of these modules significantly improves the overall scheduling efficiency. The system features a high degree of technological integration, combining cloud-edge collaboration, multi-scale modeling, and multi-algorithm optimization technologies. It can cope with distributed power sources and load fluctuations while balancing economy, reliability, and environmental protection, making it suitable for the complex operation scenarios of smart grids.

[0047] This system and method effectively overcome the shortcomings of the prior art: Addressing the deficiency of single-timescale modeling in existing technologies, it constructs hourly and minute-level sub-models through a dual-layer timescale stochastic scheduling modeling module, and achieves deep coupling through a scale coordination unit, enabling the scheduling scheme to simultaneously adapt to long-term planning and short-term dynamic adjustments. Addressing the problems of low data interaction efficiency and poor multi-objective balance in existing technologies, the power grid-cloud-edge collaborative data acquisition module improves data interaction efficiency by dynamically adjusting transmission frequency and priority. The stochastic saddle point approximate optimization calculation module and the high-dimensional objective Pareto scheduling analysis module work together to generate the optimal scheme after stability verification, solving the multi-objective balance and solution stability problems, and comprehensively improving the scientific nature and practicality of scheduling.

[0048] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0049] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A resource scheduling system based on smart grid, characterized in that, Comprise: The power grid cloud edge collaborative data acquisition module acquires distributed power output data, load demand data, transmission line impedance data and energy storage device charging and discharging state data in the smart grid through the data interaction channel between the edge node and the cloud server, and transmits the multi-dimensional data collected to the double-time-scale random scheduling modeling module; The double-time-scale random scheduling modeling module constructs a random scheduling model including an intraday hourly scheduling sub-scale and an intraday minute-level scheduling sub-scale based on the received multi-dimensional data, and transmits the model parameters to the random saddle point approximation optimization calculation module; the random saddle point approximation optimization calculation module optimizes the received model parameters by using a random saddle point approximation optimization algorithm, and outputs the optimized scheduling decision variables to the high-dimensional target Pareto scheduling analysis module; the high-dimensional target Pareto scheduling analysis module constructs a high-dimensional target Pareto scheduling model based on the received scheduling decision variables, in combination with economic, reliability and environmental protection target parameters of the smart grid resource scheduling, and performs non-dominated sorting analysis, and outputs a Pareto optimal scheduling scheme to the resource scheduling instruction generation module; the resource scheduling instruction generation module converts the Pareto optimal scheduling scheme into specific scheduling instructions for distributed power, load nodes, energy storage devices and transmission lines, and transmits them to the smart grid execution unit; the scheduling effect feedback adjustment module collects the actual operation data of the smart grid execution unit, compares and analyzes the expected data corresponding to the scheduling instructions output by the resource scheduling instruction generation module, and feeds back the deviation data to the power grid cloud edge collaborative data acquisition module, the double-time-scale random scheduling modeling module, the random saddle point approximation optimization calculation module, the high-dimensional target Pareto scheduling analysis module and the resource scheduling instruction generation module, respectively, for dynamic adjustment of the parameters of each module.

2. The resource scheduling system based on smart grid according to claim 1, wherein, The double-time-scale random scheduling model constructed by the double-time-scale random scheduling modeling module has an expression as follows: hourly scheduling sub-scale, hourly scheduling decision variable, total time period of hourly scheduling sub-scale; minute-level scheduling sub-scale, minute-level scheduling decision variable, total time period of minute-level scheduling sub-scale in the hour; probability of the th scenario, total number of scenarios; cost function of hourly scheduling sub-scale, including distributed power generation cost coefficient, power transmission loss cost coefficient; cost function of minute-level scheduling sub-scale, including energy storage device charging and discharging cost coefficient, load reduction penalty cost coefficient; time weight coefficient of the minute in the hour; coordination coefficient of bi-level time scale scheduling decision variable; scheduling decision variable of the minute under the condition of the hour in the th scenario.

3. The resource scheduling system based on smart grid of claim 1, wherein, The random saddle point approximation optimization algorithm used by the random saddle point approximation optimization calculation module is expressed as follows: ,in, This represents the minimization variable in the optimization problem, corresponding to the power output allocation variable in smart grid resource scheduling. for The feasible domain; This represents the maximization variable in the optimization problem, corresponding to the load demand allocation variable in smart grid resource scheduling. for The feasible domain; Represents a random variable The mathematical expectation, These are random parameters in a smart grid, including random fluctuations in distributed generation output and random fluctuations in load demand. for The probability distribution; The objective function for random saddle point optimization includes the matching coefficient between power output and load demand, and the transmission line capacity constraint coefficient. They represent the minimization variables respectively. With maximizing variables The regularization coefficient; They represent the first During the next iteration and The value of .

4. The resource scheduling system based on smart grid of claim 1, wherein, The expression for the high-dimensional objective Pareto scheduling model constructed by the high-dimensional objective Pareto scheduling analysis module is as follows: ,in, The decision variables representing high-dimensional objective Pareto scheduling include distributed power generation output, energy storage device charging and discharging power, and load transfer amount; This represents the economic objective function. Cost per unit of electricity generated, for Power output at all times The charging and discharging cost of energy storage units, for Energy storage charging and discharging power at all times, To reduce costs per unit load, for Load reduction amount at any time; Represents the reliability objective function. for Energy storage and discharge power at all times for Constant load demand, It is a very small positive number; This represents the environmental protection objective function. Emissions per unit of electricity generated The equivalent emissions reduction per unit load; These are the minimum and maximum constraints for the power output, respectively. These are the minimum and maximum constraints for load demand, respectively. The apparent power of the transmission line. This is a constraint on the maximum apparent power of transmission lines.

5. The resource scheduling system based on smart grid of claim 1, wherein, The data acquisition and transmission model expression of the power grid cloud-edge collaborative data acquisition module is as follows: ,in, express Aggregated data received in real time from the cloud; Indicates the total number of edge nodes; express Time of the first The data weight coefficient of each edge node is related to the communication bandwidth and data integrity of the edge node; express Time of the first Raw data collected by each edge node, including distributed power output and load demand data; express Time of the first The data noise coefficient of each edge node; express Time of the first Noise data for each edge node; This represents matrix multiplication. express Time of the first A data correction matrix for each edge node is used to eliminate systematic errors during the data acquisition process.

6. The resource scheduling system based on smart grid of claim 1, wherein, The instruction conversion model expression of the resource scheduling instruction generation module is: ,in, express At any time for the first Scheduling instructions for each execution unit; This represents the total number of smart grid execution units, including distributed power sources, energy storage devices, and load nodes. Indicates the first The instruction translation coefficient matrix of each execution unit is related to the type and rated parameters of the execution unit; express At any time for the first Pareto optimal scheduling parameters for each execution unit; Indicates the first Feedback adjustment coefficient matrix for each execution unit; express Time of the first Feedback data from each execution unit, including actual operating power, voltage, and current data; This represents the Hadamard product operation; express Time of the first The state coefficient of each execution unit is set to 1 when the execution unit operates normally and 0.5 when the execution unit operates at a reduced rate.

7. The resource scheduling system based on smart grid of claim 1, wherein, The high-dimensional target Pareto scheduling analysis module includes a Pareto solution generation unit, a non-dominated sorting calculation unit, an optimal solution screening unit and a solution stability verification unit; the Pareto solution generation unit receives the scheduling decision variables, combines the economic, reliability and environmental protection target parameters, iterates under multi-objective optimization constraints, generates an initial Pareto solution set, substitutes the target function and excludes solutions exceeding the parameter boundary; the non-dominated sorting calculation unit judges the dominance relationship of each solution in the initial solution set, compares the target function values to divide the non-dominated levels, calculates the crowding distance and considers the distance difference between adjacent solutions in each target dimension; The optimal solution screening unit selects candidate optimal solutions according to the sorting results and crowding distance, and verifies and excludes solutions that do not conform to real-time constraints in combination with the transmission line load rate and the remaining capacity of the energy storage device; The solution stability verification unit substitutes the optimal solution into the model, changes the power output and load demand fluctuation, observes the change amplitude of the target function value, and returns to regenerate the solution set if the threshold is exceeded.

8. The resource scheduling system based on smart grid of claim 1, wherein, The power grid cloud edge collaborative data acquisition module comprises an edge data acquisition unit, a data preprocessing unit, a cloud data receiving unit, and a data interaction control unit; the edge data acquisition unit collects power output, load demand, line current and voltage, and energy storage charging and discharging state data at a preset interval through sensors and intelligent instruments; The data preprocessing unit identifies abnormal values of original data using a statistical bias method, replaces the abnormal values with adjacent data interpolation, and converts heterogeneous data into a unified format; the cloud data receiving unit receives standardized data through an encrypted link, calculates a checksum to compare the data from the sending end, and requests retransmission if the data are inconsistent; the data interaction control unit dynamically adjusts the data transmission frequency and priority based on the communication state of the edge node and the cloud load, reduces the frequency of non-standard data when the bandwidth is insufficient, and preferentially receives standard data when the load is high.

9. The resource scheduling system based on smart grid of claim 1, wherein, The double-time-scale random scheduling modeling module comprises a hourly model construction unit, a minute-level model construction unit, a scale coordination unit, and a model parameter updating unit; the hourly model construction unit determines hourly scheduling segments, load and power output prediction values based on historical and real-time data, and constructs a model with power output, line capacity constraints and minimum operating cost as the target; The minute-level model construction unit divides each hour into minute-level periods, collects high-frequency fluctuation data of load and power output, constructs a model with tracking hourly plan as the target and considering the energy storage response capability; The scale coordination unit couples the hourly model output as the boundary condition of the minute-level model and sets the energy storage charging and discharging plan adjustment amount for model coupling and coordination; The model parameter updating unit rolls over and corrects the load prediction error and power output fluctuation coefficient in the model based on the feedback bias data through statistical analysis.

10. A resource scheduling method based on a smart grid, characterized in that, The method is applied to the resource scheduling system based on a smart grid of claim 1, comprising the following steps: first, the power grid cloud edge collaborative data acquisition module collects power output, load demand, line impedance and energy storage state data through edge nodes, transmits the data to the cloud server for aggregation and unified format, and forms a standardized data source; second, the double-time-scale random scheduling modeling module calls the data source, extracts hourly and minute-level calibration parameters, constructs a random scheduling model containing two sub-models, embeds the power output, load demand fluctuation probability distribution function, determines the objective function and constraint conditions; third, the random saddle point approximation optimization calculation module receives the model parameters, initializes the iteration variables and times, iteratively optimizes according to the algorithm steps, calculates the objective function value and constraint satisfaction each time, and outputs the scheduling decision variables after convergence or reaching the number of times; fourth, the high-dimensional target Pareto scheduling analysis module substitutes the decision variables into the model, calculates the objective function value, divides the levels by non-dominated sorting and calculates the crowding distance, and selects the optimal scheduling scheme; fifth, the resource scheduling instruction generation module decomposes the optimal scheme into control parameters, converts them into scheduling instructions according to the execution unit protocol, and transmits them to each execution unit through the communication network; sixth, the scheduling effect feedback adjustment module collects the actual running data of the execution unit, compares it with the expected data to calculate the bias, classifies and transmits it to each module, and each module adjusts the parameters or logic according to the bias for dynamic optimization.

Images