A power grid dispatching backtracking verification method, device and equipment and a storage medium

Abstract

The application provides a power grid scheduling backtracking verification method and device, equipment and a storage medium, the method comprising: obtaining a plurality of data entries carrying time stamps from a data acquisition unit, and determining whether all time stamps are not greater than the current scheduling time by a non-expected constraint determination unit; after verification, a convex quadratic programming model is constructed by a convex optimization modeling unit, and an optimal scheduling decision sequence is generated by a solving engine; then, the time stamps of the data used in the solving process are backtracked and verified by a result verification unit, and the scheduling instruction is output after confirming that all time stamps are not greater than the current time. The method ensures that the scheduling is based on current and historical real-time data through double time stamp verification, avoids the robustness problem caused by relying on future prediction, and guarantees a global optimal solution by using convex optimization, thereby meeting the real-time requirements of hour-level rolling scheduling.

Description

Technical Field

[0001] This application belongs to the field of power grid dispatching, and in particular relates to a power grid dispatching backtracking verification method, device, equipment and storage medium. Background Technology

[0002] like Figure 3 and Figure 4 As shown, with the high proportion of new energy grid connection, large-scale access of electric vehicles, and large-scale operation of high-power fast charging facilities, the fluctuation range of grid load has increased significantly, and the inertia level has continued to decline. Frequent start-ups and shutdowns of thermal power units and peak shaving have led to high power generation dispatch costs, and the contradiction between grid operation stability and economy has become increasingly prominent. Energy storage systems, with their rapid charging and discharging capabilities, have become a core means to smooth load fluctuations and reduce dispatch costs, and energy storage economic dispatch technology has become a core research direction for smart grid operation control.

[0003] Existing economic dispatch methods for energy storage mostly employ a predictive data plus deterministic optimization architecture. This involves acquiring 24-hour load forecast data in advance to build an optimization model and solve for the optimal charging and discharging strategy, which is currently the most mainstream technical approach in engineering applications. The closest existing technology to this invention is CN202510059352.2, a method for economic dispatch of shared energy storage power stations participating in grid ancillary services. This method is based on a renewable energy output probability model and an electricity and heat load probability model. It uses Latin hypercube sampling to randomly sample scenario samples from different time periods within a day, identifying several representative scenarios. These representative scenarios are then pieced together to form a daily scenario curve set, which is then input into a daily operating cost model to obtain the cost-optimal operating strategy for the shared energy storage power station. This method can, to some extent, solve problems such as the impact of weather on wind and solar power output and fluctuations in daily real-time load dispatch, thereby improving the utilization rate of renewable energy and reducing the cost of power generation.

[0004] The existing technology CN202510059352.2 employs a probabilistic model sampling, daily scenario stitching, and offline cost optimization approach. While it can achieve economic dispatch in shared energy storage auxiliary service scenarios, its architecture and model design have inherent flaws: It lacks unexpected causal constraints, and its decisions rely on future predicted scenarios. The optimization decisions of this method are based on the sampling scenario curves of the entire future 24 hours, essentially relying on future prediction data, which violates the causal logic of real-time grid operation. In actual operation, when the output of new energy sources and the load level deviate from the predicted scenario, the adaptability of the dispatch strategy decreases significantly, the model's robustness is poor, and it cannot be directly used for real-time online dispatch. Furthermore, the stochastic programming model is non-convex, resulting in insufficient solution efficiency and optimality. This method uses a multi-scenario stochastic programming architecture, and the overall model is non-convex, making it prone to getting trapped in local optima. Moreover, Latin hypercube sampling and multi-scenario stitching involve large computational loads and long processing times, making it difficult to meet the real-time requirements of hourly rolling dispatch and unable to adapt to the needs of online grid control. Summary of the Invention

[0005] The purpose of this application is to overcome the deficiencies in the prior art and provide a power grid dispatch backtracking verification method, device, equipment and storage medium.

[0006] This application provides a power grid dispatch backtracking verification method, including:

[0007] Multiple data entries are acquired from the data acquisition unit, each data entry carrying a collection timestamp;

[0008] The unexpected constraint determination unit verifies the multiple data entries and generates a first verification result. The first verification result is used to indicate whether the collection timestamps of the multiple data entries are all not greater than the current scheduling time.

[0009] If the first verification result is yes, the convex optimization modeling unit constructs a convex quadratic programming model based on the multiple data entries, and the solution engine solves the convex quadratic programming model to generate the optimal scheduling decision sequence.

[0010] The result verification unit performs backtracking verification on the solution process of the solution engine and generates a second verification result. The second verification result is used to indicate whether the collection timestamp of all data entries called in the solution process is not greater than the current scheduling time.

[0011] If the second verification result is yes, the optimal scheduling decision sequence is output as a scheduling instruction.

[0012] Optionally, multiple data entries are acquired from the data acquisition unit, each data entry carrying an acquisition timestamp, including:

[0013] While writing each data entry to the local cache, the data acquisition unit generates and binds a native hard timestamp for that data entry. The native hard timestamp is directly generated by the hardware clock of the data acquisition unit at the time of writing.

[0014] When verifying the multiple data entries, the unexpected constraint determination unit reads the native hard timestamp bound to each data entry one by one and compares the native hard timestamp with the current scheduling time.

[0015] Optionally, the result verification unit performs backtracking verification on the solution process of the solution engine to generate a second verification result, including:

[0016] The solver engine records a call record each time a data entry is invoked. The call record contains the data identifier of the invoked data entry and the acquisition timestamp carried by the data entry. All call records constitute the data source call log.

[0017] When performing backtracking verification, the result verification unit reads all call records in the data source call log and checks one by one whether the collection timestamp in each call record is not greater than the current scheduling time.

[0018] Optionally, a convex optimization modeling unit constructs a convex quadratic programming model based on the plurality of data entries, including:

[0019] The convex quadratic programming model includes unit ramping constraints and energy storage state of charge time-series recursive constraints. The unit ramping constraints are linear inequalities used to limit the variation in unit output between adjacent scheduling times. The energy storage state of charge time-series recursive constraints are linear equations used to calculate the state of charge at the current scheduling time based on the state of charge at the previous scheduling time and the charging and discharging power at the current scheduling time.

[0020] The solver engine simultaneously satisfies the unit ramp-up constraint and the energy storage state-of-charge timing recursion constraint during the same solver process.

[0021] Optionally, the result verification unit performs backtracking verification on the solution process of the solution engine to generate a second verification result, including:

[0022] Before the result verification unit performs the backtracking verification, the result verification unit also performs first-level verification and second-level verification;

[0023] The first-level verification is used to check whether all decision variables in the optimal scheduling decision sequence satisfy the constraints of the convex quadratic programming model. If the degree of violation of any constraint exceeds the first tolerance, the optimal scheduling decision sequence is determined to be invalid.

[0024] The secondary verification is used to calculate the power balance residual, which is the absolute value of the difference between the algebraic sum of the output of all units and the charging and discharging power of energy storage and the load demand. If the power balance residual exceeds the second tolerance, the optimal scheduling decision sequence is determined to be invalid.

[0025] The result verification unit performs the backtracking verification only after both the first-level verification and the second-level verification have passed.

[0026] Optionally, the solution engine solves the convex quadratic programming model to generate an optimal scheduling decision sequence, including:

[0027] The solution engine employs the original-dual interior-point method and enables a numerical stability correction mechanism during the solution process.

[0028] The numerical stability correction mechanism is used to adjust the iteration step size or introduce a regularization term when numerical ill-conditioning is detected, so as to ensure that the solution process converges to a feasible solution.

[0029] Optionally, the optimal scheduling decision sequence is output as a scheduling instruction, including:

[0030] After outputting the optimal scheduling decision sequence, the load period division unit calculates the daily average load based on the actual load data of the previous day in the multiple data entries, and divides the load of each scheduling time of the day into low period, peak period and flat load period based on the daily average load.

[0031] Based on the optimal scheduling decision sequence and the load period division results, a differentiated control strategy is generated. The differentiated control strategy includes: issuing charging commands to the energy storage system during the off-peak period; issuing discharging commands to the energy storage system during the peak period; and keeping the energy storage system idle during the flat load period.

[0032] This application also provides a power grid dispatch retrospective verification device, comprising:

[0033] The acquisition module is used to acquire multiple data entries from the data acquisition unit, each data entry carrying an acquisition timestamp;

[0034] The first verification module is used to verify the multiple data entries by the unexpected constraint determination unit and generate a first verification result. The first verification result is used to indicate whether the collection timestamps of the multiple data entries are all not greater than the current scheduling time.

[0035] The decision module is used to construct a convex quadratic programming model by the convex optimization modeling unit based on the multiple data entries when the first verification result is yes, and to solve the convex quadratic programming model by the solution engine to generate the optimal scheduling decision sequence.

[0036] The second verification module is used to backtrack and verify the solution process of the solution engine by the result verification unit, and generate a second verification result. The second verification result is used to indicate whether the collection timestamp of all data entries called in the solution process is not greater than the current scheduling time.

[0037] The instruction module is used to output the optimal scheduling decision sequence as a scheduling instruction when the second verification result is yes.

[0038] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the methods described above.

[0039] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform any of the methods described above.

[0040] The beneficial effects of this application are:

[0041] This application provides a power grid dispatch backtracking verification method, comprising: acquiring multiple data entries from a data acquisition unit, each data entry carrying an acquisition timestamp; verifying the multiple data entries by an unexpected constraint determination unit to generate a first verification result, the first verification result indicating whether the acquisition timestamps of the multiple data entries are all not greater than the current dispatch time; if the first verification result is yes, constructing a convex quadratic programming model by a convex optimization modeling unit based on the multiple data entries, and solving the convex quadratic programming model by a solution engine to generate an optimal dispatching decision sequence; backtracking verification of the solution process of the solution engine by a result verification unit to generate a second verification result, the second verification result indicating whether the acquisition timestamps of all data entries called in the solution process are all not greater than the current dispatch time; and outputting the optimal dispatching decision sequence as a dispatching instruction if the second verification result is yes. This application verifies the data entry timestamps by using an unexpected constraint judgment unit to ensure that scheduling decisions are based solely on current and historical real-time data. After verification, a convex optimization modeling unit constructs a convex quadratic programming model for solution. This achieves that the decision fully conforms to the causal logic of the real-time operation of the power grid, avoiding the problem of poor model robustness caused by relying on future prediction data. At the same time, convex optimization ensures the global optimum solution, meeting the real-time requirements of hourly rolling scheduling. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the power grid dispatching process in this application;

[0043] Figure 2 This is a schematic diagram of the improved IEEE 9-node power system described in this application.

[0044] Figure 3 This is a schematic diagram showing the power output of all four generators in the energy storage-free system of this application.

[0045] Figure 4 This is a schematic diagram showing the power output of all four generators in the energy storage system within this application. Detailed Implementation

[0046] Exemplary embodiments of the present disclosure will now be provided in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is to be understood that various forms of implementation of the present disclosure are intended and should not be limited to the embodiments set forth herein. Rather, the embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0047] This application provides a power grid dispatch backtracking verification method, which is applied to the field of economic dispatch control of smart grid energy storage systems. It is used to solve the problems of existing dispatch models relying on future prediction data, which leads to decisions that violate causal logic, non-convex models being prone to getting trapped in local optima, insufficient source-storage coordination, and lack of closed-loop control mechanisms.

[0048] Please refer to Figure 1 and Figure 2 As shown, the power grid dispatch backtracking verification method described in this application includes:

[0049] S101. Obtain multiple data entries from the data acquisition unit, each data entry carrying an acquisition timestamp.

[0050] The data acquisition unit consists of a power grid monitoring and control terminal, a generator data acquisition device, and an energy storage BMS system, and is responsible for collecting real-time operating data of the power grid, generator sets, and energy storage system.

[0051] On the power grid side, real-time load data for the entire network is collected. Bus voltage and system frequency; on the generator side, real-time output of four generator units is collected. Unit ramp rate and operating status;

[0052] On the energy storage side, the real-time state of charge of the energy storage system is collected. Charging and discharging power and Equipment operating status.

[0053] The data acquisition unit only caches data from the current moment and historical moments, and automatically filters out future prediction data, providing a data foundation for unexpected constraints.

[0054] While writing each data entry to the local cache, the data acquisition unit generates and binds a native hard timestamp for that data entry. The native hard timestamp is directly generated by the hardware clock of the data acquisition unit at the time of writing.

[0055] For each scheduling time ,in Values Data acquisition and preprocessing are performed cyclically on an hourly basis. The real-time data acquisition unit simultaneously collects three types of dynamic data, including the total active load of the entire network. Actual output of the 4 generator units at the previous moment Current state of charge of energy storage system With real-time charge and discharge power and .

[0056] Regarding data caching management, only data with timestamps less than or equal to the current time is cached. Real-time and historical data are stored in a local cache, automatically intercepting and blocking all timestamps greater than [a certain value]. Load forecasting and output forecasting data are used to implement unexpected constraints at the data source level.

[0057] In terms of data preprocessing, the following methods are adopted: The criteria remove abnormal jump data, complete data time sequence alignment and validity verification, and ensure that the data accuracy of the input model meets the scheduling requirements.

[0058] After the system is powered on, a full module hardware self-test is first performed to confirm that the real-time data acquisition unit, unexpected constraint judgment unit, convex optimization modeling unit, optimization solution unit, scheduling execution unit, and status feedback monitoring unit are communicating normally and have no hardware faults. Then, the static basic parameters of the system are entered.

[0059] Regarding generator set parameters, enter the following in sequence: tower( The power generation cost coefficient of thermal power units , , Upper and lower limits of contribution , Upper limit of ramp speed per unit time Lower limit of ramp rate per unit time .

[0060] Regarding energy storage system parameters, the maximum charging power of the energy storage system was recorded. Maximum discharge power Upper and lower limits of state of charge , Charging efficiency Discharge efficiency Initial state of charge Rated capacity .

[0061] Regarding scheduling rule configuration, set the global optimization cycle. Single-step scheduling time Control parameters such as convergence accuracy threshold, constraint violation threshold, and voltage deviation threshold.

[0062] After the parameters are configured and verified to be correct, the system automatically enters normal operation mode. The entire system relies on the improved IEEE 9-bus power system as the application platform. The system is configured with four thermal power generating units G1~G4 and four energy storage systems B1 at the bus. Through unexpected constraint control, convex optimization global solution, and source-storage coordinated regulation, the system achieves economical and efficient operation of the power grid.

[0063] S102. The unexpected constraint determination unit verifies the plurality of data entries and generates a first verification result. The first verification result is used to indicate whether the collection timestamps of the plurality of data entries are all not greater than the current scheduling time.

[0064] The unexpected constraint determination unit is the core constraint control module, responsible for verifying the compliance of the data source for scheduling decisions. This unit has built-in unexpected constraint determination logic, meaning that at any given time... The scheduling decision only allows the use of to Importing real-time or historical data that has already been collected is prohibited. And future forecast data.

[0065] When verifying the multiple data entries, the unexpected constraint determination unit reads the native hard timestamp bound to each data entry one by one and compares the native hard timestamp with the current scheduling time.

[0066] Specifically, the unexpected constraint determination unit traverses all data entries to be called by the current modeling unit, checks the data collection timestamp of each entry, and verifies that the timestamps of all data meet the requirements. This ensures that no data from future moments enters the modeling process.

[0067] In terms of verifying the causal logic of decisions, verify unexpected constraint expressions:

[0068]

[0069] This principle establishes that scheduling decisions at any given time can only be based on information acquired up to that time and cannot utilize future forecast data, ensuring that decisions conform to the causal logic of the actual operation of the power grid.

[0070] in, For the first The entire dataset on which time-of-slot scheduling decisions depend; For the first Time to the A collection of all historical and real-time data collected at any given moment.

[0071] Regarding the verification result judgment and branch processing, if all data meets the timestamp requirements and the decision-making link does not call future data, a "verification qualified" flag is generated, and the system enters the convex optimization model construction stage; if future time data is detected, a data compliance alarm is immediately triggered, the current modeling process is terminated, and data collection and caching are re-executed until the verification is passed.

[0072] This unit performs real-time backtracking of the data source call logs for modeling. If it detects the introduction of future data, it immediately triggers an interception alarm, terminates the modeling process, and re-collects data to ensure that the scheduling decisions conform to the causal logic of the real-time operation of the power grid.

[0073] S103. If the first verification result is yes, the convex optimization modeling unit constructs a convex quadratic programming model based on the multiple data entries, and the solution engine solves the convex quadratic programming model to generate the optimal scheduling decision sequence.

[0074] The convex optimization modeling unit is responsible for constructing a standard convex optimization scheduling model with the goal of minimizing the total power generation cost of the entire network. All constraints are linear equations or inequalities, and the whole problem is a convex quadratic programming problem, ensuring a globally optimal solution.

[0075] The convex quadratic programming model includes unit ramping constraints and energy storage state of charge time-series recursive constraints. The unit ramping constraints are linear inequalities used to limit the variation in unit output between adjacent scheduling times. The energy storage state of charge time-series recursive constraints are linear equations used to calculate the state of charge at the current scheduling time based on the state of charge at the previous scheduling time and the charging and discharging power at the current scheduling time.

[0076] The solver engine simultaneously satisfies the unit ramp-up constraint and the energy storage state-of-charge timing recursion constraint during the same solver process.

[0077] The convex optimization modeling unit explicitly defines all continuous decision variables in the model, including the unit's active power generation. Energy storage charging power Energy storage and discharge power Energy storage state of charge .

[0078] Regarding the construction of the objective function, the optimization objective is to minimize the total power generation cost of all thermal power units in the entire network within a 24-hour period. A quadratic convex objective function is constructed, which is as follows:

[0079]

[0080] in, The total power generation dispatch cost for the entire network over 24 hours is expressed in yuan. , , The first The cost coefficients of the secondary, primary, and constant terms of the thermal power unit are obtained by fitting the coal consumption characteristics of the unit. For the first Taiwanese crew Active power generation at any given time, measured in MW.

[0081] In terms of constructing constraints one by one, the power balance equation constraint, the upper and lower limits of unit output and the ramp inequality constraint, the energy storage charging and discharging power, the SOC boundary and the time sequence recursion inequality constraint are constructed in sequence. All constraints are linear in form and constitute a convex feasible region.

[0082] The power balance constraint is a linear equation:

[0083]

[0084] in, for Total active power load of the entire network at any time, in MW; For energy storage charging power, The values ​​represent energy storage discharge power, all in MW.

[0085] The unit output constraint is a linear inequality, including:

[0086]

[0087] in, and The upper and lower limits of the unit's output. and These are the upper and lower limits for the unit's ramp rate, both in MW / h.

[0088] The constraints for energy storage operation are linear inequalities, including:

[0089]

[0090]

[0091]

[0092] in, for The state of charge of the stored energy at any given time, with a value range of [value range missing]. ; and These are the upper and lower limits of the energy storage's state of charge; and This refers to the maximum charge and discharge power of energy storage, measured in MW. and The energy storage charging and discharging efficiency is defined as follows: ; This refers to the rated capacity of the energy storage system, expressed in MWh.

[0093] In terms of model convexity verification, the second-order partial derivatives of the objective function were calculated to verify that the Hessian matrix is ​​positive semi-definite, confirming that it is a strictly convex quadratic function; all constraints were determined to be linear convex functions, and the feasible region is a convex set; finally, the model was confirmed to be a standard convex quadratic programming problem, which can guarantee that the solution can obtain a globally unique optimal solution.

[0094] In terms of model standardization and regularization, the objective function and all constraints are uniformly regularized into the standard form of convex quadratic programming. The objective function is regularized into a standard format containing a quadratic coefficient matrix and a linear coefficient vector. Equality constraints and inequality constraints are regularized into standard forms of coefficient matrices and constant vectors, respectively. The upper and lower bound parameters of decision variables are encapsulated to form a standardized model file that can be directly input into the solution engine.

[0095] The solution engine has a built-in embedded convex optimization solution engine, which uses the original dual interior point method as the core algorithm and is responsible for solving the convex optimization model and verifying the results.

[0096] The solver engine enables a numerical stability correction mechanism during the solution process. This mechanism is used to adjust the iteration step size or introduce a regularization term when numerical ill-conditioning is detected, so as to ensure that the solution process converges to a feasible solution.

[0097] After the solver engine completes initialization, the core solver algorithm is configured as the primal-dual interior-point method, and the running parameters are set synchronously, including the convergence accuracy threshold. The maximum number of iterations is Next, enable the numerical stability correction mechanism and disable the future data extrapolation channel; enable parallel computing mode to adapt to the computing needs of the 24-hour long-cycle scheduling model.

[0098] In terms of execution of the primal-dual interior-point method iterative solution, relaxation variables are introduced to transform all inequality constraints into equality constraints, and the Lagrangian function of the primal-dual problem is constructed. Based on the KKT optimality condition, a system of Newton's linear equations is established to solve for the update directions of the primal and dual variables in the current iteration step. The optimal iteration step size is calculated using a line search method to ensure that all variables remain within the feasible region during the iteration process. The primal, dual, and complementary gap variables are updated iteratively, and the three convergence indices of primal residual, dual residual, and complementary gap are monitored in real time. When all three indices are less than the convergence accuracy threshold, or when the maximum number of iterations is reached, the iteration is automatically terminated, and the solution result is output.

[0099] The solver engine internally comprises a model warping module, a parameter configuration module, an iterative solution module, and a result verification module. The model warping module is responsible for converting the convex optimization model into a standard convex quadratic programming format compatible with the solver engine; the parameter configuration module is responsible for customizing the solver parameters to adapt to the low latency requirements of hourly grid scheduling; the iterative solution module uses the primal-dual interior-point method to iteratively solve for the global optimum; and the result verification module is responsible for three levels of verification of the constraint compliance, power balance, and compliance with unexpected constraints of the solution results.

[0100] S104. The result verification unit performs backtracking verification on the solution process of the solution engine and generates a second verification result. The second verification result is used to indicate whether the collection timestamp of all data entries called in the solution process is not greater than the current scheduling time.

[0101] The result verification unit performs three levels of verification.

[0102] Before the result verification unit performs the backtracking verification, the result verification unit also performs first-level verification and second-level verification.

[0103] The first-level verification is used to check whether all decision variables in the optimal scheduling decision sequence satisfy the constraints of the convex quadratic programming model. If the degree of violation of any constraint exceeds the first tolerance, the optimal scheduling decision sequence is determined to be invalid.

[0104] The first-level verification is a constraint compliance verification, which verifies each decision variable output from the solution to ensure that it satisfies all operational constraints. The constraint violation threshold is set to... ; Verify whether the unit output is within the acceptable range. Within the range, check whether the ramp rate meets the limit requirements; verify whether the energy storage charging and discharging power does not exceed the rated value, and whether the state of charge is within the specified range. Within the interval; if there is a constraint exceeding the limit and the violation degree exceeds the threshold, the solution is deemed invalid, and the convergence accuracy is automatically adjusted before restarting the solution.

[0105] The secondary verification is used to calculate the power balance residual, which is the absolute value of the difference between the algebraic sum of the output of all units and the charging and discharging power of energy storage and the load demand. If the power balance residual exceeds the second tolerance, the optimal scheduling decision sequence is determined to be invalid.

[0106] The second-level verification is the power balance accuracy verification, which verifies the degree to which the power balance constraints of the entire network are met and calculates the power balance residuals:

[0107]

[0108] in, For the first The power balance residual of the entire network at the time of dispatch, in MW, represents the magnitude of the deviation between the total output of the generating units, the net output power of energy storage and the load of the entire network; The summation operator calculates the cumulative active power output of all four generator sets in the system. For the first tower( Thermal power units in Active power generation at any given time, measured in MW; For energy storage systems in The discharge power at time t is a non-negative continuous variable, and its unit is MW; For energy storage systems in The charging power at any given time is a non-negative continuous variable, and its unit is MW; For the first Total active power load of the entire network at the time of scheduling, in MW; This is an absolute value operator used to calculate the absolute value of power deviation. It only measures the magnitude of the deviation and does not distinguish between the direction of power surplus or deficit.

[0109] like If the power balance is deemed acceptable, the solution is deemed invalid; if the residual exceeds the limit, the solution is returned to the corrected model coefficients and the solution is solved again.

[0110] The result verification unit performs the backtracking verification only after both the first-level verification and the second-level verification have passed.

[0111] The solver engine records a call record each time a data entry is invoked. This call record includes the data identifier of the invoked data entry and the data acquisition timestamp carried by that data entry. All call records constitute the data source call log. During backtracking verification, the result verification unit reads all call records in the data source call log and checks each call record's data acquisition timestamp against the current scheduling time. The third-level verification is an unexpected constraint backtracking verification, which backtracks the entire solver's data source call log to confirm that the solver did not invoke any future time prediction data, and the decision was entirely based on... to The data collected at any given time strictly meets the requirements for causal decision-making.

[0112] S105. If the second verification result is yes, output the optimal scheduling decision sequence as a scheduling instruction.

[0113] After all three levels of verification are passed, the analytical solution results generate the optimal scheduling instruction set, which includes the hourly optimal output sequence of the four generating units, the hourly charging and discharging power sequence of the energy storage, and the hourly state of charge sequence of the energy storage, and is output to the scheduling execution unit.

[0114] The scheduling execution unit parses the optimal scheduling command into control signals recognizable by the equipment and sends them to the unit controllers of the four generator sets and the PCS controller of the energy storage converter, respectively, confirming that the equipment has received the command and feeding back a ready signal. After outputting the optimal scheduling decision sequence, the load period division unit calculates the daily average load based on the actual load data of the previous day from the multiple data entries, and divides the load at each scheduling time of the day into low-peak periods, peak periods, and flat-load periods based on the daily average load.

[0115] Specifically, the daily average load is obtained based on the actual load statistics of the previous day. Runtime segments are divided according to the following rules: when When, it is determined to be a period of low load; when When the load is at its peak, it is considered a peak load period; the rest of the time is considered a normal load period.

[0116] Based on the optimal scheduling decision sequence and the load time period division results, a differentiated control strategy is generated, the differentiated control strategy including:

[0117] During the off-peak period, a charging command is issued to the energy storage system;

[0118] During the peak period, a discharge command is issued to the energy storage system;

[0119] During the flat load period, the energy storage system is kept idle.

[0120] In terms of time-of-use differentiated regulation, during off-peak hours, the four generating units are controlled to maintain their low-cost output range, and high-cost units are not started. Charging commands are issued to the energy storage PCS, and the energy storage is controlled to absorb surplus power from the grid at the optimal charging power. The state of charge of the energy storage is gradually increased, realizing energy storage during low-price periods. During peak hours, the operating generating units are controlled to maintain their current output without increasing output, avoiding the start-up of high-cost units. Discharge commands are issued to the energy storage PCS, and the energy storage is controlled to transmit power to the outside at the optimal discharge power, sharing the grid load gap and realizing discharge decompression during high-price periods. During flat-load periods, the generating units are controlled to output steadily, the energy storage is kept in a static state, the charging and discharging power is 0, the state of charge is kept constant, and the system is operating in a low-loss steady state.

[0121] Regarding the confirmation of execution effect, within 1 minute after the instruction is issued, the actual output of the unit and the actual charging and discharging power of the energy storage are collected and compared with the dispatch instruction. If the deviation is less than 2%, it is determined that the execution is in place; if the deviation exceeds the standard, a secondary adjustment is triggered to ensure that the dispatch instruction is accurately implemented.

[0122] The dispatch execution unit consists of a generator set controller and an energy storage converter (PCS), responsible for receiving optimal dispatch instructions and executing on-site control. In terms of generator set control, it issues power regulation instructions to the four generator sets to control their operation within the low-cost range; in terms of energy storage control, it issues charging and discharging instructions to the energy storage PCS, enabling charging during off-peak hours and discharging during peak hours, achieving source-storage synergy.

[0123] The status feedback monitoring unit continuously collects three types of operating data at a sampling frequency of minutes: voltage of each bus on the grid side, system frequency, and total load; actual output of the four generating units on the generator side, main steam pressure, and equipment operating status; and real-time state of charge, charging and discharging power, battery cell temperature, and SOC deviation on the energy storage side.

[0124] Regarding limit exceedance detection and anomaly identification, the operating status is determined one by one according to the preset thresholds. Scenarios that trigger anomalies include: unit output exceeding limits ( or ); Unit ramping exceeds limit, hourly output change exceeds ramping limit; Energy storage SOC exceeds limit ( or ); Grid voltage exceeds limits, bus voltage deviation exceeds rated value. .

[0125] Regarding dynamic constraint correction and re-optimization triggering, if parameter exceedance or abnormal operating conditions are detected, the current abnormal boundary is immediately locked, and the corresponding constraint conditions in the model are dynamically corrected; the instant re-optimization process is triggered, returning to the convex optimization model construction stage to rebuild the model, solve, and issue instructions to quickly eliminate the exceedance state; if the abnormality cannot be eliminated through scheduling adjustment, the equipment protection alarm is immediately triggered to notify the operation and maintenance personnel to intervene.

[0126] Regarding steady-state maintenance and timing progression, if all parameters are within the normal range, the system maintains its current scheduling state; when the next scheduling time arrives... When the time comes, it will automatically jump to the data acquisition and caching stage and enter the next round of scheduling.

[0127] The key point of this invention is to construct an integrated energy storage economic dispatch architecture. By using timestamp verification to limit decision-making to using only current and historical data, and adopting convex quadratic programming modeling across the entire domain, it takes into account both the causal authenticity of decision-making and the global optimal solution. It designs a unit-energy storage joint constraint system, which integrates unit ramp-up constraints, output boundaries, and energy storage SOC and charge / discharge power constraints into a unified convex optimization model to achieve refined regulation of source and storage coordination.

[0128] This application also provides a power grid dispatch retrospective verification device, comprising:

[0129] The acquisition module is used to acquire multiple data entries from the data acquisition unit, each data entry carrying an acquisition timestamp;

[0130] The first verification module is used to verify the multiple data entries by the unexpected constraint determination unit and generate a first verification result. The first verification result is used to indicate whether the collection timestamps of the multiple data entries are all not greater than the current scheduling time.

[0131] The decision module is used to construct a convex quadratic programming model by the convex optimization modeling unit based on the multiple data entries when the first verification result is yes, and to solve the convex quadratic programming model by the solution engine to generate the optimal scheduling decision sequence.

[0132] The second verification module is used to backtrack and verify the solution process of the solution engine by the result verification unit, and generate a second verification result. The second verification result is used to indicate whether the collection timestamp of all data entries called in the solution process is not greater than the current scheduling time.

[0133] The instruction module is used to output the optimal scheduling decision sequence as a scheduling instruction when the second verification result is yes.

[0134] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0135] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the above-described method.

[0136] The above embodiments are provided to enable those skilled in the art to understand and apply this application. Those skilled in the art will readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without inventive effort. Therefore, this application is not limited to the above embodiments, and any improvements and modifications made to this application based on the disclosure thereof should be within the scope of protection of this application.

Claims

1. A power grid dispatch backtracking verification method, characterized in that, include: Multiple data entries are acquired from the data acquisition unit, each data entry carrying a collection timestamp; The unexpected constraint determination unit verifies the multiple data entries and generates a first verification result. The first verification result is used to indicate whether the collection timestamps of the multiple data entries are all not greater than the current scheduling time. If the first verification result is yes, the convex optimization modeling unit constructs a convex quadratic programming model based on the multiple data entries, and the solution engine solves the convex quadratic programming model to generate the optimal scheduling decision sequence. The result verification unit performs backtracking verification on the solution process of the solution engine and generates a second verification result. The second verification result is used to indicate whether the collection timestamp of all data entries called in the solution process is not greater than the current scheduling time. If the second verification result is yes, the optimal scheduling decision sequence is output as a scheduling instruction.

2. The method according to claim 1, characterized in that, Multiple data entries are acquired from the data acquisition unit, each data entry carrying an acquisition timestamp, including: While writing each data entry to the local cache, the data acquisition unit generates and binds a native hard timestamp for that data entry. The native hard timestamp is directly generated by the hardware clock of the data acquisition unit at the time of writing. When verifying the multiple data entries, the unexpected constraint determination unit reads the native hard timestamp bound to each data entry one by one and compares the native hard timestamp with the current scheduling time.

3. The method according to claim 1, characterized in that, The result verification unit performs backtracking verification on the solution process of the solution engine, generating a second verification result, including: The solver engine records a call record each time a data entry is called. The call record contains the data identifier of the called data entry and the collection timestamp carried by the data entry. All call records constitute the data source call log. When performing backtracking verification, the result verification unit reads all call records in the data source call log and checks one by one whether the collection timestamp in each call record is not greater than the current scheduling time.

4. The method according to claim 1, characterized in that, A convex quadratic programming model is constructed by the convex optimization modeling unit based on the multiple data entries, including: The convex quadratic programming model includes unit ramping constraints and energy storage state of charge time-series recursive constraints. The unit ramping constraints are linear inequalities used to limit the variation in unit output between adjacent scheduling times. The energy storage state of charge time-series recursive constraints are linear equations used to calculate the state of charge at the current scheduling time based on the state of charge at the previous scheduling time and the charging and discharging power at the current scheduling time. The solver engine simultaneously satisfies the unit ramp-up constraint and the energy storage state-of-charge timing recursion constraint during the same solver process.

5. The method according to claim 1, characterized in that, The result verification unit performs backtracking verification on the solution process of the solution engine, generating a second verification result, including: Before the result verification unit performs the backtracking verification, the result verification unit also performs first-level verification and second-level verification; The first-level verification is used to check whether all decision variables in the optimal scheduling decision sequence satisfy the constraints of the convex quadratic programming model. If the degree of violation of any constraint exceeds the first tolerance, the optimal scheduling decision sequence is determined to be invalid. The secondary verification is used to calculate the power balance residual, which is the absolute value of the difference between the algebraic sum of the output of all units and the charging and discharging power of energy storage and the load demand. If the power balance residual exceeds the second tolerance, the optimal scheduling decision sequence is determined to be invalid. The result verification unit performs the backtracking verification only after both the first-level verification and the second-level verification have passed.

6. The method according to claim 1, characterized in that, The solution engine solves the convex quadratic programming model to generate an optimal scheduling decision sequence, including: The solution engine employs the original-dual interior-point method and enables a numerical stability correction mechanism during the solution process. The numerical stability correction mechanism is used to adjust the iteration step size or introduce a regularization term when numerical ill-conditioning is detected, so as to ensure that the solution process converges to a feasible solution.

7. The method according to claim 1, characterized in that, Output the optimal scheduling decision sequence as a scheduling instruction, including: After outputting the optimal scheduling decision sequence, the load period division unit calculates the daily average load based on the actual load data of the previous day in the multiple data entries, and divides the load of each scheduling time of the day into low period, peak period and flat load period based on the daily average load. Based on the optimal scheduling decision sequence and the load period division results, a differentiated control strategy is generated. The differentiated control strategy includes: issuing charging commands to the energy storage system during the off-peak period; issuing discharging commands to the energy storage system during the peak period; and keeping the energy storage system idle during the flat load period.

8. A power grid dispatch backtracking verification device, characterized in that, include: The acquisition module is used to acquire multiple data entries from the data acquisition unit, each data entry carrying an acquisition timestamp; The first verification module is used to verify the multiple data entries by the unexpected constraint determination unit and generate a first verification result. The first verification result is used to indicate whether the collection timestamps of the multiple data entries are all not greater than the current scheduling time. The decision module is used to construct a convex quadratic programming model by the convex optimization modeling unit based on the multiple data entries when the first verification result is yes, and to solve the convex quadratic programming model by the solution engine to generate the optimal scheduling decision sequence. The second verification module is used to backtrack and verify the solution process of the solution engine by the result verification unit, and generate a second verification result. The second verification result is used to indicate whether the collection timestamp of all data entries called in the solution process is not greater than the current scheduling time. The instruction module is used to output the optimal scheduling decision sequence as a scheduling instruction when the second verification result is yes.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any one of the methods described in claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform any one of the methods described in claims 1 to 7.

Images