A power supply guarantee optimization method considering dynamic adaptation of backup capacity

Abstract

The application discloses a power supply guarantee optimization method considering dynamic adaptation of backup capacity, relates to the technical field of dynamic adaptation of backup capacity, and comprises the following steps: acquiring a backup capacity demand value, comparing the backup capacity demand value with a current actual backup capacity, and obtaining an initial deviation; adjusting the backup capacity configuration in real time according to the initial deviation, and executing the adjustment through a multi-time-scale hierarchical optimization scheduling mechanism; and based on a use deviation of an actual use value of the backup capacity from a predicted demand value, adopting model predictive control combined with time delay modeling to realize adaptive optimization of the backup capacity; through the construction of a multi-time-scale prediction model, real-time deviation feedback and an adaptive optimization scheduling method, the application realizes accurate prediction and dynamic adjustment of load and renewable energy output, thereby solving the problems of inaccurate prediction and response lag of backup capacity scheduling and the difficulty in balancing economy and reliability.

Description

Technical Field

[0001] This invention relates to the field of dynamic adaptation of reserve capacity technology, and more specifically, to a supply optimization method that takes into account dynamic adaptation of reserve capacity. Background Technology

[0002] With the increasing proportion of renewable energy in the power system, the uncertainty and volatility of power grid operation have significantly increased. Traditional reserve capacity dispatching methods typically rely on static planning or fixed reserve capacity reservation strategies, determining reserve capacity benchmarks through empirical values ​​or historical average load data. While these methods can guarantee a certain level of power supply reliability in long-term operation, they suffer from the following main problems: Existing methods often employ load forecasting or power generation forecasting on a single time scale, lacking comprehensive modeling of load and renewable energy output across multiple time scales (yearly, daily, hourly, and minute-by-minute). This makes it difficult to accurately reflect short-term rapid fluctuations and long-term trends. Furthermore, traditional reserve capacity dispatching lacks feedback and adjustment mechanisms for real-time system status. Even with forecasting errors, capacity correction typically relies on manual experience, failing to respond quickly to sudden changes in load or renewable energy output. This can lead to insufficient or excessive reserve capacity allocation, thereby affecting the accuracy and reliability of system forecasting. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a supply optimization method that takes into account dynamic adaptation of reserve capacity. Through real-time deviation feedback and adaptive optimization scheduling, it achieves accurate prediction and dynamic adjustment of load and renewable energy output, thereby solving the problems of inaccurate reserve capacity scheduling prediction, response lag, and difficulty in balancing economy and reliability.

[0004] To achieve the above objectives, the present invention provides the following technical solution: This application provides a supply guarantee optimization method that takes into account dynamic adaptation of reserve capacity. The method includes: obtaining the reserve capacity demand value, comparing the reserve capacity demand value with the current actual reserve capacity to obtain the initial deviation; adjusting the reserve capacity configuration in real time according to the initial deviation, and performing the adjustment through a multi-time-scale hierarchical optimization scheduling mechanism; and achieving adaptive optimization of reserve capacity by using model predictive control combined with time delay modeling based on the usage deviation between the actual usage value and the predicted demand value of the reserve capacity.

[0005] In one embodiment, the step of obtaining the reserve capacity demand value includes: obtaining annual-scale forecast results and mapping the annual-scale forecast results to daily-scale forecast data; calculating a preliminary reserve capacity baseline value based on the daily-scale forecast data; obtaining hourly-scale forecast data based on the hourly-scale load-output ratio allocation factor and the daily-scale forecast data, and correcting the preliminary reserve capacity baseline value according to the ratio allocation factor to obtain an hourly reserve capacity baseline value; using the hourly-scale forecast data and the hourly reserve capacity baseline value as boundary conditions, performing ultra-short-term correction using LSTM, and outputting minute-level reserve capacity demand values.

[0006] In one embodiment, LSTM is used for ultra-short-term correction to output minute-level reserve capacity demand values. This includes: acquiring and preprocessing minute-level real-time monitoring data to form a real-time feature vector; constructing a minute-level online prediction model based on the feature vector and a long short-term memory network; updating the weights of the prediction model online using new training samples through incremental learning; predicting the reserve capacity demand increment for the next few minutes based on the updated prediction model; and generating minute-level reserve capacity demand values ​​by combining the reserve capacity demand increment with the boundary conditions.

[0007] In one embodiment, the incremental learning method is mini-batch gradient descent.

[0008] In one embodiment, real-time adjustment of standby capacity configuration based on initial deviation includes: dynamically generating a deviation threshold based on adaptive control and statistical error; triggering real-time adjustment of standby capacity when the initial deviation exceeds the deviation threshold; the real-time adjustment of standby capacity includes: defining an adjustable resource strategy set and establishing a multi-objective optimization function to calculate the comprehensive score of each strategy; determining the strategy priority according to the comprehensive score and adjusting the standby capacity according to the executed strategy.

[0009] In one embodiment, the deviation threshold is dynamically generated based on adaptive control and statistical error. Specifically, the following steps are taken: the historical load prediction error sequence and the current system operating status are obtained; the prediction error standard deviation is calculated based on the historical load prediction error sequence; an adaptive deviation threshold is dynamically generated based on the prediction error standard deviation and the current system operating status; the above steps are executed in a rolling manner to form a rolling closed-loop control until the threshold converges, and the final minute-level reserve capacity deviation threshold is output.

[0010] In one embodiment, adjustments are performed through a multi-timescale hierarchical optimization scheduling mechanism. Specifically, the adjusted reserve capacity is aggregated into an hourly initial capacity, and the hourly reserve capacity demand value is obtained. The deviation is compared with the actual hourly reserve capacity, and an hourly rolling scheduling plan is generated through a multi-objective optimization function. The scheduling result of the hourly rolling scheduling plan is used as the input of the daily optimization model to solve for the daily reserve capacity scheduling scheme. The daily scheduling scheme is aggregated into an annual initial capacity, and a long-term reserve capacity allocation plan is generated through an annual optimization objective function.

[0011] In one embodiment, based on the deviation between the actual usage value and the predicted demand value of the reserve capacity, adaptive optimization of the reserve capacity is achieved by combining model predictive control with time-delay modeling. Specifically, this involves: establishing a dynamic model of the reserve capacity with feedback lag; predicting the capacity demand for future periods based on the dynamic model to obtain an initial capacity prediction value; optimizing the problem through model predictive control based on the initial capacity prediction value to obtain the optimal scheduling strategy; executing reserve capacity allocation based on the optimal scheduling strategy and collecting the actual usage deviation; feeding the actual usage deviation back to the dynamic model, correcting the dynamic model parameters and model predictive control parameters in real time, and updating the initial capacity prediction value; and performing rolling optimization based on the updated initial capacity prediction value to update the scheduling strategy for the next period.

[0012] In one embodiment, the actual usage deviation is fed back to the dynamic model to correct the dynamic model parameters and model predictive control parameters in real time, and to update the initial capacity prediction value. Specifically, based on the operating deviation information, Bayesian estimation is used to update the posterior distribution and confidence interval of the dynamic model parameters; when the system operation tends to be stable, the parameter correction method is switched to Kalman filter estimation; the weight parameters of the model predictive control are adjusted online based on the reinforcement learning algorithm; and the capacity prediction value for future periods is updated according to the corrected model parameters and weight parameters.

[0013] In one embodiment, the weight parameters of model predictive control are adjusted online based on a reinforcement learning algorithm. Specifically, the system operating deviation, its statistical characteristics, and the system operating state are combined into a state vector; an action space containing the adjustment of the model predictive control weight parameters is defined; a reward function containing the cost of reserve capacity usage and the amount of prediction deviation is constructed; a weight regulator is trained based on the state vector, action space, and reward function to learn the optimal weight adjustment strategy; at the beginning of each time window, the trained weight regulator is invoked to output the optimal weight combination based on the real-time state vector.

[0014] This invention generates an initial deviation by comparing the reserve capacity demand with the actual capacity in real time, and drives a multi-time-scale hierarchical optimization mechanism based on this deviation. This effectively solves the capacity mismatch problem caused by insufficient prediction accuracy and response delay in traditional scheduling, achieving rapid closed-loop regulation with minute-level accuracy. By introducing model predictive control and time-delay modeling, and considering the impact of communication delay and actuator response lag on the control effect, it overcomes the insufficient adaptability of traditional PID control and other methods in the high-inertia scenario of power systems, significantly reducing overshoot and settling time. Using the continuous deviation between the actual usage value and the predicted value as a feedback signal, and through a model predictive control rolling optimization strategy, the system can dynamically track changes in operating status and automatically correct control parameters, significantly improving the robustness and self-learning capability of reserve capacity management in complex operating environments. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a supply optimization method that takes into account dynamic adaptation of reserve capacity, provided for an embodiment of this application.

[0016] Figure 2 A line graph showing minute-level deviation thresholds provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] Reference Figure 1 As shown in the diagram, this invention provides a supply guarantee optimization method that takes into account dynamic adaptation of reserve capacity, comprising the following steps: S1 collects real-time power system operation data and environmental data, establishes a multi-time-scale prediction model, and outputs the reserve capacity demand value. The multi-time-scale includes annual, daily, hourly, and minute-level scales. The annual and daily scales are used to generate the baseline value of reserve capacity, and the hourly and minute-level scales are used to dynamically correct the baseline value.

[0019] The power system operation data includes load demand, power output, energy storage device power status, standby generator availability, and grid operation topology, while the environmental data includes temperature, wind speed, and solar irradiance.

[0020] In this embodiment, real-time power system operation data and environmental data are collected, and a multi-time-scale prediction model is established to output the reserve capacity demand value, specifically: The power system operation data and environmental data are preprocessed, and the processed data are used to establish an annual-scale load forecasting model and a power generation output forecasting model using a regression model. The annual load forecasting model outputs the trend value of the daily load curve throughout the year, while the power generation forecasting model outputs the annual power generation trend of renewable energy. Obtain annual-scale forecast results, which include annual maximum load and annual average renewable output; The annual-scale forecast results are decomposed into daily characteristic components, and daily-scale load trend factors and renewable output correction factors are established. The annual-scale forecast results are refined into daily-scale forecast data by mapping the daily-scale load trend factor and the renewable energy output correction factor. The daily-scale forecast data includes daily-scale base load forecast values ​​and daily-scale base renewable energy output forecast values. The specific calculation formula for the daily-scale basic load forecast is as follows:

[0021] The specific calculation formula for the daily-scale basic renewable energy output forecast is as follows:

[0022] In the formula, This represents the daily-scale basic load forecast. Forecast values ​​of basic renewable energy output on a daily scale. This represents the maximum annual load value. For the average annual renewable energy output, This is the daily-scale load trend factor, representing the relative proportion of daily-scale load throughout the year, obtained by normalizing the annual load curve. The renewable energy output correction factor represents the ratio of daily renewable energy output to annual average output, obtained from meteorological forecasts and historical power curve fitting.

[0023] Calculate the preliminary reserve capacity baseline value based on daily-scale forecast data; The specific calculation formula for the preliminary reserve capacity baseline value is as follows:

[0024] In the formula, This is a preliminary baseline value for reserve capacity. This is the annual-scale safety redundancy coefficient, used to reflect annual load fluctuations and forecast uncertainties.

[0025] Based on historical load time series characteristics and meteorological data, the hourly load and output ratio allocation factor is obtained; Hourly-scale forecast data is obtained by combining the proportional allocation factor with daily-scale forecast data. The hourly-scale forecast data includes hourly base load forecasts and hourly base renewable energy output forecasts. The specific calculation formula for the hourly basic load forecast is as follows:

[0026] The specific calculation formula for the hourly baseline renewable energy output forecast is as follows:

[0027] In the formula, This is the hourly base load forecast value. Forecast values ​​of hourly-based renewable energy output. This is the hourly-scale load allocation factor. The output ratio allocation factor on an hourly scale.

[0028] The hourly reserve capacity baseline value is obtained by correcting the initial reserve capacity baseline value with hourly meteorological data using the aforementioned proportional allocation factor. The specific calculation formula for the hourly reserve capacity baseline value is as follows:

[0029]

[0030]

[0031] In the formula, This is the baseline value for initial standby capacity per hour. The weight of the h-th hour in the reserve capacity allocation. This is the revised baseline value for hourly reserve capacity. The nonlinear response function between meteorology and load can be obtained using multiple regression methods. Temperature in the hourly range. At the speed of light in hours, Hourly wind speed, The load fluctuation coefficient represents the rate of change of load in that hour.

[0032] Using hourly-scale forecast data and hourly reserve capacity baseline values ​​as boundary conditions, combined with minute-level real-time monitoring data, LSTM online prediction is used for ultra-short-term correction, outputting minute-level reserve capacity demand values ​​to guide real-time reserve capacity dynamic adaptation.

[0033] It should be noted that by establishing a hierarchical forecasting system covering four time scales—year, day, hour, and minute—the method achieves full-cycle adaptive optimization of reserve capacity, from long-term planning to real-time dynamic adjustment. At the yearly and daily scales, this method comprehensively considers load growth trends and renewable energy output patterns to generate a reserve capacity benchmark value with global constraints. At the hourly and minute scales, it combines real-time meteorological and operational monitoring data, utilizing proportional allocation and LSTM dynamic correction mechanisms to refine and update the benchmark value, thereby achieving high-precision, dynamic, and closed-loop optimization of reserve capacity configuration. This scheme not only significantly improves the timeliness and accuracy of reserve capacity forecasting but also effectively balances the safety, economy, and renewable energy absorption capacity of the power system, providing an implementable and scalable intelligent power supply optimization method for achieving safe and stable operation of the power grid under conditions of high proportion of clean energy.

[0034] Furthermore, using hourly-scale forecast data and hourly reserve capacity baseline values ​​as boundary conditions, combined with minute-level real-time monitoring data, LSTM online forecasting is employed for ultra-short-term correction, outputting minute-level reserve capacity demand values, specifically: Using hourly-scale forecast data and hourly reserve capacity baseline values ​​as boundary conditions, we set upper and lower limits for forecasts and initial states. Acquire minute-level real-time monitoring data, including load changes, remaining energy storage capacity, actual output of renewable energy, and start / stop status of standby generators; Denoising and normalization are performed on minute-level real-time monitoring data to form real-time feature vectors; A minute-level online prediction model based on Long Short-Term Memory (LSTM) network is constructed. The LSTM input includes real-time feature vectors and hourly reserve capacity baseline values. The hidden layer state of the LSTM carries load trend information from the previous few minutes, thus realizing time-dependent modeling. The real-time feature vector, the hourly reserve capacity baseline value, and the corresponding LSTM prediction output are combined to form a new training sample. The model weights are then updated by mini-batch gradient descent on the new training sample using the current weights of the online prediction model. The specific calculation formula for the updated model weights is as follows:

[0035] In the formula, To update the model weights at the current time step, These are the model weights from the previous time step. For learning rate, The gradient of the loss function is given by the loss function. Regarding model weights The gradient is given, and the loss function is the mean squared error.

[0036] The incremental reserve capacity demand in the next few minutes is predicted based on the minute-level online prediction model after the model weights are updated. The hourly reserve capacity baseline value is adjusted minute-by-minute based on the incremental reserve capacity demand, and boundary conditions are met to generate minute-by-minute reserve capacity demand values.

[0037] The specific calculation formula for the minute-level reserve capacity requirement is as follows:

[0038] The boundary conditions are specifically as follows:

[0039]

[0040] In the formula, This represents the standby capacity requirement in minutes. To meet the increased demand for backup capacity, The allowable deviation range is set at the hour level to ensure that minute-level reserve capacity adjustments respond to real-time deviations without exceeding hourly boundaries, thus guaranteeing scheduling stability.

[0041] S2 compares the required reserve capacity with the current actual reserve capacity to obtain the initial deviation, adjusts the reserve capacity configuration in real time based on the initial deviation, and executes the adjustment through a multi-time-scale hierarchical optimization scheduling mechanism.

[0042] In this embodiment, the required reserve capacity is compared with the current actual reserve capacity to obtain an initial deviation. The reserve capacity configuration is then adjusted in real time based on the initial deviation. Specifically: Obtain the current actual reserve capacity and compare it with the minute-level reserve capacity demand value to obtain the initial deviation between the current actual reserve capacity and the demand value; The system dynamically generates minute-level reserve capacity deviation thresholds through adaptive control and statistical error drive. If the initial deviation exceeds the deviation threshold, the reserve capacity is adjusted in real time. The reserve capacity is adjusted in real time, specifically as follows: Define a set of adjustable resource strategies and establish a multi-objective optimization function, which includes scheduling cost, response time, equipment status impact and reliability risk; The specific calculation formula for the multi-objective optimization function is as follows:

[0043] In the formula, For a multi-objective optimization function, For scheduling costs, For response time, Due to equipment status, For reliability risks, , , , These are the weighting coefficients.

[0044] The adjustable resource strategy set includes energy storage device charging and discharging strategies, standby generator start-up and shutdown strategies, and flexible load adjustment strategies.

[0045] The comprehensive score of each strategy is calculated based on the multi-objective optimization function, and the strategies are prioritized according to their scores from low to high. The strategy adjustments are executed sequentially according to the order to obtain the updated reserve capacity. Updating the actual reserve capacity can be understood as executing various adjustment strategies according to the multi-objective optimization order based on the current reserve capacity, and accumulating the increment of the reserve capacity by all strategies to obtain the updated reserve capacity.

[0046] It should be noted that comparing the required reserve capacity with the actual current reserve capacity and adjusting the reserve capacity configuration in real time enables dynamic adaptive management of reserve capacity. On the one hand, this method can respond promptly to load fluctuations and uncertainties in renewable energy output, ensuring that the system has sufficient reserve capacity at all times, thereby improving power supply reliability. On the other hand, by introducing a multi-objective optimization function to prioritize and efficiently schedule adjustable resources such as energy storage, backup generators, and flexible loads, it can optimize economy and equipment operating status while meeting reliability requirements, avoiding ineffective operations and resource waste, thus achieving safe, economical, and flexible multi-timescale hierarchical reserve capacity scheduling.

[0047] Furthermore, such as Figure 2 As shown, the deviation threshold for minute-level reserve capacity is dynamically generated through adaptive control and statistical error drive, specifically as follows: The system acquires historical load forecasting error sequences and current system operating status. The load forecasting error sequences are obtained by the difference between minute-level load forecast values ​​and actual measured load values. The current system operating status includes energy storage SOC, standby generator availability, and adjustable load capacity. Calculate the standard deviation of the forecast error based on the load forecast error sequence; An adaptive deviation threshold is established based on the standard deviation of the prediction error and the current system state; The adaptive deviation threshold is calculated using the following formula:

[0048] In the formula, For adaptive deviation threshold, Based on the baseline deviation threshold, , For adjustment coefficients, The system state influence function considers the remaining energy storage capacity and minute-level renewable output. This represents the standard deviation of the prediction error.

[0049] The adaptive deviation threshold establishment process is repeated every minute to form a rolling closed-loop control until a preset number of times is reached, at which point convergence is achieved, generating a minute-level deviation threshold for reserve capacity. As historical prediction errors and system state changes, the deviation threshold is adaptively updated in real time, making reserve capacity adjustment both fast and economical.

[0050] It should be noted that the deviation threshold can be dynamically adjusted based on historical load prediction errors and the current system status, so that reserve capacity adjustment is only triggered when the actual deviation exceeds the threshold, thereby avoiding frequent and ineffective operations and improving the sensitivity and economy of system response. At the same time, when load or renewable output fluctuations increase, the threshold automatically increases to tolerate short-term fluctuations and ensure dispatch stability; when fluctuations decrease, the threshold decreases to improve dispatch accuracy and realize minute-level dynamic adaptive closed-loop reserve capacity management, which not only ensures power supply reliability but also optimizes equipment operation and dispatch costs.

[0051] Furthermore, adjustments are implemented through a multi-time-scale hierarchical optimization scheduling mechanism, specifically as follows: The updated reserve capacity is aggregated into hourly initial capacity, and combined with hourly load forecasts and renewable output, the hourly reserve capacity demand is obtained. The specific calculation formula for the hourly reserve capacity requirement is as follows:

[0052] In the formula, This represents the hourly standby capacity requirement. For safety reserves.

[0053] Based on the deviation between the actual hourly reserve capacity and the demand value, a multi-objective optimization function is used to adjust the adjustable resource strategy set to form an hourly rolling scheduling plan. The scheduling results of the hourly rolling scheduling plan are used as the initial capacity input for the daily level. A daily optimization model is established by taking into account daily load forecast, renewable energy output, energy storage charge and discharge cycle count and equipment operating costs. The specific calculation formula for the daily-level optimization model is as follows:

[0054] In the formula, For daily optimization models, The cost of generating electricity in the h-th hour. The energy storage cost for the h-th hour is... This is a penalty value for equipment lifespan consumption. This is the lifespan penalty weighting coefficient.

[0055] The daily backup capacity scheduling plan is output through the daily optimization model, so that while meeting the reliability requirements, it also takes into account the economy and the extension of energy storage life. The daily scheduling results are aggregated into an initial capacity plan for the year. Combined with annual load forecasts and renewable energy output, an optimization objective function for the year is established. The optimization objective function for the year includes minimizing annual operating costs, extending the service life of standby equipment, and ensuring that the system reliability indicators meet the long-term supply requirements. Solve the grade-level optimization objective function to form a long-term reserve capacity allocation and operation plan, providing a reference boundary for daily and hourly scheduling.

[0056] It should be noted that by constructing a hierarchical optimization scheduling mechanism with four time scales—minute, hour, day, and year—dynamic, adaptive, and closed-loop management of reserve capacity is achieved. At the minute and hourly levels, it can quickly respond to load fluctuations and changes in renewable energy output, ensuring short-term security and immediate supply. At the daily level, it combines the number of energy storage charge-discharge cycles and equipment operating costs to achieve a balance between economic efficiency and equipment lifespan. At the yearly level, it further coordinates long-term operating costs, equipment lifespan, and system reliability, providing boundary conditions and reference schemes for medium- and long-term scheduling. Overall, this scheme balances real-time response capability, economic efficiency, equipment lifespan management, and long-term reliability, making power system reserve capacity scheduling both safe and reliable, as well as efficient and economical, adaptable to complex and ever-changing grid operating environments.

[0057] S3 is based on the usage deviation between the actual usage value and the predicted demand value of the reserve capacity, and achieves adaptive optimization of the reserve capacity through model predictive control and time delay modeling.

[0058] In this embodiment, based on the deviation between the actual usage and the predicted demand of the reserve capacity, adaptive optimization of the reserve capacity is achieved through model predictive control and time-delay modeling, specifically as follows: Based on historical load data, updated reserve capacity operation records and system structure, a dynamic model of reserve capacity with feedback lag is established. The model can be a discrete time-delay difference equation. Based on the aforementioned dynamic model of reserve capacity and historical data, the capacity demand at each future time step is predicted using machine learning methods to obtain the initial capacity prediction value. In the predicted future time window Within this framework, an MPC optimization problem is constructed using the initial capacity prediction and system constraints as input. The system constraints include upper and lower capacity limits, system safety margin constraints, and switching rate constraints. The specific calculation formula for the MPC optimization problem is as follows:

[0059] In the formula, Use a cost function for reserve capacity. The backup capacity scheduling strategy at time step t, To predict the deviation.

[0060] The optimal scheduling strategy is obtained by solving the MPC optimization problem using the quadratic programming method. Based on the optimal scheduling strategy, reserve capacity allocation is performed, and reserve capacity usage and load data are collected to obtain the actual usage deviation between the actual usage and the initial capacity prediction value. Using actual usage deviations as feedback input, combined with a dynamic model, the parameters of the dynamic model and the weight parameters of MPC are corrected, and the capacity prediction values ​​for each future time step are updated. After each time window ends, MPC rolling optimization is performed again based on the updated capacity prediction and system status, and the standby capacity scheduling strategy for the next time window is updated. Repeated deviation information acquisition and rolling optimization steps form a real-time feedback and rolling optimization closed loop, enabling the standby capacity configuration to dynamically respond to load fluctuations and time lag characteristics, thereby improving system reliability and resource utilization efficiency.

[0061] It should be noted that by using the deviation between the actual usage of reserve capacity and the predicted demand as feedback, combined with model predictive control (MPC) and dynamic modeling with time delays, adaptive optimization of reserve capacity is achieved. The advantages of this method are: it can identify and correct the deviation between capacity configuration and actual load demand in real time, enabling reserve capacity to dynamically respond to load fluctuations and system hysteresis characteristics; simultaneously, through rolling optimization and prediction error correction, it improves system reliability and safety margin, reduces reserve capacity waste, achieves efficient resource utilization, and balances economy and security, thereby optimizing operating costs while ensuring power supply stability.

[0062] Furthermore, the actual usage deviation is used as feedback input, and combined with the dynamic model, the parameters of the dynamic model and the MPC weight parameters are corrected to update the capacity prediction values ​​for each future time step, specifically: By using the bias information as observation information and updating the parameter distribution using Bayes' theorem, the posterior distribution and confidence interval of the dynamic model parameters are obtained. In the initial stage of model operation, the dynamic model parameters are updated according to the confidence interval, so that the model can maintain adaptability to actual deviations within a wide parameter range, thereby avoiding overfitting and too fast convergence. If the mean and variance of the deviation information are both lower than the preset threshold within a certain time window during continuous system operation, the system operation is determined to be stable. As the system operation tends to stabilize, the dynamic model parameter correction method is switched from Bayesian estimation to Kalman filter estimation. Based on the parameter update values ​​from the previous moment and the current observation information, Kalman filtering is used to perform high-precision parameter correction on the dynamic model parameters; A weight regulator is trained based on a reinforcement learning algorithm to update the MPC weight parameters; Based on the corrected model parameters and the updated MPC weight parameters, the capacity prediction values ​​for each future time step are obtained again.

[0063] It should be noted that by using Bayes' theorem to update the parameter distribution and provide a confidence interval in the early stages of the model, the uncertainty modeling and relaxed correction of the dynamic model parameters are achieved, which can effectively avoid the problem of overfitting or too fast convergence of parameters in the stage of insufficient data. After the system operation tends to stabilize, switching to Kalman filtering for high-precision recursive correction can gradually approach the true parameter values ​​within a narrower interval, thereby improving prediction accuracy and convergence speed while maintaining the model's adaptability, and achieving a dynamic optimization effect that balances robustness and accuracy.

[0064] Furthermore, a weight regulator is trained based on a reinforcement learning algorithm to update the MPC weight parameters, specifically as follows: The deviation information and its statistical characteristics are combined with the system operating state to form a state vector, which is used as the input state of the reinforcement learning model. Define the action space as the set of adjustments for MPC weight parameters; Construct a reward function, which includes the cost of using reserve capacity and the prediction deviation. The higher the reward value, the better the system economy and accuracy under the weight configuration. Based on the state vector, action set and reward function, a weight regulator is trained using a deep Q-network method so that it can learn the optimal weight adjustment strategy from historical operating experience. At the beginning of each time window, the weight adjuster is invoked to output the optimal weight combination based on the latest state and update the MPC optimization problem.

[0065] It should be noted that the advantage of training the weight regulator through reinforcement learning algorithms to adaptively update the MPC weight parameters is that it can take deviation information and system operating status as input, and continuously learn and optimize the weight adjustment strategy by combining historical operating experience, thereby avoiding the subjectivity and rigidity of manually setting weights. At the same time, by incorporating economic efficiency and prediction accuracy into the optimization objective through the reward function, the weight allocation can be dynamically adjusted according to the system status, realizing long-term optimization and real-time adaptation of reserve capacity configuration, and improving the flexibility, robustness and global performance of MPC optimization.

[0066] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0067] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0068] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0069] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0070] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0071] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A supply guarantee optimization method that takes into account dynamic adaptation of reserve capacity, characterized in that, include: Obtain annual-scale prediction results and map them to daily-scale prediction data; Based on daily-scale forecast data, a preliminary baseline value for reserve capacity is calculated. Based on the hourly load and output ratio allocation factor, hourly forecast data is obtained by combining daily forecast data, and the initial reserve capacity baseline value is corrected according to the ratio allocation factor to obtain the hourly reserve capacity baseline value. Using hourly forecast data and hourly reserve capacity baseline values ​​as boundary conditions, LSTM is used for ultra-short-term correction to output minute-level reserve capacity demand values. The initial deviation is obtained by comparing the required reserve capacity with the current actual reserve capacity. Based on adaptive control and statistical error, a deviation threshold is dynamically generated. When the initial deviation exceeds the deviation threshold, the reserve capacity is adjusted in real time. The real-time adjustment of the reserve capacity includes: defining a set of adjustable resource strategies, establishing a multi-objective optimization function, calculating the comprehensive score of each strategy; determining the strategy priority according to the comprehensive score, and adjusting the reserve capacity according to the executed strategy. The method of dynamically generating the deviation threshold based on adaptive control and statistical error is as follows: Obtain historical load forecasting error sequences and current system operating status; Calculate the standard deviation of the forecast error based on the historical load forecast error sequence; Based on the standard deviation of the prediction error and the current system operating status, an adaptive deviation threshold is dynamically generated. A rolling closed-loop control is formed by performing a rolling deviation threshold generation step until the threshold converges, and the final minute-level deviation threshold of the standby capacity is output. Adjustments are performed through a multi-time-scale hierarchical optimization scheduling mechanism; Based on the deviation between the actual usage and the predicted demand of reserve capacity, adaptive optimization of reserve capacity is achieved by combining model predictive control with time delay modeling.

2. The supply guarantee optimization method considering dynamic adaptation of reserve capacity according to claim 1, characterized in that, The method of using LSTM for ultra-short-term correction to output minute-level reserve capacity demand values ​​includes: Acquire and preprocess minute-level real-time monitoring data to form real-time feature vectors; Based on feature vectors, a minute-level online prediction model based on a long short-term memory network is constructed. The weights of the prediction model are updated online using incremental learning with new training samples. The updated forecasting model predicts the incremental reserve capacity demand over the next few minutes. By combining the incremental reserve capacity demand with the aforementioned boundary conditions, a minute-level reserve capacity demand value is generated.

3. The supply optimization method considering dynamic adaptation of reserve capacity according to claim 2, characterized in that, The incremental learning method is the mini-batch gradient descent method.

4. The supply optimization method considering dynamic adaptation of reserve capacity according to claim 1, characterized in that, The adjustment is performed through a multi-time-scale hierarchical optimization scheduling mechanism, specifically as follows: The adjusted reserve capacity is aggregated into an hourly initial capacity, and the hourly reserve capacity demand value is obtained. The deviation is compared with the actual hourly reserve capacity, and an hourly rolling scheduling plan is generated through a multi-objective optimization function. The scheduling results of the hourly rolling scheduling plan are used as the input of the daily optimization model to obtain the daily reserve capacity scheduling scheme. The daily scheduling schemes are aggregated into grade-level initial capacities, and a long-term reserve capacity allocation plan is generated through the grade-level optimization objective function.

5. The supply optimization method considering dynamic adaptation of reserve capacity according to claim 1, characterized in that, The deviation between the actual usage and predicted demand of reserve capacity is used to achieve adaptive optimization of reserve capacity through model predictive control combined with time-delay modeling. Specifically: Establish a dynamic model of reserve capacity with feedback lag; Based on the dynamic model of reserve capacity, the capacity demand for future periods is predicted to obtain the initial capacity forecast value; The optimal scheduling strategy is obtained by using model prediction to optimize the control problem based on the initial capacity prediction value. Reserve capacity allocation is performed based on the optimal scheduling strategy, and actual usage deviations are collected; The actual usage deviation is fed back to the dynamic model, the dynamic model parameters and model prediction control parameters are corrected in real time, and the initial capacity prediction value is updated. Rolling optimization is performed based on the updated initial capacity predictions, and the scheduling strategy for the next time period is updated.

6. The supply optimization method considering dynamic adaptation of reserve capacity according to claim 5, characterized in that, The process of feeding back actual usage deviations to the dynamic model, correcting the dynamic model parameters and model prediction control parameters in real time, and updating the initial capacity prediction value specifically involves: Based on the operational deviation information, Bayesian estimation is used to update the posterior distribution and confidence interval of the dynamic model parameters; When the system operation tends to be stable, switch the parameter correction method to Kalman filter estimation; Online adjustment of weight parameters for model predictive control based on reinforcement learning algorithms; The capacity forecast for future periods is updated based on the revised model parameters and weight parameters.

7. The supply optimization method considering dynamic adaptation of reserve capacity according to claim 6, characterized in that, The online adjustment of the weight parameters for model prediction control based on the reinforcement learning algorithm specifically involves: The system operating deviation, its statistical characteristics, and the system operating state are combined into a state vector; Define the action space that includes the adjustment of the model prediction control weight parameters; Construct a reward function that includes the cost of using reserve capacity and the amount of prediction deviation; The weight regulator is trained based on the state vector, action space, and reward function to learn the optimal weight adjustment strategy. At the beginning of each time window, the trained weight adjuster is invoked to output the optimal weight combination based on the real-time state vector.

Images