Thermal power storage combined peak shaving control method and system considering storage life

Abstract

The application relates to the field of electric energy storage control, and specifically discloses a thermal power storage combined peak shaving control method and system considering storage life, which firstly utilizes extended Kalman filtering to perform online estimation on battery SoH, and quantifies the uncertainty of the estimated value through a covariance matrix. Subsequently, the uncertainty information (i.e. the covariance matrix) is introduced into an optimization objective function of model predictive control to form a three-target optimization problem including economic cost, degradation cost and information cost. In this way, when formulating the storage charging and discharging and thermal power unit output strategy, the controller not only balances the current economic benefit and battery life loss, but also actively avoids and manages the uncertainty of future state estimation, and even takes a probing action to improve information accuracy.

Description

Technical Field

[0001] This application relates to the field of energy storage control, and more specifically, to a method and system for joint peak-shaving control of thermal power and energy storage considering energy storage life. Background Technology

[0002] With the advancement of dual-carbon goals and the construction of new power systems, the increasing penetration rate of new energy sources has made the demand for flexible regulation resources in the power grid increasingly urgent. Thermal power units, as the traditional main force of regulation, have slow ramp-up rates and large regulation delays, making it difficult to cope with the randomness and volatility of new energy output alone. Electrochemical energy storage, with its advantages of fast response speed and high control precision, has become an ideal partner for assisting thermal power in frequency regulation and peak shaving. The core purpose of constructing a joint thermal power and energy storage peak shaving control scheme that considers energy storage lifespan is to fully leverage the complementary characteristics of both while addressing the high lifespan cost of energy storage. During joint peak shaving, unrestricted use of energy storage for high-frequency, deep charging and discharging will lead to a sharp decline in battery health, thereby increasing the operating cost over the entire lifespan. Therefore, how to delay energy storage degradation through refined control strategies, achieving a dynamic balance between overall system economic benefits and energy storage lifespan, while meeting the grid's peak shaving needs and ensuring the economic efficiency of thermal power operation, is a key issue that urgently needs to be addressed in the field of joint thermal power and energy storage control.

[0003] However, existing combined peak-shaving control schemes for thermal power and energy storage still have significant limitations in practical applications. Most current control strategies employ a cascaded open-loop information flow architecture, where state estimation and control decision-making are artificially separated into two independent stages. Specifically, existing battery management systems (BMS) or upper-level energy management systems typically estimate the battery's SoH (Solar Hysteresis) based on ampere-hour integrals or simple empirical models, and pass this estimate as a deterministic, error-free constant to downstream control modules (such as Model Predictive Control, MPC). The drawback of this approach is that it ignores the unobservability of the battery's internal state (such as SoH) and the inherent uncertainties in the estimation process. In actual operating conditions, the estimated SoH often deviates due to the heterogeneity of individual battery cells, temperature variations, and sensor noise. When the controller blindly trusts this erroneous state value, it's like navigating under incorrect map guidance, easily leading to aggressive charging and discharging strategies (such as high-rate charging and discharging even when the battery is actually aged), resulting in actual battery life loss far exceeding expectations and even posing safety risks. Furthermore, existing optimization objective functions typically only add fuel costs and a fixed degradation penalty term, failing to perceive the impact of control behavior on the accuracy of future state estimation. This leads to the accumulation and amplification of uncertainty over long-term operation, ultimately causing theoretically optimal control strategies to become suboptimal or even ineffective decisions in the field.

[0004] Therefore, an optimized joint peak-shaving control scheme for thermal power and energy storage is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method and system for combined peak-shaving control of thermal power and energy storage, taking into account energy storage lifetime.

[0006] According to one aspect of this application, a method for joint peak-shaving control of thermal power and energy storage considering energy storage lifetime is provided, comprising:

[0007] Forecast data is constructed based on load forecast sequences and electricity price forecast sequences, and observation data is constructed based on measured battery voltage, measured battery temperature and measured battery current.

[0008] Based on the observation data and the control input of the previous time step, online SoH estimation and uncertainty quantification based on EKF are performed on the posterior state estimate and the posterior covariance matrix of the previous time step to obtain the posterior state estimate and the posterior covariance matrix of the current time step.

[0009] Based on the predicted data and the posterior state estimate at the current moment, construct the economic cost function and the degradation cost function;

[0010] Based on the posterior covariance matrix at the current time, a bi-objective MPC optimization is performed on the economic cost function and the degradation cost function to obtain the optimal power sequence;

[0011] Based on the optimal power sequence, the energy storage command power and the thermal power command power are generated.

[0012] According to another aspect of this application, a combined thermal power plant and energy storage peak-shaving control system considering energy storage lifetime is provided, comprising:

[0013] The forecast and observation data construction module is used to construct forecast data based on load forecast sequence and electricity price forecast sequence, and to construct observation data based on measured battery voltage, measured battery temperature and measured battery current.

[0014] The posterior estimation module is used to perform online SoH estimation and uncertainty quantification based on EKF on the posterior state estimate and the posterior covariance matrix of the previous time step, based on the observation data and the control input of the previous time step, to obtain the posterior state estimate and the posterior covariance matrix of the current time step.

[0015] The cost function construction module is used to construct the economic cost function and the degradation cost function based on the predicted data and the posterior state estimate at the current time.

[0016] The dual-objective MPC optimization module is used to perform dual-objective MPC optimization on the economic cost function and the degradation cost function based on the posterior covariance matrix at the current time to obtain the optimal power sequence;

[0017] The command power generation module is used to generate energy storage command power and thermal power command power based on the optimal power sequence.

[0018] Compared with existing technologies, this application provides a joint peak-shaving control method and system for thermal power and energy storage that considers energy storage lifespan. First, it uses an extended Kalman filter to estimate the battery's SoH (Solar Energy Flow) online and quantifies the uncertainty of this estimate using a covariance matrix. Then, this uncertainty information (i.e., the covariance matrix) is introduced into the objective function of model predictive control, forming a three-objective optimization problem including economic cost, degradation cost, and information cost. The information cost is specifically used to penalize control strategies that may lead to a decrease in the accuracy of future SoH estimates or an increase in uncertainty. In this way, when formulating energy storage charging / discharging and thermal power unit output strategies, the controller not only weighs current economic benefits against battery lifespan depletion but also actively avoids and manages the uncertainty of future state estimates, and even takes probing actions to improve information accuracy. This fundamentally solves the problem in traditional schemes where the controller blindly trusts erroneous state values, leading to aggressive operation and uncertainty accumulation. It achieves a dynamic balance between system economy, energy storage lifespan, and information acquisition capability, ensuring the robustness and optimality of the joint peak-shaving strategy throughout its entire lifespan. Attached Figure Description

[0019] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 This is a flowchart of a combined peak-shaving control method for thermal power and energy storage considering energy storage life according to an embodiment of this application;

[0021] Figure 2 This is a schematic diagram of data flow in the combined peak-shaving control method for thermal power and energy storage, taking into account energy storage life, according to an embodiment of this application.

[0022] Figure 3 This is a flowchart illustrating the process of online SoH estimation and uncertainty quantification of the posterior state estimate and posterior covariance matrix at the previous time based on EKF, according to the thermal power and energy storage joint peak-shaving control method considering energy storage lifetime in accordance with the embodiments of this application.

[0023] Figure 4This is a flowchart illustrating the generation of energy storage command power and thermal power command power based on the optimal power sequence in the thermal power and energy storage joint peak-shaving control method considering energy storage lifetime according to the embodiments of this application.

[0024] Figure 5 This is a block diagram of a combined thermal power and energy storage peak-shaving control system considering energy storage life according to an embodiment of this application. Detailed Implementation

[0025] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0026] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0027] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0028] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0029] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0030] To address the problem that existing thermal power-storage joint peak-shaving schemes suffer from the disconnect between state estimation and control decision-making, leading to the controller ignoring the estimation error and uncertainty of SoH (Solar Health), and consequently causing accelerated battery degradation and control strategy failure due to aggressive charging and discharging, this proposal suggests a thermal power-storage joint peak-shaving control method that considers the lifespan of the energy storage. Specifically, firstly, an input data stream incorporating load price prediction and real-time battery status is constructed. An extended Kalman filter algorithm is then used to recursively calculate the posterior state of the previous time step, resolving the current battery health state and its corresponding posterior covariance matrix in real time, thereby accurately quantifying the uncertainty of state estimation. Next, based on this uncertainty quantification, a composite objective function is constructed, including economic operating costs, battery degradation costs, and information costs representing future state uncertainties (such as the trace of the predicted covariance matrix). This multi-dimensional objective is then continuously optimized within a model predictive control framework to generate an optimal power command sequence that balances economic benefits, lifespan extension, and information uncertainty suppression. Finally, this sequence is used to generate and issue coordinated control commands for energy storage and thermal power.

[0031] In the technical solution of this application, a joint peak-shaving control method for thermal power and energy storage considering energy storage life is proposed. Figure 1 This is a flowchart of a combined peak-shaving control method for thermal power and energy storage that takes into account energy storage life, according to an embodiment of this application. Figure 2 This is a schematic diagram of data flow in a combined thermal power and energy storage peak-shaving control method considering energy storage lifetime, according to an embodiment of this application. Figure 1 and Figure 2 As shown, the combined peak-shaving control method for thermal power and energy storage considering energy storage lifetime according to an embodiment of this application includes the following steps: S100, constructing forecast data based on load forecast sequence and electricity price forecast sequence, and constructing observation data based on measured battery voltage, measured battery temperature and measured battery current; S200, based on the observation data and the control input at the previous time moment, performing online SoH estimation and uncertainty quantification based on EKF on the posterior state estimate and the posterior covariance matrix at the previous time moment to obtain the posterior state estimate and the posterior covariance matrix at the current time moment; S300, constructing an economic cost function and a degradation cost function based on the forecast data and the posterior state estimate at the current time moment; S400, performing bi-objective MPC optimization on the economic cost function and the degradation cost function based on the posterior covariance matrix at the current time moment to obtain the optimal power sequence; S500, generating the energy storage command power and the thermal power command power based on the optimal power sequence.

[0032] Specifically, in step S100, forecast data is constructed based on the load forecast sequence and the electricity price forecast sequence, and observation data is constructed based on the measured battery voltage, measured battery temperature, and measured battery current. It should be understood that since the optimal control of a combined thermal power and energy storage system relies on data input at different time scales, model predictive control requires boundary conditions in the future time domain to plan long-term economic dispatch, while extended Kalman filtering requires real-time physical feedback to correct the current internal state estimate. If unprocessed raw multi-source heterogeneous data is used directly, problems such as sensor noise, data loss, and timing misalignment will lead to state estimation divergence and control strategy failure. Therefore, in the technical solution of this application, forecast data is further constructed based on the load forecast sequence and the electricity price forecast sequence, and observation data is constructed based on the measured battery voltage, measured battery temperature, and measured battery current to establish a standardized data preprocessing and alignment mechanism. This ensures that the data input to subsequent algorithm modules has temporal consistency and numerical validity, providing a reliable data foundation for achieving high-precision state perception and robust optimization decision-making.

[0033] More specifically, in a concrete example of this application, the system executes a serialized processing flow including real-time measurement acquisition, prediction sequence acquisition, and data cleaning and encapsulation. First, at the current scheduling moment, real-time physical operating parameters of the battery energy storage system are synchronously acquired via an industrial fieldbus or data acquisition and monitoring control system. These parameters include battery terminal voltage measurements, charging / discharging current measurements, and battery temperature measurements. The actual control commands from the previous scheduling moment are also extracted, and these instantaneous physical quantities are combined into a raw measurement dataset by attaching a unified timestamp. Simultaneously, the load forecasting algorithm interface and the electricity market pricing interface are called in parallel to acquire the raw load forecasting sequence and the raw electricity price forecasting sequence, respectively covering the future preset scheduling time domain. For these two sets of forecast data, time resolution verification and resampling operations are performed to ensure that their time steps strictly match the system's control cycle, and data segments of corresponding length starting from the current moment are extracted to form an aligned prediction dataset. Subsequently, the raw measurement dataset is validated, outliers or null values ​​are removed through range constraint checks, and the cleaned data is structurally encapsulated to generate observation data. Similarly, the processed prediction sequence is encapsulated into prediction data. This process transforms the raw physical signals into a standardized information flow, ensuring that the observed data accurately reflects the current electrochemical state of the battery and that the predicted data accurately describes the future boundary conditions of the system.

[0034] Specifically, in step S200, based on the observed data and the control input from the previous moment, online SoH estimation and uncertainty quantification based on EKF are performed on the posterior state estimate and the posterior covariance matrix from the previous moment to obtain the posterior state estimate and the posterior covariance matrix from the current moment. It should be understood that since the battery's state of health (SoH) is an internal state variable characterizing the degree of battery aging, it cannot be directly measured physically by sensors. Furthermore, batteries exhibit highly nonlinear and time-varying characteristics under complex peak-shaving conditions, and simple time-integration or table lookup methods are insufficient to overcome the effects of accumulated errors and measurement noise. Therefore, in the technical solution of this application, online SoH estimation and uncertainty quantification based on EKF are further performed on the posterior state estimate and the posterior covariance matrix from the previous moment based on the observed data and the control input from the previous moment to obtain the posterior state estimate and the posterior covariance matrix from the current moment. This integrates the prior knowledge of the physical model with real-time observed data to achieve dynamic correction and optimal estimation of the battery's internal state. In this way, while obtaining a high-precision SoH estimate, the confidence level of the estimate can be clearly quantified through the covariance matrix, providing a complete state description containing uncertainty information for subsequent model predictive control, and avoiding aggressive or ineffective control strategies due to state estimation bias.

[0035] Figure 3 This document describes a flowchart illustrating the process of online SoH estimation and uncertainty quantification of the posterior state estimate and posterior covariance matrix at the previous time step, based on observation data and the control input at the previous time step, according to an embodiment of this application. The method considers energy storage lifetime and is used to obtain the posterior state estimate and posterior covariance matrix at the current time step. Figure 3 As shown, step S200 includes: S210, based on the battery state equation and process noise covariance matrix, performing state and covariance prediction on the control input at the previous time step, the posterior state estimate at the previous time step, and the posterior covariance matrix at the previous time step to obtain the prior state estimate at the current time step and the prior covariance matrix at the current time step; S220, based on the observation data, battery observation equation, and measurement noise covariance matrix, performing observation linearization and Kalman gain calculation on the prior state estimate at the current time step and the prior covariance matrix at the current time step to obtain the Kalman gain matrix, the predicted observation value, and the observation Jacobian matrix; S230, based on the Kalman gain matrix, the predicted observation value, the observation Jacobian matrix, and the observation data, performing state and covariance update on the prior state estimate at the current time step and the prior covariance matrix at the current time step to obtain the posterior state estimate at the current time step and the posterior covariance matrix at the current time step.

[0036] Accordingly, in step S210, based on the battery state equation and the process noise covariance matrix, state and covariance prediction is performed on the control input at the previous time step, the posterior state estimate at the previous time step, and the posterior covariance matrix at the previous time step to obtain the prior state estimate and the prior covariance matrix at the current time step. It should be understood that since the electrochemical states inside the battery (such as state of charge (SoC) and state of health (SoH)) dynamically evolve over time and with current excitation, relying solely on historical states cannot accurately describe the system behavior at the current time step, and relying solely on sensor measurements is susceptible to noise interference and bias. Therefore, in the technical solution of this application, state and covariance prediction is further performed on the control input at the previous time step, the posterior state estimate at the previous time step, and the posterior covariance matrix at the previous time step, based on the battery state equation and the process noise covariance matrix, to obtain the prior state estimate and the prior covariance matrix at the current time step. This utilizes the prior knowledge of the physical model to prospectively extrapolate the evolution trend of the system state and simultaneously quantifies the degree of uncertainty diffusion in the model prediction process. This provides high-quality prior information, including the predicted mean and predicted variance, for subsequent measurement update steps, ensuring that the Kalman filter algorithm has a reliable weighting benchmark when fusing model predictions and real-time observations, thereby improving the dynamic response speed and robustness of state estimation.

[0037] Specifically, in a concrete example of this application, the system first performs a state prediction operation, retrieving the posterior state estimate of the previous time step (including the optimal estimates of SoC and SoH at the previous time step) and the control input of the previous time step (i.e., the measured current applied to the battery at the previous time step) from the historical database. These two sets of data are then substituted into a pre-constructed nonlinear battery state equation, and the prior state estimate of the current time step is obtained through numerical calculation. This process utilizes an electrochemical mechanism model to describe the changes in battery state with charge and discharge behavior. The calculation formula is as follows:

[0038] ,

[0039] in, This represents the prior state estimate at the current moment. This represents the posterior state estimate from the previous time step. This indicates the control input from the previous moment. ( The equation represents the nonlinear battery state equation. This step enables a one-step prediction of the physical evolution of the system's internal state at the current moment, based on the system state and control input from the previous moment.

[0040] Next, a covariance prediction operation is performed. At the point where the posterior state estimate was obtained in the previous time step, the nonlinear battery state equation is linearized, and the Jacobian matrix of the state transition matrix is ​​calculated. Then, combining the posterior covariance matrix from the previous time step with the preset process noise covariance matrix, the prior covariance matrix for the current time step is calculated. This matrix represents the range of uncertainty in the state estimate resulting solely from model prediction without incorporating current observation data. Its calculation formula is as follows:

[0041] ,

[0042] in, Represents the prior covariance matrix at the current time. The Jacobian matrix represents the state transition matrix. Let represent the posterior covariance matrix of the previous time step. This represents the process noise covariance matrix. This step quantifies the extent to which the uncertainty in the system state estimate spreads over time due to model approximation errors and process noise.

[0043] Accordingly, in step S220, based on the observation data, battery observation equation, and measurement noise covariance matrix, observation linearization and Kalman gain calculation are performed on the prior state estimate and the prior covariance matrix at the current time to obtain the Kalman gain matrix, predicted observation value, and observation Jacobian matrix. It should be understood that due to the strong nonlinear coupling between the battery's terminal voltage and its internal state of charge and health state, it is difficult to handle directly using linear filtering theory. Furthermore, both the prior state predicted by the model and the measured data acquired by the sensor have varying degrees of random error in accuracy, and relying solely on either one cannot obtain the optimal estimate. Therefore, in the technical solution of this application, observation linearization and Kalman gain calculation are further performed on the prior state estimate and the prior covariance matrix at the current time based on the observation data, battery observation equation, and measurement noise covariance matrix to obtain the Kalman gain matrix, predicted observation value, and observation Jacobian matrix. This allows for a local linearization approximation of the nonlinear observation model at the current estimation operating point, and the optimal correction weight is calculated based on the ratio of the uncertainty of the prior estimate to the uncertainty of the measurement noise. This ensures the reliability of the dynamic equilibrium model's predicted values ​​and actual observations under the minimum variance criterion during the subsequent state correction process, thereby improving the accuracy and convergence of state estimation.

[0044] Specifically, in a concrete example of this application, the system first performs an observation matrix calculation operation. It extracts the measured battery current at the current moment from the observation data and, at the state space point determined by the prior state estimate at the current moment, performs a Taylor series expansion of the nonlinear battery observation equation and truncates the first-order terms. The partial derivatives of this equation with respect to the state variables are then calculated to construct the observation Jacobian matrix. This matrix describes the sensitivity of the linearization effect of small changes in the battery's internal state on the terminal voltage. Next, a predicted observation calculation operation is performed. The values ​​of the state of charge and health from the prior state estimate at the current moment, along with the measured battery current, are directly substituted into the battery observation equation to calculate the predicted observation value. This value represents the battery terminal voltage value expected by the system based solely on the physical model without incorporating corrections for the current voltage measurement. Subsequently, the system performs a Kalman gain calculation operation. It retrieves the prior covariance matrix and the measurement noise covariance matrix at the current moment, combines them with the calculated observation Jacobian matrix, and solves for the Kalman gain matrix. This matrix, as a weighting coefficient, determines the degree of confidence the system places in the measurement residuals in subsequent update steps. Its calculation formula is as follows:

[0045] ,

[0046] in, Represents the Kalman gain matrix. Represents the prior covariance matrix at the current time. Represents the observed Jacobian matrix. This represents the measurement noise covariance matrix. This step calculates the optimal gain matrix in the sense of minimum mean square error through operations on the covariance matrix, when the uncertainty of the model prediction... When it is large, the gain If the value is increased, the system will make more use of observational data for correction; conversely, it will place more trust in model predictions.

[0047] Accordingly, in step S230, based on the Kalman gain matrix, predicted observations, observed Jacobian matrix, and observed data, the prior state estimate and prior covariance matrix at the current time are updated to obtain the posterior state estimate and posterior covariance matrix at the current time. It should be understood that, due to the influence of electrochemical model bias, time-varying parameters, and environmental noise during battery system operation, the prior estimates obtained solely through state equation derivation often deviate from the true state, while relying solely on sensor data cannot directly obtain the internal state of the battery and is easily affected by observation noise. Therefore, in the technical solution of this application, the prior state estimate and prior covariance matrix at the current time are further updated based on the Kalman gain matrix, predicted observations, observed Jacobian matrix, and observed data to obtain the posterior state estimate and posterior covariance matrix at the current time. This allows the optimal weights determined by the Kalman gain matrix to feed back the deviation between the measured voltage and the model-predicted voltage to the prior state for closed-loop correction. In this way, the accumulated error of open-loop prediction can be eliminated, the optimal estimate of the battery's state of charge and health at the current moment can be obtained under the minimum variance criterion, and the covariance matrix representing the confidence of the estimate can be updated simultaneously, providing a precise state benchmark with uncertainty quantification for subsequent model predictive control.

[0048] Specifically, in a concrete example of this application, the system first performs observation residual calculation and state correction operations. It extracts the measured battery voltage at the current moment from the observation data, performs a difference operation between this voltage and the predicted observation value to obtain the observation residual, then uses the Kalman gain matrix to weight this residual, and finally adds the weighted correction amount to the prior state estimate at the current moment, thereby obtaining the measurement-corrected posterior state estimate at the current moment. This step transforms the measured information of the battery terminal voltage into a direct correction to the internal state of charge and health. The calculation formula is as follows:

[0049] ,

[0050] in, This represents the posterior state estimate at the current time. This represents the prior state estimate at the current moment. Represents the Kalman gain matrix. This represents the measured battery voltage in the observation data. This represents the predicted observation. This formula shows that the posterior estimate equals the prior estimate plus the voltage prediction error weighted by the Kalman gain, achieving closed-loop correction of the internal state using external observations.

[0051] Next, the system performs a posterior covariance matrix update operation. This involves subtracting the product of the Kalman gain matrix and the observation Jacobian matrix from the identity matrix to construct a shrinkage coefficient matrix. This coefficient is then left-multiplied by the current-time prior covariance matrix to obtain the current-time posterior covariance matrix. The calculation formula is as follows:

[0052] ,

[0053] in, Let I denote the posterior covariance matrix at the current time, and let I denote the identity matrix. Represents the Kalman gain matrix. Represents the observed Jacobian matrix. This represents the prior covariance matrix at the current time. Mathematically, this step quantifies the degree to which the uncertainty of the system state estimate decreases after introducing new observation data; that is, the observation information causes the covariance matrix to shrink in a specific direction, thereby improving the confidence of the state estimate.

[0054] Specifically, in step S300, an economic cost function and a degradation cost function are constructed based on the predicted data and the posterior state estimate at the current moment. It should be understood that, since the combined thermal power and energy storage peak-shaving system not only faces time-varying fluctuations in grid load and electricity prices, but is also constrained by the nonlinearity of battery aging on the energy storage side, considering only a single dimension of cost or ignoring the current health of the batteries would lead to short-sighted scheduling strategies or excessive asset lifespan losses. Therefore, in the technical solution of this application, an economic cost function and a degradation cost function are further constructed based on the predicted data and the posterior state estimate at the current moment. This maps future market boundary conditions and current physical state information into a mathematically optimizable objective function, establishing a quantitative relationship between system operation behavior and comprehensive economic costs. This ensures that the subsequent optimization controller, from a full life-cycle perspective, dynamically balances fuel consumption, grid interaction expenditures, and energy storage lifespan losses, formulating an optimal scheduling strategy that balances short-term operating benefits and long-term asset preservation.

[0055] More specifically, in this embodiment of the application, the construction of an economic cost function and a degradation cost function based on predicted data and the posterior state estimate at the current moment includes: constructing a fuel cost function for thermal power units and a grid interaction cost for each time period based on predicted data and a thermal power unit cost model; summing the fuel cost function for thermal power units and the grid interaction cost for each time period to obtain the economic cost function; extracting the current health state estimate from the posterior state estimate at the current moment; constructing an energy storage degradation cost function for each time period based on the current health state estimate and a battery degradation model; and summing the energy storage degradation cost functions for each time period to obtain the degradation cost function.

[0056] In a specific example of this application, the economic cost sub-item construction operation is first performed. This involves extracting the electricity price forecast sequence and load forecast sequence for the next N time periods from the forecast data. Then, combined with a pre-defined thermal power unit cost model, the fuel cost function of the thermal power units and the grid interaction cost for each time period are defined. The thermal power unit fuel cost function is typically modeled as a quadratic function of the thermal power unit output to reflect the coal consumption characteristics under different load rates. The grid interaction cost is determined by the product of the grid interaction power and the time-of-use electricity price. Next, the economic cost aggregation operation is performed, summing all the thermal power unit fuel cost functions and grid interaction costs for the next N forecast time periods to obtain the economic cost function. Its calculation formula is as follows:

[0057] ,

[0058] in, Represents the economic cost function. Indicates the prediction time-domain index. Indicates the current moment. This indicates the prediction time domain length, and a, b, and c represent the fuel cost coefficients of thermal power units. This indicates the power output sequence of thermal power units. Represents the electricity price forecast series. This represents the power sequence of the power grid interaction. This formula unifies the nonlinear coal consumption characteristics of thermal power and the time-varying electricity pricing mechanism of the electricity market into a scalar function of the power variable, providing a clear economic indicator for subsequent optimization.

[0059] Subsequently, a health status extraction operation is performed, parsing the current health status estimate from the posterior state estimate at the current moment. This value reflects the actual aging degree of the battery (such as the capacity decay ratio). Finally, a degradation cost construction and aggregation operation is performed. Based on the current health status estimate and a preset battery degradation model, an energy storage degradation cost function for each time period is constructed, and these functions are summed in the time domain to obtain the degradation cost function. In this model, the degradation cost is designed to be proportional to the square of the charge / discharge power and inversely proportional to the current health status, meaning that the more aged the battery, the higher the unit degradation cost will be for the same charge / discharge behavior. The calculation formula is as follows:

[0060] ,

[0061] in, Represents the degradation cost function. This represents the replacement cost per unit capacity. This represents the magnification degradation factor. This represents an estimate of the current health status. This represents the sequence of energy storage charge and discharge power. Represents the SOC degradation coefficient. Represents the energy storage state of charge sequence. Represents the ideal SOC. This represents the scheduling time interval. This formula, by introducing the current health status as the denominator, achieves protective pricing for aging batteries. That is, as battery health declines, the controller senses higher operating costs, thus automatically limiting high-power charging and discharging behavior and extending the battery's remaining lifespan.

[0062] Specifically, in step S400, based on the posterior covariance matrix at the current time, a bi-objective MPC optimization is performed on the economic cost function and the degradation cost function to obtain the optimal power sequence. It should be understood that since the estimated value of the battery health state is not an absolutely accurate constant, but a random variable with a certain confidence interval, and the choice of control strategy directly affects the observability and estimation accuracy of the system state at future times, optimizing only the determined state value would ignore the risks brought by estimation errors, leading to aggressive control decisions or state divergence. Therefore, in the technical solution of this application, a bi-objective MPC optimization is further performed on the economic cost function and the degradation cost function based on the posterior covariance matrix at the current time to obtain the optimal power sequence. This introduces the uncertainty quantification index of state estimation into the optimization objective, prompting the controller to actively select a control path that can reduce the uncertainty of future states while pursuing economic and lifespan benefits. In this way, a dynamic balance can be achieved between system operating economy, energy storage lifespan protection, and state information acquisition, preventing control failure caused by the accumulation of state estimation errors and improving the robustness of the joint peak-shaving strategy throughout its entire lifespan.

[0063] More specifically, in this embodiment, based on the posterior covariance matrix at the current time, a bi-objective MPC optimization is performed on the economic cost function and the degradation cost function to obtain the optimal power sequence. This includes: constructing an information cost function based on the posterior covariance matrix at the current time, the battery state equation, the process noise covariance matrix, and the posterior state estimate at the current time; constructing an objective function based on the information cost function, the economic cost function, and the degradation cost function; and constraining and solving the objective function to obtain the optimal power sequence.

[0064] In a specific example of this application, the information cost construction operation is first performed. Starting with the posterior covariance matrix at the current time, the evolution of the covariance matrix is ​​iteratively predicted using the Jacobian matrix of the battery state equation on the future trajectory and a preset process noise covariance matrix, until the end of the prediction time domain. During this process, the system calculates the trace of the covariance matrix at the end of the prediction time domain and defines it as the information cost function. The smaller the value of this function, the lower the uncertainty of the future state estimation. Its calculation formula is as follows:

[0065] ,

[0066] in, Represents the information cost function. This represents the prediction time-domain terminal covariance matrix recursively derived from the posterior covariance matrix at the current time. This represents the trace operation of a matrix. This step transforms the abstract uncertainty of future information into a scalar cost that can be minimized, giving the controller the incentive to actively probe to improve information accuracy.

[0067] Next, economic weight coefficients, degradation weight coefficients, and information weight coefficients are introduced. The economic cost function, degradation cost function, and information cost function are then weighted and summed to construct a comprehensive objective function that includes short-term operating costs, long-term lifespan depreciation, and the value of information acquisition. The calculation formula is as follows:

[0068] ,

[0069] in, Describe the objective function. , , These are the weighting coefficients for economy, degradation, and information, respectively. Represents the economic cost function. This represents the degradation cost function. This formula establishes a multi-dimensional value assessment system, enabling the optimization problem to simultaneously consider reducing fuel costs, delaying battery degradation, and maintaining high confidence in state estimation.

[0070] Subsequently, the power balance constraint (the sum of thermal power and energy storage output equals the load forecast), thermal power unit output and ramping constraints, and energy storage charging / discharging power and state of charge constraints are aggregated into the above objective function to construct a nonlinear programming problem. The system calls a nonlinear programming solver (such as IPOPT) to solve this problem, obtaining the sequence of control variables that minimizes the objective function while satisfying all physical constraints, i.e., the optimal power sequence. This process unifies the complex mechanistic model, multi-source cost trade-offs, and physical constraints within a single mathematical framework, outputting a future optimal scheduling strategy that is both economical and robust.

[0071] Specifically, in step S500, the energy storage command power and the thermal power command power are generated based on the optimal power sequence. It should be understood that since the optimal power sequence derived from model predictive control (MPC) is generated based on predicted data, directly executing the first element of this sequence completely ignores the almost inevitable real-time deviation between the actual load and the predicted value at the scheduling moment. This design separates the tasks of prediction optimization and real-time control at two different time scales, failing to form an effective feedback correction closed loop to cope with the uncertainty of prediction. At a deeper level, this mechanism ignores the fundamental difference between thermal power units and energy storage systems in terms of response speed and regulation costs. The millisecond-level rapid response capability of energy storage makes it an ideal asset for handling high-frequency, sudden fluctuations, while the minute-level slow regulation characteristics of thermal power units mean that frequent changes will significantly increase additional costs. The original mechanism passively applies the pressure of correcting all real-time deviations to the slow-responding thermal power units, not only causing the idleness and waste of the high-quality regulation resource of energy storage, but also directly leading to the system's actual power failing to accurately track the actual load, thus causing control accuracy loss and grid assessment risks. To overcome the aforementioned shortcomings, this solution proposes a real-time adaptive correction method for MPC commands based on dynamic marginal cost in an optional embodiment. This method no longer treats the MPC planning results as rigid final commands, but rather as the long-term economically optimal operating baseline, and introduces a real-time, closed-loop feedback correction layer based on economic principles. In this technical solution, energy storage command power and thermal power command power are further generated based on the optimal power sequence, thereby introducing a real-time adaptive correction mechanism based on dynamic marginal cost. The MPC planning results are treated as the long-term economically optimal operating baseline, and the actual load deviation is quantified using real-time feedback. Then, a dynamic allocation factor is calculated based on the real-time marginal ratio of thermal power fuel cost to energy storage degradation cost, intelligently allocating the power deviation to the most economical regulation resource. This enables a shift from predictive optimization open-loop control to real-time feedback closed-loop control, ensuring that each correction to the real-time deviation achieves economic optimality after comprehensively considering instantaneous fuel cost and long-term asset depreciation. This significantly improves the system's response speed and tracking accuracy to load prediction errors while maximizing the overall economic benefits throughout its entire lifecycle.

[0072] Figure 4 This document presents a flowchart illustrating the generation of energy storage command power and thermal power command power based on an optimal power sequence in a combined thermal power and energy storage peak-shaving control method considering energy storage lifetime, according to embodiments of this application. Figure 4As shown, step S500 includes: S510, performing real-time deviation quantification and baseline command extraction on the measured load power, predicted load power, and optimal power sequence to obtain the energy storage baseline power, thermal power baseline power, and real-time power deviation; S520, calculating the real-time correction dynamic allocation factor based on the energy storage baseline power and thermal power baseline power to obtain the dynamic allocation factor; S530, performing command fusion correction and issuance on the energy storage baseline power, thermal power baseline power, and real-time power deviation based on the dynamic allocation factor to obtain the energy storage command power and thermal power command power.

[0073] Accordingly, in step S510, real-time deviation quantification and baseline command extraction are performed on the measured load power, predicted load power, and optimal power sequence to obtain the energy storage baseline power, thermal power baseline power, and real-time power deviation. It should be understood that, because traditional model-based predictive control command dispatch mechanisms typically execute the first element of the optimal sequence derived from predicted data optimization, this plan-and-execute mode completely ignores the almost inevitable real-time deviation between the actual load and the predicted value at the scheduling moment. This leads to the separation of the two tasks at different time scales—predictive optimization and real-time control—failing to form an effective feedback correction closed loop to address the uncertainty of prediction. Furthermore, by ignoring the fundamental differences in response speed and regulation costs between thermal power units and energy storage systems, the pressure to correct all real-time deviations is passively applied to the slow-responding thermal power units, resulting in control accuracy loss and grid assessment risks. Therefore, in the technical solution of this application, real-time deviation quantification and baseline command extraction are further performed on the measured load power, predicted load power, and optimal power sequence to obtain the energy storage baseline power, thermal power baseline power, and real-time power deviation. This is to take into account that any effective feedback correction must be based on the accurate quantification of the deviation, and to regard the planning result of MPC as no longer a rigid final command, but as a long-term economically optimal operating baseline, and to transform the unavoidable prediction uncertainty into a clear and quantifiable real-time control target. In this way, a clear input signal can be provided for subsequent intelligent and dynamic correction, so that the entire control system has the ability to perceive changes in the real world and ensure that every correction of the real-time deviation is based on accurate quantified data.

[0074] Specifically, in a particular example of this application, the measured load power of the power grid is first synchronously collected at scheduling time k, and the first element is extracted from the optimal power sequence passed in the previous step as the energy storage baseline power. And based on the load forecast values ​​used during optimization. The baseline power of the thermal power unit can be calculated by back-calculating the power balance relationship. The calculation formula is as follows:

[0075] ,

[0076] in, This represents the baseline output command of the thermal power unit at time k. Let k be the load forecast value. This is the baseline output command for energy storage at time k. This step establishes the optimal operating point under ideal prediction conditions, serving as a reference benchmark for subsequent deviation allocation.

[0077] Next, the difference between the actual load and the predicted load is calculated to obtain the real-time power deviation Δ. This can be expressed as a formula:

[0078] ,

[0079] in, This represents the real-time power deviation at time k. Let k be the measured load value at time k. This step concretizes the uncertainty at the forecasting level into a physical power deficit or surplus, providing a clear control deviation signal for the subsequent correction layer based on dynamic marginal cost. In this way, the unavoidable forecasting uncertainty is transformed into a clear and quantifiable real-time control objective. This provides a clear input signal for subsequent intelligent and dynamic corrections, enabling the entire control system to perceive changes in the real world.

[0080] Accordingly, in step S520, a real-time correction dynamic allocation factor is calculated based on the energy storage baseline power and the thermal power baseline power to obtain the dynamic allocation factor. It should be understood that after quantifying the total deviation, an intelligent strategy must be designed to determine which regulation resource should undertake the correction task, rather than simply allocating it evenly or at a fixed rate, because the regulation costs of thermal power units and energy storage systems differ significantly at different operating points. Therefore, in the technical solution of this application, a real-time correction dynamic allocation factor is further calculated based on the energy storage baseline power and the thermal power baseline power to obtain the dynamic allocation factor. This utilizes the principle of marginal analysis to calculate the marginal regulation costs of thermal power and energy storage assets at the current baseline operating point, and miniaturizes and applies the classic economic dispatch principle to instantaneous control. This enables the allocation of deviation correction tasks to have high scenario adaptability, ensuring that each correction to the real-time deviation is an economically optimal decision after comprehensively considering instantaneous fuel costs and long-term asset depreciation, thereby fundamentally improving the system's operational economy in uncertain environments.

[0081] Specifically, in this embodiment of the application, the dynamic allocation factor is calculated in real time based on the energy storage baseline power and the thermal power baseline power to obtain the dynamic allocation factor, including: determining the thermal power marginal cost based on the thermal power baseline power; determining the energy storage marginal cost based on the energy storage baseline power; and determining the dynamic allocation factor based on the thermal power marginal cost and the energy storage marginal cost.

[0082] More specifically, the marginal cost of thermal power is determined based on the baseline power output. It should be understood that after quantifying the total deviation, an intelligent strategy must be designed to determine which regulatory resource should undertake the correction task, rather than simply allocating it evenly or at a fixed rate. Furthermore, the fuel cost characteristics of thermal power units are typically a quadratic function of output, meaning that their regulation cost is not a constant value but exhibits significant nonlinear fluctuations with changes in the current baseline operating point. Therefore, in the technical solution of this application, the marginal cost of thermal power is further determined based on the baseline power output. This allows for the differentiation of the fuel cost function of the thermal power unit at the current baseline output point, thereby accurately quantifying the instantaneous fuel cost change rate caused by increasing or decreasing the unit output of the thermal power unit under the current specific operating conditions. This allows for the capture of the economic characteristics differences of thermal power units at different load rates, providing accurate quantitative basis for subsequent real-time comparison with the marginal cost of energy storage systems, ensuring that the system can identify and utilize the low-cost regulation range of thermal power units for deviation correction.

[0083] In a specific example of this application, the system reads the baseline power of thermal power plants at the current scheduling moment and calls a preset quadratic function model of the fuel cost of thermal power units. The coefficients in this model reflect the coal consumption characteristics of the units. Subsequently, the system performs differentiation to calculate the cost function at the point of tangency with the baseline power of thermal power plants. The derivative value is used to obtain the marginal cost of thermal power. The calculation formula is as follows:

[0084] ,

[0085] in, This represents the marginal adjustment cost of the thermal power unit at time k. For fuel cost function, For thermal power output, This represents the baseline power output of thermal power plants. and These are the coefficients of its quadratic fuel cost model. This step, through analytical calculation, accurately measures the instantaneous fuel cost required for a thermal power unit to generate an additional unit of power at a specific operating point, providing a real-time economic indicator for the calculation of the dynamic allocation factor on the thermal power side.

[0086] More specifically, the marginal cost of energy storage is determined based on the baseline power of the energy storage system. It should be understood that since the regulation cost of an energy storage system is not reflected in immediate fuel consumption, but rather in implicit cycle life loss, and this loss exhibits a strongly non-linear relationship with the current health state and power output, simply treating energy storage as a zero-marginal-cost resource would lead to overuse of batteries during the aging process, thus accelerating their decommissioning. Therefore, in the technical solution of this application, the marginal cost of energy storage is further determined based on the baseline power of the energy storage system. This allows for the partial derivative calculation of the power-related portion of its degradation cost function at the baseline output point. This directly links the real-time regulation behavior of energy storage to its long-term life loss, quantifying the long-term economic cost required for energy storage to undertake an additional unit of regulation task under the current battery condition. This lays the foundation for subsequent economic comparisons with thermal power on the same level.

[0087] In a specific example of this application, the system performs a differential operation on the degradation cost function to calculate the marginal cost of energy storage. That is, the power-related part of its degradation cost function at the baseline output point. The system first calls a pre-built energy storage degradation cost function, which describes the functional relationship between power, health status, and economic losses. Then, the system calculates the partial derivative of this function with respect to the energy storage output variable, substitutes the current energy storage baseline power into the derivative function, and finally obtains the marginal cost of energy storage by taking the absolute value. The calculation formula is as follows:

[0088] ,

[0089] in, This represents the marginal adjustment cost of energy storage at time k. For the degradation cost function, As a variable in energy storage output, For energy storage baseline power, It is the replacement cost per unit capacity. It is the magnification degradation factor. This is an estimate of the current health status. This refers to the scheduling time interval. This step transforms the implicit depreciation cost of the battery into an explicit marginal price signal through mathematical derivation, particularly in the formula... Located in the denominator, this indicates that when the battery health is poor, the marginal cost increases significantly, thus economically suppressing the high-power utilization of aging batteries. This mechanism directly links the real-time adjustment behavior of energy storage to its long-term lifespan degradation, quantifying the long-term economic cost of having energy storage undertake an additional unit of adjustment task under the current battery condition.

[0090] More specifically, the dynamic allocation factor is determined based on the marginal cost of thermal power and the marginal cost of energy storage. It should be understood that if the allocation of real-time power deviation correction tasks adopts a fixed proportion or simple averaging method, it will be unable to adapt to the significantly different marginal adjustment costs of thermal power units and energy storage systems under different operating conditions, thus causing the system operation to deviate from the economically optimal trajectory. Therefore, in the technical solution of this application, the dynamic allocation factor is further determined based on the marginal cost of thermal power and the marginal cost of energy storage, thereby constructing a weighting coefficient that can reflect the relative economic advantages of the two adjustment resources in real time, minimizing the classic economic dispatch principle and applying it to instantaneous control. This enables the allocation of deviation correction tasks to have a high degree of scenario adaptability, ensuring that every correction amount for real-time deviation is an economically optimal decision after comprehensively weighing the increase in instantaneous fuel costs of thermal power and the long-term asset depreciation of energy storage, thereby fundamentally improving the system's full life-cycle operational economy in the face of load forecast uncertainty.

[0091] In a specific example of this application, the system performs a ratio construction operation for the dynamic allocation factor. The system retrieves the marginal adjustment cost of the thermal power unit and the marginal adjustment cost of the energy storage system at time k, calculated in previous steps. A preset, minimal positive number is introduced to prevent the denominator from being zero during the calculation. Subsequently, the system uses the marginal cost of thermal power as the numerator and the sum of the marginal costs of thermal power and energy storage as the denominator, and calculates the dynamic allocation factor at time k through division. The dynamic allocation factor is determined by the following formula:

[0092] ,

[0093] in, For the marginal cost of thermal power, For the marginal cost of energy storage, It is to prevent small positive numbers with a denominator of zero. This is a dynamic allocation factor. The allocation logic defined by this formula indicates that when the marginal adjustment cost of thermal power is higher than that of energy storage (e.g., thermal power is in a high coal consumption range or energy storage is in a healthy and low-loss range), the larger the calculated allocation factor value, the higher the proportion of subsequent correction tasks allocated to energy storage. Conversely, the proportion of energy storage actions is reduced, thus achieving optimal matching of macroeconomics at the micro-level of real-time control. In this way, the classic economic dispatch principle is miniaturized and applied to instantaneous control. The dynamic characteristics of this factor make the allocation of deviation correction tasks highly adaptable to different scenarios. The purpose and effect of this step is to ensure that every correction to real-time deviations is an economically optimal decision after comprehensively considering instantaneous fuel costs and long-term asset depreciation, thereby fundamentally improving the system's operational economy under uncertain environments.

[0094] Accordingly, in step S530, based on the dynamic allocation factor, the energy storage baseline power, thermal power baseline power, and real-time power deviation are fused and corrected with commands to obtain the energy storage command power and the thermal power command power. It should be understood that after calculating the economically optimal correction strategy (i.e., the dynamic allocation factor), this strategy must be combined with the original long-term planning baseline, and the final generated physical commands must strictly comply with the physical safety constraints of the equipment. Otherwise, the theoretically optimal solution may cause equipment overload or safety accidents during actual execution. Therefore, in the technical solution of this application, the energy storage baseline power, thermal power baseline power, and real-time power deviation are further fused and corrected with commands based on the dynamic allocation factor to obtain the energy storage command power and the thermal power command power. This serves as the final barrier from the computational domain to the physical domain, safely and accurately transforming the economically optimal correction strategy into executable physical control commands. That is, after calculating the optimal correction strategy, it is necessary to fuse this strategy with the original baseline plan and ensure that the final generated physical commands are within the safe operating range of the equipment. In this way, a complete closed loop can be completed from perceiving deviations and making intelligent decisions to precise execution, ensuring that the system maintains stable, efficient and safe operation while responding to load fluctuations in real time.

[0095] Specifically, in one particular example of this application, the system first performs a deviation decomposition operation, utilizing the dynamic allocation factor calculated in the preceding steps. The total real-time power deviation Decomposed into the correction amount that the energy storage system should bear. Correction amount to be borne by thermal power units This step uses mathematical calculations to allocate the correction tasks to the two subsystems in an economically viable proportion. The calculation formula is as follows:

[0096] ,

[0097] ,

[0098] in, The amount of correction that energy storage should bear, For dynamic allocation factors, For real-time power deviation, This is the amount of correction that thermal power plants should bear.

[0099] Next, the system performs command fusion and safety verification operations, adding the calculated correction amounts to the corresponding baseline power to generate the final command power. Specifically, for energy storage commands, the system introduces a saturation limiting function for mandatory safety verification to ensure that the final issued command absolutely does not exceed the physical power limit of the energy storage device (such as the maximum power of the inverter or the maximum charge / discharge capacity of the battery). The calculation formula is as follows:

[0100] ,

[0101] ,,

[0102] in, and These are the final power commands issued to the energy storage system and the thermal power unit, respectively. It is a saturation limiting function. and These are the baseline power for energy storage and thermal power, respectively. and These are the lower and upper limits of the physical power of energy storage. This step acts as the final gate from the computational domain to the physical domain, safely and accurately transforming the economically optimal correction strategy calculated in the previous step into executable physical control commands, which are then issued to the field devices. This completes the closed loop from sensing deviations and intelligent decision-making to precise execution. This step ensures that the commands finally issued to the field devices not only include economic corrections for real-time deviations but also strictly comply with physical safety boundaries, ensuring the stability, efficiency, and safety of the system operation.

[0103] Through the above preferred embodiments, a fundamental transformation has been achieved from a rigid control framework relying on prediction and open-loop execution to a flexible control architecture with two time scales, combining long-term planning and real-time feedback. The technical effects are reflected in a significantly enhanced system response speed and suppression capability to load forecasting errors, and the ability of the power plant's actual output power to more accurately track grid dispatch commands, thereby greatly improving the system's control accuracy and operational stability. The ultimate technical objective is to ensure that every adjustment behavior of the system balances short-term operating costs and long-term asset health by dynamically correcting real-time deviations at each dispatching moment, thus maximizing the comprehensive economic benefits of the thermal power-storage combined peak-shaving system throughout its entire lifecycle while addressing grid uncertainties.

[0104] In summary, the combined peak-shaving control method for thermal power and energy storage considering energy storage lifespan, according to the embodiments of this application, is clarified. First, an extended Kalman filter is used to estimate the battery's SoH (Solar Energy Flow) online, and the uncertainty of this estimate is quantified using the covariance matrix. Subsequently, this uncertainty information (i.e., the covariance matrix) is introduced into the optimization objective function of model predictive control, forming a three-objective optimization problem including economic cost, degradation cost, and information cost. The information cost is specifically used to penalize control strategies that may lead to a decrease in the accuracy of future SoH estimation or an increase in uncertainty. In this way, when formulating energy storage charging / discharging and thermal power unit output strategies, the controller not only weighs current economic benefits against battery life loss but also actively avoids and manages the uncertainty of future state estimates, and even takes probing actions to improve information accuracy. This fundamentally solves the problem in traditional schemes where the controller blindly trusts erroneous state values, leading to aggressive operation and uncertainty accumulation. It achieves a dynamic balance between system economy, energy storage lifespan, and information acquisition capability, ensuring the robustness and optimality of the combined peak-shaving strategy throughout its entire lifespan.

[0105] Furthermore, a combined thermal power and energy storage peak-shaving control system that takes into account energy storage lifespan is also provided.

[0106] Figure 5 This is a block diagram of a combined thermal power and energy storage peak-shaving control system considering energy storage lifetime, according to an embodiment of this application. Figure 5 As shown, the thermal power and energy storage joint peak-shaving control system 100 considering energy storage lifetime according to an embodiment of this application includes: a prediction and observation data construction module 110, used to construct prediction data based on load prediction sequence and electricity price prediction sequence, and to construct observation data based on measured battery voltage, measured battery temperature and measured battery current; a posterior estimation module 120, used to perform online SoH estimation and uncertainty quantification based on EKF on the posterior state estimate and the posterior covariance matrix of the previous time based on the observation data and the control input of the previous time to obtain the posterior state estimate and the posterior covariance matrix of the current time; a cost function construction module 130, used to construct an economic cost function and a degradation cost function based on the prediction data and the posterior state estimate of the current time; a bi-objective MPC optimization module 140, used to perform bi-objective MPC optimization on the economic cost function and the degradation cost function based on the posterior covariance matrix of the current time to obtain the optimal power sequence; and a command power generation module 150, used to generate energy storage command power and thermal power command power based on the optimal power sequence.

[0107] As described above, the thermal power and energy storage combined peak shaving control system 100 considering energy storage lifetime according to the embodiments of this application can be implemented in various wireless terminals, such as servers with thermal power and energy storage combined peak shaving control algorithms considering energy storage lifetime. In one possible implementation, the thermal power and energy storage combined peak shaving control system 100 considering energy storage lifetime according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the thermal power and energy storage combined peak shaving control system 100 considering energy storage lifetime can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the thermal power and energy storage combined peak shaving control system 100 considering energy storage lifetime can also be one of the many hardware modules of the wireless terminal.

[0108] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for joint peak-shaving control of thermal power and energy storage considering energy storage lifetime, characterized in that, include: Forecast data is constructed based on load forecast sequences and electricity price forecast sequences, and observation data is constructed based on measured battery voltage, measured battery temperature and measured battery current. Based on the observation data and the control input of the previous time step, online SoH estimation and uncertainty quantification based on EKF are performed on the posterior state estimate and the posterior covariance matrix of the previous time step to obtain the posterior state estimate and the posterior covariance matrix of the current time step. Based on the predicted data and the posterior state estimate at the current moment, construct the economic cost function and the degradation cost function; Based on the posterior covariance matrix at the current time, a bi-objective MPC optimization is performed on the economic cost function and the degradation cost function to obtain the optimal power sequence, including: constructing an information cost function based on the posterior covariance matrix at the current time, the battery state equation, the process noise covariance matrix, and the posterior state estimate at the current time; Based on the information cost function, economic cost function, and degradation cost function, an objective function is constructed; the objective function is constrained, aggregated, and solved to obtain the optimal power sequence. Based on the optimal power sequence, the energy storage command power and the thermal power command power are generated.

2. The method for joint peak-shaving control of thermal power and energy storage considering energy storage life as described in claim 1, characterized in that, Based on observation data and the control input from the previous time step, online SoH estimation and uncertainty quantification based on EKF are performed on the posterior state estimate and posterior covariance matrix from the previous time step to obtain the posterior state estimate and posterior covariance matrix from the current time step, including: Based on the battery state equation and process noise covariance matrix, state and covariance predictions are performed on the control input at the previous time step, the posterior state estimate at the previous time step, and the posterior covariance matrix at the previous time step to obtain the prior state estimate and the prior covariance matrix at the current time step. Based on the observation data, battery observation equation and measurement noise covariance matrix, observation linearization and Kalman gain calculation are performed on the current time prior state estimate and the current time prior covariance matrix to obtain the Kalman gain matrix, predicted observation value and observation Jacobian matrix; Based on the Kalman gain matrix, predicted observations, observed Jacobian matrix, and observation data, the state and covariance of the prior state estimate and the prior covariance matrix at the current time are updated to obtain the posterior state estimate and the posterior covariance matrix at the current time.

3. The method for joint peak-shaving control of thermal power and energy storage considering energy storage life as described in claim 1, characterized in that, Based on the predicted data and the posterior state estimate at the current moment, the economic cost function and the degradation cost function are constructed, including: Based on the predicted data and the thermal power unit cost model, the fuel cost function of thermal power units and the power grid interaction cost for each time period are constructed. The economic cost function is obtained by summing the fuel cost function of thermal power units and the grid interaction cost of each time period. Extract the current health status estimate from the posterior state estimate at the current time step; Based on the current health status estimate and battery degradation model, construct the energy storage degradation cost function for each time period; The energy storage degradation cost function is obtained by summing the energy storage degradation cost functions for each time period.

4. The method for joint peak-shaving control of thermal power and energy storage considering energy storage life as described in claim 1, characterized in that, Based on the optimal power sequence, the energy storage command power and the thermal power command power are generated, including: Real-time deviation quantification and baseline command extraction are performed on the measured load power, predicted load power and optimal power sequence to obtain the energy storage baseline power, thermal power baseline power and real-time power deviation; The dynamic allocation factor is calculated by real-time correction based on the energy storage baseline power and the thermal power baseline power. Based on the dynamic allocation factor, the baseline power of energy storage, the baseline power of thermal power, and the real-time power deviation are fused, corrected, and issued to obtain the commanded power of energy storage and the commanded power of thermal power.

5. The method for joint peak-shaving control of thermal power and energy storage considering energy storage life as described in claim 4, characterized in that, The dynamic allocation factor is calculated in real time based on the energy storage baseline power and the thermal power baseline power to obtain the dynamic allocation factor, including: Determine the marginal cost of thermal power based on the baseline power output; Determine the marginal cost of energy storage based on the baseline power of energy storage; The dynamic allocation factor is determined based on the marginal cost of thermal power and the marginal cost of energy storage.

6. The method for joint peak-shaving control of thermal power and energy storage considering energy storage lifetime as described in claim 5, characterized in that, Determining the dynamic allocation factor based on the marginal cost of thermal power and the marginal cost of energy storage includes: determining the dynamic allocation factor using the following formula, wherein the formula is: ， in, For the marginal cost of thermal power, For the marginal cost of energy storage, It is to prevent small positive numbers with a denominator of zero. This is a dynamic allocation factor.

7. The method for joint peak-shaving control of thermal power and energy storage considering energy storage life as described in claim 4, characterized in that, Based on a dynamic allocation factor, command fusion correction and issuance are performed on the energy storage baseline power, thermal power baseline power, and real-time power deviation to obtain the energy storage command power and thermal power command power. This includes command fusion correction using the following formula: ， ， ， ， in, For real-time power deviation, For dynamic allocation factors, For the amplitude limiting function, For energy storage command power, and These are the lower and upper limits of the physical power of energy storage. Baseline output command, The amount of correction that energy storage should bear, The amount of correction to be borne by thermal power plants. This refers to the commanded power output of thermal power plants.

8. A combined thermal power plant and energy storage peak-shaving control system considering energy storage lifespan, characterized in that, include: The forecast and observation data construction module is used to construct forecast data based on load forecast sequence and electricity price forecast sequence, and to construct observation data based on measured battery voltage, measured battery temperature and measured battery current. The posterior estimation module is used to perform online SoH estimation and uncertainty quantification based on EKF on the posterior state estimate and the posterior covariance matrix of the previous time step, based on the observation data and the control input of the previous time step, to obtain the posterior state estimate and the posterior covariance matrix of the current time step. The cost function construction module is used to construct the economic cost function and the degradation cost function based on the predicted data and the posterior state estimate at the current time. The dual-objective MPC optimization module is used to perform dual-objective MPC optimization on the economic cost function and the degradation cost function based on the current time posterior covariance matrix to obtain the optimal power sequence. This includes: constructing the information cost function based on the current time posterior covariance matrix, the battery state equation, the process noise covariance matrix, and the current time posterior state estimate. Based on the information cost function, economic cost function, and degradation cost function, an objective function is constructed; the objective function is constrained, aggregated, and solved to obtain the optimal power sequence. The command power generation module is used to generate energy storage command power and thermal power command power based on the optimal power sequence.

Images