Thermal power storage frequency modulation auxiliary service benefit maximization method

Abstract

The invention belongs to the technical field of power system automation and energy storage regulation and control, particularly relates to a thermal power energy storage frequency modulation auxiliary service benefit maximization method, and aims to solve the problems of disjunction between technical response and economic benefit, large equipment loss and poor market adaptability in a traditional frequency modulation strategy. According to the method, a multi-time-scale instruction analysis model is constructed, thermal power coal consumption and energy storage loss characteristics are fused, a dynamic cost-income coupling function is established, and thermal storage power is distributed in real time based on a rolling time domain optimization algorithm; and dynamically correcting model parameters by executing feedback and an online learning mechanism, and synchronously generating a market bidding strategy and benefit pre-judgment report. The system supports state-of-charge boundary adaptive adjustment and equipment collaborative aging evaluation, and is deployed at an edge computing node to realize millisecond-level response. According to the invention, the economic benefit of frequency modulation is obviously improved by 15-25%, the score of frequency modulation performance is more than 95, the energy storage life is prolonged by more than 30%, the fluctuation of unit income is reduced by 40%, and full-link closed-loop intelligent regulation and control and market competitiveness enhancement are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power system automation and energy storage regulation, and particularly relates to a method for maximizing the benefit of thermal power energy storage frequency modulation auxiliary service. BACKGROUND

[0002] With the increasing demand for frequency stability and regulation flexibility of the power system, the coordinated participation of thermal power units and energy storage systems in frequency modulation auxiliary service has become a key path to improve the response capability of the power grid. Traditional thermal power frequency modulation relies on the climbing rate and inertia response of the unit itself, and the regulation process is obviously lagging, which is difficult to meet the requirements of suppressing frequency fluctuations in seconds or even milliseconds under high proportion of new energy access. Especially in the operation scenario of multi-time scale frequency modulation instruction superposition and severe load fluctuation, there is often a structural conflict between the power distribution strategy of the thermal storage combined system and the economic optimization goal, which puts forward higher requirements for the dynamic adaptation ability, cost accounting accuracy and benefit maximization mechanism of the control algorithm.

[0003] However, the existing scheduling model mostly uses static weight distribution or empirical threshold control, which is difficult to establish a dynamic balance between real-time electricity price fluctuations, equipment aging loss and frequency modulation performance evaluation indexes, resulting in frequent charging and discharging of the energy storage system or excessive regulation of the thermal power unit, which shortens the service life of the equipment and reduces the overall benefit. At the same time, the multi-source data (such as AGC instruction, SOC state, coal consumption curve) in the response process of the frequency modulation instruction lacks a unified value mapping framework, and the traditional method cannot realize the precise coupling of frequency modulation action and economic return. In addition, the benefit evaluation relies on post-statistics and linear regression, which cannot predict the optimal bidding strategy under the nonlinear market mechanism, resulting in the competitiveness and profit space of the system in the auxiliary service market being severely compressed.

[0004] Therefore, a method for maximizing the benefit of thermal power energy storage frequency modulation auxiliary service is expected. SUMMARY

[0005] The purpose of the present application is to provide a method for maximizing the benefit of thermal power energy storage frequency modulation auxiliary service, which can effectively solve the problems in the background art.

[0006] To achieve the above purpose, the technical solution adopted by the present application is: The application discloses a method for maximizing the benefits of a thermal power energy storage frequency modulation auxiliary service, which comprises the following specific steps: step (1), constructing a multi-time scale frequency modulation instruction analysis model: based on historical data and real-time data of power grid automatic power generation control instructions, a sliding window mechanism is adopted to perform multi-scale decomposition on the instructions at a second level, a minute level and an hour level, power change rates, fluctuation frequencies and duration characteristics at various time scales are extracted, and an instruction mapping relationship matched with a thermal power unit climbing capacity and a response speed of an energy storage system is established; step (2), establishing a dynamic cost-benefit coupling function: a coal consumption characteristic curve of a thermal power unit, a charging and discharging cycle loss model of an energy storage system, real-time auxiliary service market price signals and frequency modulation performance evaluation rules are fused to construct a dynamic coupling function with the economic value of a unit frequency modulation action as a target, wherein a thermal power regulation cost is quantified according to a load rate and a coal consumption differential relationship, and an energy storage loss cost is quantified according to a state of charge variation and a cycle depth nonlinear mapping; step (3), designing an adaptive power distribution optimization algorithm: the dynamic cost-benefit coupling function is taken as an optimization target, the current state of charge of the energy storage system, the operation constraints of the thermal power unit and the multi-scale characteristics of the frequency modulation instructions are combined, and a rolling time domain optimization framework is adopted to solve the optimal power distribution proportion of the thermal power and the energy storage in each scheduling period, so that the economic benefits are maximized under the premise that the frequency modulation performance indexes are met; step (4), implementing frequency modulation action execution and real-time feedback correction: thermal power unit output adjustment instructions and energy storage system charging and discharging power instructions are generated according to the optimization results, actual response deviations, equipment state parameters and market settlement data in the execution process are synchronously collected, and the weight coefficients and model parameters in the cost-benefit coupling function are dynamically corrected through an online learning mechanism; and step (5), generating an auxiliary service bidding strategy and benefit prediction report: based on historical optimization results and market clearing rules, a nonlinear regression and scenario simulation method is adopted to predict the expected benefits and risk levels under different bidding capacities and price combinations, an optimal bidding strategy and a benefit prediction report for the next 24 hours are output, and the benefit prediction report is used to guide the market participation decision-making of the next day.

[0007] Preferably, the time length of the sliding window in step (1) is dynamically adjusted according to the fluctuation intensity of the frequency modulation instructions, the second-level window length is 5-15 seconds, the minute-level window length is 3-10 minutes, the hour-level window length is 30-60 minutes, and the window overlap rate is 50%, so as to ensure the continuity and stability of the feature extraction.

[0008] Preferably, the coal consumption characteristic curve of the thermal power unit in step (2) is obtained by fitting field thermal test data, a cubic spline interpolation method is adopted to construct a continuous mapping relationship between a load rate and unit coal consumption, and the calculation precision of the coal consumption differential cost reaches 0.1 grams of standard coal per kilowatt hour; the cycle loss model of the energy storage system is established based on an exponential attenuation relationship between a battery health state and a deep discharge frequency, and the equivalent loss cost increases by 1.8 times for each 10% increase in cycle depth.

[0009] Preferably, the rolling horizon optimization scheduling period in step (3) is 2 seconds, the prediction horizon is 60 seconds, and the optimization variables include the incremental thermal power With energy storage power The constraints include the minimum technical output of the thermal power unit, the maximum ramp rate, the upper and lower limits of the state of charge of the energy storage system (set to 20% to 80%), and the charge and discharge power limits. The optimization solution uses an interior point method, and the time consumption of a single calculation is less than 500 milliseconds.

[0010] Preferably, the actual response deviation in step (4) is defined as the absolute value integral of the difference between the automatic generation control instruction and the actual output of the thermal storage combined system. The frequency regulation performance evaluation index includes the regulation rate, the regulation accuracy and the response time, and the weight coefficients are automatically updated according to the latest evaluation rules of the power grid. The online learning mechanism uses the recursive least squares method, and the market sensitivity parameters in the cost-benefit coupling function are updated once every 10 scheduling periods.

[0011] Preferably, the nonlinear regression model in step (5) uses a Gaussian process regression algorithm, and the input features include historical winning prices, frequency regulation capacity demand, new energy output fluctuation rate and system reserve margin. The output is the expected unit capacity benefit; scene simulation sets not less than 1000 market scenarios, covering the price fluctuation range of 0.5 times to 2.0 times the current average price, and the benefit prediction report includes the expected value, standard deviation and 95% confidence interval.

[0012] Preferably, the method further comprises a dynamic adjustment mechanism for the state of charge safety boundary of the energy storage system: according to the future 2-hour load forecast and the frequency regulation instruction intensity forecast, the lower limit and the upper limit of the state of charge are adjusted in real time. When the predicted frequency regulation demand is severe, the state of charge operating range is expanded to 15% to 85% to improve the system regulation margin, and the weight of the cycle loss cost is increased to prevent excessive use.

[0013] Preferably, the method further comprises a coordinated aging evaluation module for the thermal power unit and the energy storage system: based on the cumulative number of adjustments, the duration of deep adjustment and the temperature stress data, the equivalent aging factors of the thermal power boiler and the energy storage battery are calculated respectively. When any device aging factor exceeds the preset threshold (thermal power is 0.85, and energy storage is 0.90), the priority of the device in power distribution is automatically reduced, and the device maintenance suggestion is marked in the benefit prediction report.

[0014] Preferably, the method is deployed on the edge computing node of the thermal storage combined control system, and has bidirectional communication capability with the power grid dispatching master station, the power plant monitoring system and the energy storage energy management system. The communication protocol uses the general standard of the power system, the data refresh frequency is not less than 1 times per second, and the overall response delay of the system is less than 1 second.

[0015] Compared with the prior art, the present application has the following beneficial effects: Realize the precise coupling of frequency modulation action and economic value By constructing a dynamic cost-benefit coupling function, the coal consumption of thermal power, the loss of energy storage, the market price and the evaluation rules are unified into the optimization framework, so that each frequency modulation action corresponds to a clear economic value evaluation, solving the problem of disconnection between technical response and economic benefit in traditional methods, and the frequency modulation benefit can be improved by 15% to 25%.

[0016] Improve the adaptability of multi-time scale frequency modulation instructions Adopting a multi-scale instruction analysis and rolling optimization mechanism, the system can simultaneously cope with high-frequency fluctuations at the second level and trend changes at the hour level, and in the high penetration scenario of new energy, the frequency modulation performance evaluation score is stable above 95 points, which is more than 20 percentage points higher than the static allocation strategy.

[0017] Extend the service life of key equipment Through the online feedback correction and aging evaluation module, the deep charge and discharge of the energy storage system and the frequent large adjustment of the thermal power unit are effectively avoided, the cycle life of the energy storage system is extended by more than 30%, the thermal stress damage of the key components of the thermal power unit is reduced by 25%, and the operation and maintenance cost is significantly reduced.

[0018] Enhance the competitiveness of auxiliary service market Based on the benefit prediction and bidding strategy generation of the nonlinear market mechanism, the system has forward-looking market participation ability, and can still maintain the stability of the income in the period of severe price fluctuations, and the bidding strategy optimization can improve the winning rate by 18% and reduce the standard deviation of the unit capacity income fluctuation by 40%.

[0019] Realize the whole-link closed-loop intelligent regulation and control From instruction analysis, optimization allocation, execution feedback to strategy generation, a complete data-driven closed loop is formed, and the system has self-learning, self-adapting and self-optimizing capabilities, and can continuously maintain the state of maximum benefit in complex operating environment without manual intervention. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 It is the schematic diagram of the overall technical scheme architecture of the present application. DETAILED DESCRIPTION

[0021] Embodiment 1 Please refer to Figure 1 In order to make the purpose, technical scheme and advantages of the present application more clear and obvious, the present application will be further described in detail in combination with specific embodiments.

[0022] Currently, in the process that the thermal power energy storage combined system participates in the frequency modulation auxiliary service of the power grid, there are problems of disconnection between frequency modulation action and economic benefits, insufficient response ability of multi-time scale instruction, accelerated equipment aging and lack of forward-looking market bidding strategy and the like. In view of the above technical problems, the present application provides a thermal power energy storage frequency modulation auxiliary service benefit maximization method, and is applied to a thermal power energy storage frequency modulation auxiliary service benefit maximization method.

[0023] In the above-mentioned method for maximizing the benefits of thermal power energy storage frequency modulation auxiliary services, the step (1) of constructing a multi-time scale frequency modulation instruction analysis model: based on the historical data and real-time data of the automatic generation control instruction of the power grid, a sliding window mechanism is used to perform multi-scale decomposition of the instruction at the second, minute and hour levels, the power change rate, fluctuation frequency and duration characteristics at each time scale are extracted, and an instruction mapping relationship matching the climbing ability of the thermal power unit and the response speed of the energy storage system is established. Specifically, the time length of the sliding window in step (1) is dynamically adjusted according to the fluctuation intensity of the frequency modulation instruction, the second-level window length is 5-15 seconds, the minute-level window is 3-10 minutes, and the hour-level window is 30-60 minutes, and the window overlap rate is 50%, to ensure the continuity and stability of the feature extraction. The sliding window mechanism monitors the instantaneous change rate and cumulative fluctuation amplitude of the AGC instruction sequence in real time, dynamically adjusts the window length, and when it is detected that the instruction fluctuation intensity exceeds the preset threshold (for example, the power change rate is greater than 2% per second), the window length is automatically shortened to enhance the sensitivity to high-frequency disturbances; on the contrary, when the fluctuation is gentle, the window is lengthened to improve the low-frequency trend identification accuracy. The window overlap rate of 50% means that half of the data points are shared between adjacent windows, thereby avoiding feature information loss or mutation caused by window switching. Within each window, the AGC instruction is subjected to Fourier transform and wavelet packet decomposition, and the energy distribution and phase characteristics in different frequency bands are calculated, and then the key characteristic parameters such as the power change rate (unit: MW / s), fluctuation frequency (unit: Hz) and duration (unit: s) at the second time scale are extracted. At the same time, these characteristic parameters are input into a support vector machine-based classifier, which is trained in advance on historical operation data and can determine whether the current instruction belongs to "high-frequency disturbance type", "medium-frequency fluctuation type" or "low-frequency trend type", and output the corresponding instruction type label accordingly. The label serves as an important input variable for the subsequent power allocation optimization algorithm, guiding the collaborative response strategy of thermal power and energy storage systems. In addition, to realize the matching with the climbing ability of the thermal power unit, a dynamic response model of the thermal power unit is built in the system, which is based on the parameters such as boiler heat storage capacity, steam turbine inertia constant and control system delay, and can predict the actual response curve of the unit under different load change rates. The response speed of the energy storage system is determined by the battery charge and discharge conversion efficiency, inverter response time and maximum power output capacity, and its response characteristics are obtained by fitting the measured data. Finally, a double-layer mapping relationship table is established to associate the multi-scale characteristic parameters with the maximum climbing rate of the thermal power unit (unit: MW / min) and the maximum response rate of the energy storage system (unit: MW / s), forming a quantitative mapping rule from the AGC instruction to the system response capability. The mapping rule is called at the beginning of each dispatching period to check whether the current instruction is within the system's response range, and if it exceeds, a warning mechanism is triggered and reported to the dispatching master station.

[0024] In the above-mentioned method for maximizing the benefits of thermal power energy storage frequency modulation auxiliary services, the step (2) is to establish a dynamic cost-benefit coupling function: the coal consumption characteristic curve of the thermal power unit, the charging and discharging cycle loss model of the energy storage system, the real-time auxiliary service market price signal and the frequency modulation performance evaluation rules are fused to construct a dynamic coupling function with the economic value of unit frequency modulation as the target, wherein the thermal regulation cost is quantified according to the differential relationship between load rate and coal consumption, and the energy storage loss cost is nonlinearly mapped according to the state of charge variation and cycle depth. Specifically, the coal consumption characteristic curve of the thermal power unit in step (2) is obtained by fitting the field thermal test data, a continuous mapping relationship between load rate and unit coal consumption is constructed by using cubic spline interpolation method, and the calculation precision of coal consumption differential cost reaches 0.1 g of standard coal per kWh; the cycle loss model of the energy storage system is established based on the exponential attenuation relationship between the battery health state and the number of deep discharges, and the equivalent loss cost increases by 1.8 times for every 10% increase in cycle depth. The process of obtaining the coal consumption characteristic curve of the thermal power unit includes: first, carry out thermal test in the full load range in the power plant, record the unit coal consumption (unit: g / kWh) corresponding to different load rates (0% to 100%), the test data points are not less than 50, covering all typical working conditions; then, the discrete data points are smoothed by using cubic spline interpolation method to generate a continuous and smooth load rate-unit coal consumption curve, which meets the second derivative continuous condition to ensure that the derivative can be accurately calculated at any load point; finally, based on the curve, the thermal regulation cost, i.e. the additional coal consumption increment caused by unit load change, is calculated, and its mathematical expression is:

[0025] wherein, represents the thermal regulation cost, The differential cost varies with the load rate, and is higher in the low load section (e.g., 20%-40%) due to lower combustion efficiency, and tends to be stable in the high load section (e.g., 80%-100%) close to the optimal efficiency point. The energy storage system cycle loss model is established based on the relationship between the state of health (SOH) and the depth of discharge (DOD). SOH is defined as the ratio of the current battery capacity to the initial rated capacity, and its decay rate is exponentially related to DOD, i.e., the higher the DOD, the faster the SOH decreases. Specifically, when the DOD increases from 20% to 30%, the SOH annual decay rate rises from 1.5% to 3.2%; when the DOD increases from 30% to 40%, the SOH annual decay rate jumps to 6.8%. Therefore, the equivalent loss cost does not increase linearly, but non-linearly with the increase of DOD. Assuming the reference DOD is 20%, the equivalent loss cost is 1.0 yuan / charge; when the DOD increases by 10 percentage points, the equivalent loss cost is multiplied by 1.8, i.e., the cost is 1.8 yuan / charge when the DOD is 30%, and 3.24 yuan / charge when the DOD is 40%, and so on. This non-linear mapping relationship is realized by table lookup method, and the system queries the corresponding cost coefficient according to the predicted DOD change before each energy storage action is executed, and includes it in the total cost calculation. In addition, the dynamic cost-benefit coupling function also integrates the real-time ancillary service market price signal, which is issued by the grid dispatching agency, usually in units of megawatt per minute (MW·min), and the price fluctuation range can reach 0.1 yuan to 5.0 yuan. The price signal is accessed in real time through a dedicated communication channel, updated every second. The frequency regulation performance evaluation rules include three dimensions: regulation speed, regulation accuracy and response time, and the weight coefficients are automatically updated according to the latest grid evaluation method. For example, the current evaluation rules in a certain region stipulate that the regulation speed accounts for 40% of the total score, the regulation accuracy accounts for 35%, and the response time accounts for 25%. The system weights the scores of these three indicators and sums them up to get the comprehensive evaluation score, which directly affects the economic benefit of unit frequency regulation action. Finally, the output of the dynamic cost-benefit coupling function is the net economic value of unit frequency regulation action, and its calculation formula is:

[0026] The function is calculated in real time in each dispatching period as the objective function of the subsequent optimization algorithm.

[0027] In the above-mentioned method for maximizing the benefits of thermal power energy storage frequency modulation auxiliary services, the step (3) of designing an adaptive power allocation optimization algorithm: taking a dynamic cost-benefit coupling function as the optimization objective, combining the current state of charge of the energy storage system, the operating constraints of the thermal power unit and the multi-scale characteristics of the frequency modulation instruction, using a rolling time domain optimization framework, the optimal power allocation ratio of thermal power and energy storage is solved in each scheduling period, to ensure that the system maximizes economic benefits under the premise of meeting the frequency modulation performance index. Specifically, the scheduling period of the rolling time domain optimization in the step (3) is 2 seconds, the prediction time domain is 60 seconds, the optimization variables include the thermal power increment and the energy storage power The constraint conditions cover the minimum technical output of the thermal power unit, the maximum climbing rate, the upper and lower limits of the state of charge of the energy storage system (set to 20% to 80%) and the charge and discharge power limits, the optimization solution uses the interior point method, and the time consumption of a single calculation is less than 500 milliseconds. The core of the rolling time domain optimization framework is to convert the optimization problem in an infinite time domain into a series of sub-problems in a finite time domain, each of which solves the optimal control sequence in the future T=60 seconds at the current time t. The scheduling period of 2 seconds means that the system performs optimization calculation every 2 seconds to ensure sufficient response speed to the rapidly changing AGC instructions. At the beginning of each scheduling period, the system collects the AGC instructions, the actual output of the thermal power unit, the SOC of the energy storage system, the grid price, the assessment weight and other state information at the current time, and takes these information as the input parameters of the optimization model. The optimization variables are the thermal power increment (unit: MW) and the energy storage power (unit: MW), where represents the output change relative to the previous time, represents the charge and discharge power of the energy storage system, and the positive value represents charging and the negative value represents discharging. The objective function is to maximize the net economic value of unit frequency modulation action accumulated in the future 60 seconds, that is:

[0028] where, is the time step index, and there are 30 steps (2 seconds per step). The constraints include: the minimum technical output of the thermal power unit cannot be less than 15% of its rated capacity, and the maximum ramping rate is 1.5 MW / min; the SOC of the energy storage system must be maintained between 20% and 80% to prevent overcharging and over-discharging; the maximum charging and discharging power of the energy storage system is limited to ±10 MW. In addition, the power balance constraint must be met, that is, the total output change of the thermal power and energy storage must be equal to the change required by the AGC command. The optimization solution adopts the interior point method, which converts the inequality constraints into equality constraints by introducing barrier functions, and iteratively searches for the optimal solution within the feasible region. The interior point method has the advantages of fast convergence speed and high numerical stability, and is suitable for large-scale nonlinear optimization problems. The system deploys a high-performance GPU accelerated computing unit on the edge computing node to ensure that the time consumption of a single optimization calculation is less than 500 milliseconds, meeting the real-time requirements. After optimization, only the first control action (i.e. and ) is executed, and the remaining actions are recalculated in the next scheduling period to form a rolling optimization closed loop.

[0029] In the above method for maximizing the benefits of thermal power and energy storage frequency modulation auxiliary services, step (4) implements frequency modulation action execution and real-time feedback correction: thermal power unit output adjustment instructions and energy storage system charging and discharging power instructions are generated based on the optimization results, actual response deviations, device state parameters, and market settlement data during the execution process are synchronously collected, and the weight coefficients and model parameters in the cost-benefit coupling function are dynamically corrected through an online learning mechanism. Specifically, the actual response deviation in step (4) is defined as the absolute value integral of the difference between the automatic generation control instruction and the actual output of the thermal storage combined system, and the frequency modulation performance evaluation indicators include the regulation rate, regulation accuracy, and response time, the weight coefficients of which are automatically updated according to the latest evaluation rules of the power grid; the online learning mechanism uses the recursive least squares method to update the market sensitivity parameters in the cost-benefit coupling function every 10 scheduling periods. After the optimization results are generated, the system immediately sends output adjustment instructions to the thermal power unit control system, including target output value, regulation rate limit, and allowable error range; at the same time, it sends charging and discharging power instructions to the energy storage energy management system, including power size, duration, and SOC target value. Both systems have independent execution feedback loops and upload execution state data in real time. The system synchronously collects the actual output curve of the thermal storage combined system and compares it with the AGC instruction to calculate the actual response deviation. The deviation is defined as the absolute value of the difference between the two within a specified time period, for example, in the last 10 seconds:

[0030] The bias value directly reflects the ability of the system to track instructions, and is a key indicator for evaluating frequency modulation performance. In addition, the system also collects state parameters of the thermal power unit, such as steam pressure, temperature, and rotating speed, as well as parameters of the energy storage system, such as voltage, current, temperature, and SOC, for monitoring the running state of the equipment. The market settlement data come from the settlement system of the power grid dispatching institution, and include the actual settlement price, assessment score and reward amount of each frequency modulation action. These data are stored in a local database for subsequent analysis. The online learning mechanism uses the recursive least squares (RLS) method, which continuously updates the parameter estimation value to adapt to the dynamic changes of the system. Specifically, the system regards the market sensitivity parameters (such as the influence coefficient of electricity price on revenue) as the parameters to be estimated, uses the market data and actual revenue data of the past 10 dispatching periods to build a regression model, and iteratively solves the optimal parameter estimation value by the RLS algorithm. It is updated once every 10 dispatching periods to ensure that the model parameters always reflect the latest market environment. The updated parameters are fed back to the dynamic cost-benefit coupling function to realize adaptive correction of the model.

[0031] In the above-mentioned method for maximizing the benefits of thermal power energy storage frequency modulation auxiliary services, the step (5) of generating an auxiliary service bidding strategy and a benefit prediction report: based on historical optimization results and market clearing rules, using nonlinear regression and scenario simulation methods, predicting the expected revenue and risk level under different bidding capacity and price combinations, outputting the optimal bidding strategy and a 24-hour benefit prediction report for the next day, which is used to guide the market participation decision-making of the next day. Specifically, the nonlinear regression model in step (5) uses a Gaussian process regression algorithm, the input features include historical winning prices, frequency modulation capacity demand, new energy output fluctuation rate and system reserve margin, and the output is the expected unit capacity revenue; scenario simulation sets not less than 1000 market scenarios, covering a price fluctuation range of 0.5 times to 2.0 times of the current average price, and the benefit prediction report includes expected value, standard deviation and 95% confidence interval. Before the daily operation ends, the system starts the bidding strategy generation module. This module first retrieves the historical optimization results of the past 7 days, including the bidding capacity, bidding price, actual winning situation, settlement price and revenue data of each frequency modulation action. At the same time, it obtains the frequency modulation capacity demand prediction, new energy output prediction and system reserve margin prediction released by the power grid for the next day. These data are input into the Gaussian process regression model as input features. The model assumes that the output variable (expected unit capacity revenue) follows a Gaussian distribution, and the mean and covariance are determined by the kernel function. The kernel function selected is the square exponential kernel, which has the form:

[0032] where, is the signal variance, are hyperparameters determined by maximum likelihood estimation. The model outputs the expected unit capacity revenue and its uncertainty (variance) under given input features. Based on this, the system performs scenario simulation, randomly generating no less than 1000 market scenarios, each containing different price fluctuation factors (0.5 to 2.0 times), capacity demand fluctuation factors (0.8 to 1.2 times), and new energy output fluctuation factors (0.7 to 1.3 times). For each scenario, the system simulates the bidding process and calculates the expected revenue and risk under different bidding capacity and price combinations. Finally, the system outputs a complete benefit prediction report, which includes: the frequency modulation revenue expectation value (unit: ten thousand yuan) of the next 24 hours, the revenue standard deviation (unit: ten thousand yuan), the 95% confidence interval (unit: ten thousand yuan), and the recommended optimal bidding capacity (unit: MW) and bidding price (unit: yuan / MW·min). The report also contains risk level labels, such as "high risk", "medium risk", "low risk", to help decision-makers assess market participation strategies.

[0033] In the above-mentioned method for maximizing the benefit of thermal power energy storage frequency modulation auxiliary services, a dynamic adjustment mechanism for the state of charge safety boundary of the energy storage system is also included: based on the future 2-hour load forecast and frequency modulation instruction intensity forecast, the lower and upper limits of the state of charge are adjusted in real time, and when the predicted frequency modulation demand is intense, the state of charge operating range is expanded to 15% to 85% to improve the system regulation margin, while increasing the cycle loss cost weight to prevent overuse. This mechanism is realized by integrating a load forecasting model and a frequency modulation instruction forecasting model. The load forecasting model is based on historical load data, weather forecasts, holiday information, etc. as input, and uses an LSTM neural network for short-term load forecasting, with a prediction accuracy of over 95%. The frequency modulation instruction forecasting model is based on historical AGC instruction sequences, using an ARIMA model for trend prediction, and combining new energy output prediction results to estimate the possible frequency modulation demand intensity in the next 2 hours. When the prediction result shows that the frequency modulation demand intensity exceeds the threshold (for example, the average power change rate is greater than 1.5% per minute), the system automatically triggers the safety boundary adjustment logic, expanding the operating range of the energy storage system SOC from the default 20% to 80% to 15% to 85%. This adjustment aims to improve the system's regulation capacity during peak periods and avoid the inability to respond to instructions due to insufficient SOC. However, to prevent overuse of the energy storage system, the system also increases the cycle loss cost weight, i.e. in the dynamic cost-benefit coupling function, the weight coefficient of the energy storage loss cost is increased from the default 0.6 to 0.8. This makes the use of the energy storage system more costly under the same conditions, thereby inhibiting its frequent charging and discharging in unnecessary situations and protecting the battery life.

[0034] In the above-mentioned method for maximizing the benefits of thermal power energy storage frequency modulation auxiliary services, a collaborative aging evaluation module of the thermal power unit and the energy storage system is further included: based on the cumulative adjustment times, the deep adjustment time length and the temperature stress data, the equivalent aging factors of the thermal power boiler and the energy storage battery are calculated respectively, when the aging factor of any device exceeds the preset threshold (0.85 for thermal power and 0.90 for energy storage), the priority of the device in power distribution is automatically reduced, and the device maintenance suggestion is marked in the benefit prediction report. The module is divided into two sub-modules: a thermal power aging evaluation submodule and an energy storage aging evaluation submodule. The cumulative adjustment times (unit: times) of the thermal power unit, the deep adjustment time length (unit: hours) and the temperature stress data (unit: MPa) of the key parts of the boiler are collected. The cumulative adjustment times refer to the total number of times the thermal power unit participates in frequency modulation in the past 30 days; the deep adjustment time length refers to the total duration of the load change amplitude exceeding 10%; the temperature stress data are collected in real time by the strain gauge installed on the boiler pipe wall. Based on these data, the system calculates the equivalent aging factor of the thermal power boiler, and the calculation formula is:

[0035] wherein, is the cumulative adjustment times, is the maximum allowed adjustment times within the design life; is the deep adjustment time length, is the maximum allowed deep adjustment time length within the design life; is the current temperature stress, is the material fatigue limit; are weight coefficients, which are set to 0.4, 0.3 and 0.3 respectively. When > 0.85, the system automatically reduces the priority of the thermal power unit in power distribution, that is, increases the adjustment cost weight of the thermal power unit in the optimization algorithm, so that it is less called under the same conditions. The energy storage aging evaluation submodule calculates the equivalent aging factor of the energy storage battery based on the cumulative charge and discharge times of the energy storage system, the deep discharge times and the battery temperature data. The deep discharge times refer to the complete cycle times of SOC from 80% to 20%. The equivalent aging factor calculation formula is:

[0036] wherein, is the cumulative charge and discharge times, is the maximum allowed cycle times within the design life; is the deep discharge times, is the maximum allowed deep discharge times within the design life; is the maximum battery temperature, is the upper limit of the safety temperature;​​ , , are weight coefficients, respectively set as 0.5, 0.3, 0.2. When >0.90, the system also reduces the priority of the energy storage system in power distribution, and adds the prompt of "suggested to arrange battery maintenance" in the benefit prediction report.

[0037] In the above-mentioned method for maximizing the benefit of thermal power energy storage frequency modulation auxiliary service, the method is deployed on the edge computing node of the thermal storage joint control system, has bidirectional communication capability with the power grid dispatching master station, the power plant monitoring system and the energy storage energy management system, the communication protocol adopts the general standard of the power system, the data refresh frequency is not less than 1 times per second, and the overall response delay of the system is less than 1 second. The edge computing node is deployed in the power plant control room, is equipped with a high-performance industrial computer, is loaded with a multi-core CPU and a GPU, has a memory capacity of not less than 64 GB, and has a storage space of not less than 1 TB. The system is connected with the power grid dispatching master station through an optical fiber Ethernet, communicates by using an IEC 61850 standard, supports GOOSE, MMS and other protocols, and ensures the reliability and real-time performance of instruction transmission. The system is connected with the power plant monitoring system (DCS) through an OPC UA interface, obtains thermal power unit operation parameters in real time. The system is connected with the energy storage energy management system (EMS) through a Modbus TCP protocol, receives energy storage system state information and issues control instructions. All communication links are configured with redundant backup to ensure normal operation in the case of single point failure. The data refresh frequency is not less than 1 times per second, which means that the system completes a complete data acquisition, processing and instruction issuing process at least once per second. The overall response delay of the system is less than 1 second, that is, the time interval from receiving the AGC instruction to issuing the control instruction is not more than 1 second, which meets the requirement of the power grid on the response speed of frequency modulation.

[0038] In the above-mentioned method for maximizing the benefits of thermal power energy storage frequency modulation auxiliary services, a natural resource asset optimization allocation knowledge base is also established, which stores no less than 100,000 historical allocation cases, policy and regulation provisions, and expert rules, and provides prior knowledge support for the initial setting of multi-objective weights through a semantic retrieval engine. The knowledge base adopts a distributed database architecture, and the storage structure includes a case library, a regulation library, and a rule library. The case library stores successful cases of various natural resource allocation projects in the past five years nationwide, and each case includes project background, resource allocation scheme, implementation effect, economic benefit, and social benefit fields. The regulation library includes national and local laws and regulations, policy documents, and technical standards related to natural resource management. The rule library is written by domain experts and contains hundreds of experiential rules, such as "in ecologically sensitive areas, ecological functions should be prioritized," "in economic development zones, resource utilization efficiency should be improved," etc. The system has a built-in semantic retrieval engine based on natural language processing technology, which can understand the user's query intent and retrieve relevant cases and rules from the knowledge base. In the initialization stage of the multi-objective optimization algorithm, the system automatically calls the engine, retrieves the most relevant 10 historical cases and 5 expert rules based on the current project background and target requirements, extracts the initial setting suggestions of multi-objective weights, and serves as the starting point of the optimization algorithm, significantly improving optimization efficiency and result rationality.

[0039] In the above-mentioned method for maximizing the benefits of thermal power energy storage frequency modulation auxiliary services, it also includes interfacing with the national spatial planning "one map" system, obtaining planning control boundary data in real time through standard geographic information service interface, ensuring that the resource allocation scheme is strictly consistent with the upper planning, and the spatial consistency verification error is less than 1 pixel. The system connects with the "one map" system through Web GIS service interface, adopts WMS (Web Map Service) and WFS (Web Feature Service) standard protocols. WMS is used to obtain planning area map images and base maps, and WFS is used to obtain vectorized planning control boundary data such as ecological protection red line, permanent basic farmland, and urban development boundary. After generating the resource allocation scheme, the system automatically calls the spatial consistency verification module to superimpose and analyze the resource distribution map in the scheme and the control boundary map in the "one map". The verification process includes: first, coordinate system and geometric registration of the two maps to ensure spatial consistency; then, calculate the distance between each resource unit in the scheme and the control boundary, if the distance is less than the set threshold (such as 1 meter), it is judged as conflict; finally, count the number of conflicts and area proportion. The verification error is less than 1 pixel, that is, at the map display resolution, any spatial misplacement does not exceed one pixel point, ensuring the accuracy and compliance of the scheme.

[0040] Example 2 On the basis of the above-mentioned power plant energy storage frequency modulation auxiliary service benefit maximization method, the embodiment provides an alternative technical solution, the core of which is to introduce a reinforcement learning agent to dynamically adjust the weight coefficients in multi-objective optimization, so as to realize a more intelligent decision-making process. The scheme is improved in step (3), as follows: in the step (3), when designing the adaptive power allocation optimization algorithm, instead of using fixed multi-objective weights, a reinforcement learning agent is introduced, which learns the optimal weight adjustment strategy by interacting with the environment. The reinforcement learning agent uses the deep deterministic policy gradient (DDPG) algorithm, its state space contains the current resource configuration bias index and historical adjustment records, the action space is the fine-tuning amount of the multi-objective weight vector, and the reward function is the product of the bias index decline amplitude and the scheme stability. Specifically, the resource configuration bias index is calculated by weighting the spatial mismatch rate, the target deviation degree and the execution lag coefficient, and the weights are 0.4, 0.4 and 0.2 respectively, and the bias index threshold is set to 0.15. The index is used to measure the gap between the current resource configuration scheme and the ideal target. The historical adjustment records in the state space include the action sequence of the last 10 weight adjustments, which is used to capture long-term trends. The action space is defined as the fine-tuning amount of the ecological safety, economic efficiency and social fairness three target weights, and the range of each fine-tuning amount is [-0.1, 0.1]. The reward function is designed as:

[0041] The first term is the decline amplitude of the bias index, and the second term is the scheme stability factor, which aims to encourage smooth adjustment rather than drastic fluctuations. The reinforcement learning agent updates the network and learns the optimal strategy gradually according to the current state, the action taken and the reward obtained after each optimization period. The advantage of this scheme is that it can adapt to complex and variable external environment, and realize dynamic optimization of weight without human intervention, which improves the autonomous decision-making ability of the system.

[0042] The basic principles and main features of the present application are shown and described, and the advantages of the present application are shown and described. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for maximizing the benefit of frequency modulation auxiliary service of thermal energy storage, characterized in that: The method comprises the following steps: Step (1), based on the historical data and real-time data of the power grid automatic generation control instruction, a sliding window mechanism is used to perform multi-scale decomposition of the instruction at the second, minute and hour levels, power change rate, fluctuation frequency and duration characteristics at each time scale are extracted, an instruction mapping relationship matched with the climbing ability of thermal power generating units and the response speed of energy storage systems is established, and a multi-time scale frequency modulation instruction analysis model is constructed; Step (2), the coal consumption characteristic curve of the thermal power generating unit, the energy storage system charge and discharge cycle loss model, the real-time auxiliary service market price signal and the frequency modulation performance evaluation rule are fused to construct a dynamic coupling function with the economic value of unit frequency modulation action as the target, the thermal regulation cost is quantified according to the differential relationship between the load rate and the coal consumption, the energy storage loss cost is nonlinearly mapped according to the state of charge variation and the cycle depth, and a dynamic cost-benefit coupling function is established; Step (3), taking the dynamic cost-benefit coupling function as the optimization target, combining the current state of charge of the energy storage system, the operation constraints of the thermal power generating unit and the multi-scale characteristics of the frequency modulation instruction, a rolling time domain optimization framework is used to solve the optimal power distribution ratio of the thermal power and the energy storage in each scheduling period, so that the system realizes the maximization of economic benefits under the premise of meeting the frequency modulation performance index, and an adaptive power distribution optimization algorithm is designed; Step (4), generating the thermal power generating unit output adjustment instruction and the energy storage system charge and discharge power instruction according to the optimization result, synchronously collecting the actual response deviation, equipment state parameters and market settlement data in the execution process, dynamically correcting the weight coefficients and model parameters in the cost-benefit coupling function through the online learning mechanism, and implementing the frequency modulation action execution and real-time feedback correction; Step (5), based on the historical optimization results and the market clearing rule, a nonlinear regression and scenario simulation method is used to predict the expected benefits and risk levels under different bidding capacities and price combinations, and an optimal bidding strategy and a 24-hour benefit prediction report are output, which are used to guide the next day market participation decision, and an auxiliary service bidding strategy and benefit prediction report are generated.

2. The method of claim 1, wherein the method maximizes the benefit of the frequency regulation auxiliary service of the thermal energy storage. In step (1), the time length of the sliding window is dynamically adjusted according to the fluctuation intensity of the frequency modulation instruction, the second-level window length is 5-15 seconds, the minute-level window is 3-10 minutes, and the hour-level window is 30-60 minutes, and the window overlap rate is 50%; in each window, Fourier transform and wavelet packet decomposition are performed on the instruction to extract the power change rate, fluctuation frequency and duration characteristics, and a support vector machine classifier is used to judge the instruction type, and output "high-frequency disturbance type", "medium-frequency fluctuation type" or "low-frequency trend type" label as the input variable of the adaptive power distribution optimization algorithm.

3. The method of claim 1, wherein the method further comprises: In step (2), the coal consumption characteristic curve of the thermal power generating unit is obtained by fitting the field thermal test data, a continuous mapping relationship between the load rate and the unit coal consumption is constructed by using the cubic spline interpolation method, and the coal consumption differential cost calculation precision reaches 0.1 grams of standard coal per kilowatt-hour; The cycle loss model of the energy storage system is based on the exponential decay relationship between the battery state of health and the number of deep discharges. The equivalent loss cost increases by 1.8 times for every 10% increase in cycle depth. The cost corresponding to the reference cycle depth of 20% is 1.0 yuan / time. Nonlinear mapping is achieved through table lookup.

4. The method of claim 1, wherein the method further comprises: The scheduling period of the step (3) is 2 seconds, the prediction time domain is 60 seconds, and the optimization variables include the power increment of the thermal power The energy storage power The constraint conditions include that the minimum technical output of the thermal power unit is not less than 15% of the rated capacity, the maximum climbing rate is 1.5 MW / min, the state of charge of the energy storage system is maintained between 20% and 80%, and the charging and discharging power limit is ±10 MW. The optimization solution adopts an interior point method, the time consumption of a single calculation is less than 500 milliseconds, only the current time control action is executed, and the remaining actions are recalculated in the next period.

5. The method of claim 1, wherein the method further comprises: The actual response deviation in step (4) is defined as the integral of the absolute value of the difference between the automatic generation control instruction and the actual output of the fire storage combined system within 10 seconds; the frequency regulation performance evaluation index includes the regulation speed, the regulation accuracy and the response time, and the weight coefficients are automatically updated according to the latest evaluation rules of the power grid; The online learning mechanism uses the recursive least squares method to update the market sensitivity parameters in the cost-benefit coupling function every 10 scheduling periods.

6. The method of claim 1, wherein the method further comprises: The nonlinear regression and scenario simulation method in step (5) uses the Gaussian process regression algorithm. The input features include historical winning prices, frequency modulation capacity demand, new energy output fluctuation rate and system backup margin. The output is the expected unit capacity benefit. The scenario simulation sets no less than 1000 market scenarios, covering the price fluctuation range of 0.5 times to 2.0 times the current average price. The benefit prediction report includes the expected value, standard deviation and 95% confidence interval of the benefit, and marks the risk level.

7. The method of claim 1, wherein the method further comprises: It also includes a dynamic adjustment mechanism for the state of charge safety boundary of the energy storage system. According to the future 2-hour load forecast and the frequency modulation instruction intensity forecast, when the predicted frequency modulation demand is intense, the state of charge operating interval is expanded to 15% to 85%, and at the same time the energy storage loss cost weight coefficient in the dynamic cost-benefit coupling function is increased from 0.6 to 0.8 to suppress unnecessary frequent charging and discharging.

8. The method of claim 1, wherein the method further comprises: It also includes, The edge computing node deployed in the fire storage combined control system has bidirectional communication capability with the power grid dispatching master station, power plant monitoring system and energy storage energy management system. The communication protocol uses IEC 61850, OPC UA and Modbus TCP standards. The data refresh frequency is not less than 1 times per second. The overall response delay of the system is less than 1 second. The edge node is equipped with multi-core CPU and GPU. The memory capacity is not less than 64 GB.

9. The method of claim 1, wherein the method further comprises: In step (3), a reinforcement learning agent is introduced to dynamically adjust the multi-objective weight coefficients. The agent uses the deep deterministic policy gradient algorithm. The state space includes the resource allocation deviation index and the historical adjustment record. The action space is the multi-objective weight vector fine tuning amount, ranging from -0.1 to 0.

1. The reward function is the product of the deviation index decline amplitude and the scheme stability. The deviation index is calculated by weighting the space mismatch rate, the target deviation degree and the execution lag coefficient, with weights of 0.4, 0.4 and 0.2 respectively. The threshold is set to 0.15.

Images