A dynamic peak shaving method for a thermal power plant using high-temperature thermal energy storage

Abstract

The application provides a dynamic peak regulation method of a thermal power plant using high-temperature heat storage energy, and belongs to the technical field of dynamic peak regulation of thermal power plants. The application collects power grid load prediction data, heat storage tank temperature and pressure state data and steam turbine operation parameter data in real time, calculates the current heat storage amount and the residual heat storage capacity of the heat storage tank through an intelligent peak regulation decision maker, determines the optimal steam extraction ratio and heat storage release strategy based on a dynamic peak regulation game model, extracts the steam at the intermediate stage of the steam turbine to the heat storage tank for heat storage according to the optimal steam extraction ratio during the low load period of the power grid, starts the heat release system of the heat storage tank to increase the power generation power during the peak load period of the power grid to realize power compensation, adjusts the optimal steam extraction ratio in real time through the intelligent peak regulation decision maker, establishes a heat storage efficiency monitoring system to monitor the operation state of the heat storage tank and optimize the operation parameters in real time, and solves the technical problem that the optimal steam extraction ratio cannot be determined in real time.

Description

Technical Field

[0001] This invention belongs to the field of dynamic peak shaving technology for thermal power plants, and specifically relates to a dynamic peak shaving method for thermal power plants that utilizes high-temperature thermal energy storage. Background Technology

[0002] Thermal power plants use coal combustion to produce flue gas, which is then used to heat water to generate steam. This steam drives a turbine to generate electricity—this is the basic principle of traditional thermal power plants. To address peak-shaving demands caused by grid load fluctuations, current technology extracts a portion of the steam into a thermal storage tank during off-peak hours for heat storage. During peak hours, the heat storage tank releases heat to heat water, generating steam to drive the turbine. The steam does not require purification before entering the storage tank, thus saving purification costs. In actual operation, the proportion of steam extracted during off-peak hours is first limited by a maximum limit to avoid affecting the normal rotation of the turbine. Secondly, it can be adjusted in real time according to grid load conditions. For example, when the grid load decreases further during off-peak hours, the steam extraction ratio can be appropriately increased; when the grid load recovers, the extraction ratio can be decreased, but neither can exceed the maximum safe ratio. However, in existing technologies, the determination of the steam extraction ratio mainly relies on the operator's experience and pre-set fixed operating modes. This lack of intelligent, dynamic optimization based on grid load forecasts, real-time temperature and pressure conditions of the thermal storage tank, and turbine operating parameters prevents the maximization of thermal storage benefits under complex and variable grid load conditions. When grid load fluctuates frequently, a fixed extraction ratio strategy often fails to fully utilize the thermal storage capacity of the tank and cannot guarantee the optimal operating state of the turbine under various conditions. In other words, existing technologies suffer from the inability to determine the optimal steam extraction ratio in real time. Summary of the Invention

[0003] In view of this, the present invention provides a dynamic peak-shaving method for thermal power plants that utilizes high-temperature thermal energy storage, which can solve the technical problem in the prior art that the optimal steam extraction ratio cannot be determined in real time.

[0004] This invention is implemented as follows: It provides a dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage. The method includes: real-time acquisition of grid load forecast data, thermal storage tank temperature and pressure status data, and turbine operating parameter data; and calculation of the current heat storage capacity and remaining heat storage capacity of the thermal storage tank using an intelligent peak-shaving decision-maker. Based on the grid load forecast data and the thermal storage tank temperature and pressure status data, an optimal steam extraction ratio and thermal energy release strategy are determined through a dynamic peak-shaving game model. The dynamic peak-shaving game model includes an upper-level model aiming to maximize grid peak-shaving benefits and a lower-level model aiming to maximize turbine operating safety. During periods of low grid load, intermediate-stage steam from the turbine is extracted to the thermal storage tank according to the optimal steam extraction ratio. The extracted steam transfers heat to the high-temperature thermal storage medium through a heat exchanger, and the condensate is returned to the turbine circulation system. During periods of high grid load, the thermal storage tank heat release system is activated. The high-temperature thermal storage medium heats the feedwater through a heat exchanger to generate steam, which drives the turbine impeller to increase power generation, thus achieving power compensation during peak grid load periods.

[0005] Specifically, the steps of the dynamic peak-shaving game model are as follows: the objective function of the upper-level model is the product of the power compensation amount during the peak period of the power grid and the current heat storage capacity of the thermal storage tank, plus the inverse term of the operating cost minus the logarithm of the power grid load demand, plus the sine function term of the optimal steam extraction ratio. The constraints include thermal storage tank capacity constraints, power grid load balance constraints, and steam extraction ratio range constraints. The objective function of the lower-level model is the product of the turbine operating stability index and the steam temperature, plus the inverse term of the turbine vibration power minus the cosine function term of the blade angle, plus the exponential function term of the optimal steam extraction ratio. The constraints include turbine operating parameter safety range constraints, steam temperature and pressure constraints, and blade stress constraints.

[0006] The coupling term of the two objective functions is the optimal steam extraction ratio term, which represents the combined impact of the steam extraction ratio on the grid peak-shaving benefits and the turbine operation safety.

[0007] The intelligent peak-shaving decision-maker is specifically based on a deep network architecture of sequence-to-sequence model, including an input encoding layer, a multi-layer self-attention mechanism, a feedforward network layer, and an output decoding layer. The self-attention mechanism is used to process the correlation between power grid load forecast data and thermal storage tank temperature and pressure status data. The feedforward network layer is used to extract nonlinear feature patterns. The output decoding layer generates the current heat storage capacity, remaining heat storage capacity, and optimal steam extraction ratio of the thermal storage tank.

[0008] The intelligent peak shaving decision-maker adopts a sliding time window strategy to dynamically adjust the upper limit of the extraction ratio based on the power grid load forecast data. The sliding time window strategy is a data processing method that uses a fixed-length time window to analyze historical data and updates the window content over time.

[0009] Specifically, the training dataset for the intelligent peak-shaving decision-maker is established by collecting historical grid load data, thermal storage tank operating status data, turbine operating parameter data, and corresponding label data of current thermal storage capacity, remaining thermal storage capacity, and optimal steam extraction ratio of the thermal storage tank. The collected data is preprocessed, including data cleaning, normalization, and time-series alignment. The preprocessed data is then segmented according to time windows to form training samples.

[0010] Specifically, the training of the intelligent peak shaving decision-maker involves using training set data to conduct supervised learning training on the model, employing the mean squared error loss function and an adaptive moment estimation optimizer for parameter updates, setting the learning rate to 0.001, the batch size to 32, and the number of training rounds to 100. During the training process, a validation set is used to evaluate model performance and an early stopping mechanism is implemented to prevent overfitting.

[0011] The memory length parameter in the intelligent peak shaving decision-maker is determined based on the grid load prediction time window length, the thermal capacity of the thermal storage tank, and the turbine response time. When the grid load prediction time window length is 4 hours, the thermal capacity of the thermal storage tank is 500MWh, and the turbine response time is 15 minutes, the memory length parameter is set to 16 time segments.

[0012] Specifically, the extraction proportional control function is calculated based on the grid load fluctuation amplitude, the thermal storage tank temperature change rate, and the turbine power change rate to obtain the control intensity value. When the control intensity value is in the range of 0 to 0.3, a short memory length of 8 time segments is used; when the control intensity value is in the range of 0.3 to 0.7, a medium memory length of 16 time segments is used; and when the control intensity value is in the range of 0.7 to 1.0, a long memory length of 32 time segments is used.

[0013] Wherein, the memory length parameter is the time series length considered by the intelligent peak shaving decision-maker when processing historical data; the adjustment intensity value is the value used to adjust the memory length parameter calculated by the extraction proportional adjustment function; the grid load fluctuation amplitude is the degree of change of grid load within a certain period of time; the thermal storage tank temperature change rate is the rate of change of temperature inside the thermal storage tank over time; the turbine power change rate is the rate of change of turbine output power over time; and the time segment is the basic time unit for data processing by the intelligent peak shaving decision-maker.

[0014] Furthermore, after adjusting the optimal steam extraction ratio in real time through the intelligent peak-shaving decision-maker, the system also includes establishing a thermal storage efficiency monitoring system to monitor the temperature distribution, pressure changes, and heat loss of the thermal storage tank in real time, and to optimize the circulation flow rate of the high-temperature thermal storage medium and the operating parameters of the heat exchanger based on the monitoring results.

[0015] The high-temperature thermal storage medium is molten salt or ceramic granules, used for storing and releasing thermal energy. The current heat storage capacity of the thermal storage tank is the total amount of thermal energy stored within it, calculated by an intelligent peak-shaving decision-maker. The remaining thermal storage capacity is the capacity of the thermal storage tank to continue storing thermal energy, also calculated by the intelligent peak-shaving decision-maker. The grid load forecast data is the predicted information of future grid load demand, sourced from the grid dispatch system. The thermal storage tank temperature and pressure status data is real-time monitoring data of temperature distribution and pressure changes within the thermal storage tank, sourced from temperature and pressure monitoring sensors. The turbine operating parameter data includes turbine speed, power output, vibration parameters, and blade angle operating status information. The optimal steam extraction ratio is the ratio of steam extraction to total steam volume calculated using a dynamic peak-shaving game model. The thermal energy release strategy is the control scheme for releasing stored thermal energy from the thermal storage tank during peak grid load periods. The upper limit of the extraction ratio is the maximum allowable steam extraction ratio under conditions ensuring safe turbine operation. The off-peak period of the power grid load is a period of time when the demand for power grid load is relatively low; the peak period of the power grid load is a period of time when the demand for power grid load is relatively high; the power compensation is the increased power generation capacity through the release of heat energy by the thermal storage system.

[0016] This invention establishes a dynamic peak-shaving game model and an intelligent peak-shaving decision-maker, combining grid load forecast data and thermal storage tank temperature and pressure status data to achieve real-time optimization of the steam extraction ratio, solving the problem of fixed extraction ratios in traditional technologies. This invention employs a two-layer game model structure: the upper-layer model aims to maximize grid peak-shaving benefits, while the lower-layer model aims to maximize turbine operational safety. The optimal steam extraction ratio is determined through the coupled calculation of these two objective functions, ensuring both turbine safety and maximizing thermal storage benefits. Furthermore, this invention utilizes a sliding time window strategy in the intelligent peak-shaving decision-maker to dynamically adjust the upper limit of the extraction ratio based on grid load forecast data, achieving intelligent real-time adjustment of the steam extraction ratio. In summary, this invention solves the technical problem mentioned in the background technology of being unable to determine the optimal steam extraction ratio in real time. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention.

[0018] Figure 2 This is a schematic diagram of the neural network structure of the intelligent peak-shaving decision-maker involved in the present invention.

[0019] Figure 3 This is a schematic diagram of the composition of the high-temperature thermal energy storage system in the thermal power plant in Example 2.

[0020] Figure 4 This is a temperature distribution monitoring diagram inside the thermal storage tank in Example 2.

[0021] Figure 5 This is a 24-hour power regulation operation curve diagram from Example 2.

[0022] Figure 6 This is a monitoring graph of the molten salt temperature change process in Example 2. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0024] like Figure 1 The diagram shown is a flowchart of a dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage, provided by this invention. This method includes the following steps:

[0025] S01. Establish a high-temperature thermal energy storage system for thermal power plants, including a thermal storage tank, a steam extraction pipeline, a heat exchanger, and temperature and pressure monitoring sensors. The thermal storage tank is filled with a high-temperature thermal storage medium, and the steam extraction pipeline connects the intermediate stage steam extraction port of the steam turbine to the inlet of the thermal storage tank.

[0026] S02. Real-time acquisition of power grid load forecast data, thermal storage tank temperature and pressure status data, and turbine operating parameter data, and calculation of the current heat storage capacity and remaining heat storage capacity of the thermal storage tank through the intelligent peak shaving decision-maker;

[0027] S03. Based on the power grid load forecast data and the temperature and pressure status data of the thermal storage tank, the optimal steam extraction ratio and thermal storage release strategy are determined through a dynamic peak shaving game model. The dynamic peak shaving game model includes an upper-level model aimed at maximizing the peak shaving benefits of the power grid and a lower-level model aimed at maximizing the operating safety of the steam turbine.

[0028] S04. During periods of low grid load, the intermediate stage steam of the steam turbine is extracted to the heat storage tank according to the optimal steam extraction ratio. The extracted steam transfers heat to the high-temperature heat storage medium through a heat exchanger, and the condensate is returned to the steam turbine circulation system.

[0029] S05. During peak grid load periods, the heat storage tank heat release system is activated. The high-temperature heat storage medium heats the feedwater through a heat exchanger to generate steam. The steam drives the turbine impeller to increase power generation, thereby achieving power compensation during peak grid load periods.

[0030] S06. The optimal steam extraction ratio is adjusted in real time by an intelligent peak shaving decision-maker. The intelligent peak shaving decision-maker adopts a sliding time window strategy and dynamically adjusts the upper limit of the extraction ratio according to the power grid load forecast data.

[0031] S07. Establish a thermal storage efficiency monitoring system to monitor the temperature distribution, pressure changes and heat loss of the thermal storage tank in real time, and optimize the circulation flow rate of the high-temperature thermal storage medium and the operating parameters of the heat exchanger based on the monitoring results.

[0032] The existing technology involves coal-fired power plants burning coal to produce flue gas, which is then used to heat water to generate steam. This steam drives a turbine to generate electricity. During periods of low grid load, a portion of the steam is extracted into a thermal storage tank for heat storage. During periods of high grid load, the thermal storage tank releases heat to heat water and generate steam to drive the turbine. The steam does not require purification before entering the thermal storage tank, thus saving purification costs. The steam extraction ratio needs to be adjusted in real time according to the grid load while ensuring the safe operation of the turbine.

[0033] The dynamic peak-shaving game model includes an upper-level model aimed at maximizing the peak-shaving benefits of the power grid and a lower-level model aimed at maximizing the operational safety of the steam turbine. The objective function of the upper-level model is the product of the peak-period power compensation amount of the power grid and the current heat storage capacity of the thermal storage tank, plus the inverse term of the operating cost, minus the logarithmic term of the power grid load demand, plus the sine function term of the optimal steam extraction ratio. The peak-period power compensation amount of the power grid comes from the power grid load forecast data, the current heat storage capacity of the thermal storage tank comes from the intelligent peak-shaving decision-maker, the operating cost comes from the equipment operating parameters, the power grid load demand comes from the power grid load forecast data, and the optimal steam extraction ratio is used for calculation in the lower-level model. The constraints include thermal storage tank capacity constraints, power grid load balance constraints, and steam extraction ratio range constraints. The objective function of the lower-level model is the product of the turbine operating stability index and the steam temperature, plus the reciprocal of the turbine vibration power term minus the blade angle cosine function term, plus the optimal steam extraction ratio exponential function term. The turbine operating stability index is derived from turbine operating parameter data, the steam temperature from temperature and pressure monitoring sensors, the turbine vibration power from turbine operating parameter data, and the blade angle from turbine operating parameter data. The optimal steam extraction ratio is used for calculation in the upper-level model. Constraints include the safe range constraint of turbine operating parameters, steam temperature and pressure constraints, and blade stress constraints. The coupling term of the two objective functions is the optimal steam extraction ratio term, which represents the combined impact of the steam extraction ratio on grid peak-shaving benefits and turbine operating safety.

[0034] The intelligent peak-shaving decision-maker is structured as a deep network architecture based on a sequence-to-sequence model, comprising an input encoding layer, a multi-layer self-attention mechanism, a feedforward network layer, and an output decoding layer. The self-attention mechanism handles the correlation between grid load forecast data and thermal storage tank temperature and pressure status data. The feedforward network layer extracts nonlinear feature patterns. The output decoding layer generates the current heat storage capacity, remaining heat storage capacity, and optimal steam extraction ratio of the thermal storage tank. The intelligent peak-shaving decision-maker employs a sliding time window strategy to dynamically adjust the upper limit of the extraction ratio based on the grid load forecast data. The steps for establishing the training dataset for the intelligent peak-shaving decision-maker specifically include collecting historical grid load data, thermal storage tank operating status data, turbine operating parameter data, and corresponding data on the current heat storage capacity, remaining heat storage capacity, and optimal steam extraction ratio of the thermal storage tank. The system extracts proportional label data and preprocesses the collected data, including data cleaning, normalization, and time-series alignment. The preprocessed data is then divided into training samples according to time windows. Each training sample includes an input sequence and a corresponding target output sequence. The training samples are divided into an 80% training set, a 10% validation set, and a 10% test set. The training steps for the intelligent peak-shaving decision-maker specifically include supervised learning training of the model using the training set data, updating parameters using the mean squared error loss function and an adaptive moment estimation optimizer, setting the learning rate to 0.001, the batch size to 32, and the number of training epochs to 100. During training, the validation set is used for model performance evaluation, and an early stopping mechanism is implemented to prevent overfitting. After training, the model's prediction accuracy and generalization ability are evaluated on the test set.

[0035] The memory length parameter in the intelligent peak shaving decision-maker is determined based on the grid load prediction time window length, the thermal capacity of the thermal storage tank, and the turbine response time. When the grid load prediction time window length is 4 hours, the thermal capacity of the thermal storage tank is 500MWh, and the turbine response time is 15 minutes, the memory length parameter is set to 16 time segments.

[0036] Specifically, a decimation proportional control function is designed to adjust the memory length parameter of the intelligent peak shaving decision-maker. The decimation proportional control function calculates the control intensity value based on the grid load fluctuation amplitude, the thermal storage tank temperature change rate, and the turbine power change rate. When the control intensity value is in the range of 0 to 0.3, a short memory length of 8 time segments is used; when the control intensity value is in the range of 0.3 to 0.7, a medium memory length of 16 time segments is used; and when the control intensity value is in the range of 0.7 to 1.0, a long memory length of 32 time segments is used to adjust the memory length parameter of the intelligent peak shaving decision-maker.

[0037] The high-temperature thermal storage medium is molten salt or ceramic granules, used for storing and releasing thermal energy. The current heat storage capacity of the thermal storage tank is the total amount of thermal energy stored within it, calculated by an intelligent peak-shaving decision-maker. The remaining thermal storage capacity is the capacity of the thermal storage tank to continue storing thermal energy, calculated by the intelligent peak-shaving decision-maker. The grid load forecast data is the predicted information on future grid load demand, sourced from the grid dispatch system. The thermal storage tank temperature and pressure status data is real-time monitoring data of temperature distribution and pressure changes within the thermal storage tank, sourced from temperature and pressure monitoring sensors. The turbine operating parameter data includes turbine speed, power output, vibration parameters, and blade angle, among other operating status information. The optimal steam extraction ratio is the ratio of steam extraction to total steam volume calculated using a dynamic peak-shaving game model. The thermal energy release strategy is the control scheme for releasing stored thermal energy from the thermal storage tank during peak grid load periods. The upper limit of the extraction ratio is the maximum allowable steam extraction ratio under conditions ensuring safe turbine operation.

[0038] The sliding time window strategy is a data processing method that analyzes historical data using a fixed-length time window and updates the window content over time. The grid load trough period is a time period with relatively low grid load demand. The grid load peak period is a time period with relatively high grid load demand. The power compensation is the increased power generation through the release of heat energy from the thermal storage system. The memory length parameter is the length of the time series considered by the intelligent peak-shaving decision-maker when processing historical data. The adjustment intensity value is the value used to adjust the memory length parameter, calculated from the proportional regulation function. The grid load fluctuation amplitude is the degree of change in grid load over a certain period. The thermal storage tank temperature change rate is the rate of change of temperature within the thermal storage tank over time. The turbine power change rate is the rate of change of turbine output power over time. The time segment is the basic time unit for data processing by the intelligent peak-shaving decision-maker.

[0039] The specific implementation methods of the above steps are described in detail below.

[0040] The specific implementation of step S01 involves first constructing a thermal storage tank structure. The tank employs a double-layer insulation design, with an inner layer of high-temperature resistant stainless steel and an outer layer of insulation material. The tank capacity is designed to be 500 MWh, with an operating temperature range of 550℃ to 650℃ and an operating pressure range of 1.5 MPa to 2.0 MPa. Next, a steam extraction pipeline system is installed. The extraction pipeline connects to the third-stage extraction port of the intermediate-pressure cylinder of the steam turbine. The pipeline has an inner diameter of 800 mm, uses an insulated design, and is equipped with regulating valves to control the steam flow. The extraction pipeline is designed to withstand a pressure of 3.0 MPa. Following this, a heat exchanger system is installed. The heat exchanger adopts a shell-and-tube structure with a heat exchange area of ​​2000 mm. The design achieves a heat exchange efficiency greater than 85%, and the heat exchanger material is made of high-temperature resistant alloy steel. Finally, a temperature and pressure monitoring sensor network is deployed, with 20 temperature sensors and 10 pressure sensors arranged inside the heat storage tank. The sensor accuracy requirements are a temperature measurement error of less than ±2℃ and a pressure measurement error of less than ±0.05MPa, with a data acquisition frequency of once per second. The high-temperature heat storage medium filled in the heat storage tank is molten salt, a mixture of 60% sodium nitrate and 40% potassium nitrate. Molten salt exhibits good thermal stability and heat transfer performance, a wide operating temperature range, and a heat storage density greater than 1.8 MJ / kg.

[0041] The specific implementation of step S02 involves establishing a data acquisition system. A data acquisition device obtains grid load forecast data from the grid dispatching system, with a forecast time window of the next 4 hours and a data update frequency of once every 15 minutes. The forecast accuracy is required to be less than 5%. Simultaneously, temperature and pressure status data are collected from the thermal storage tank monitoring system, including temperature values ​​at 20 locations, pressure values ​​at 10 locations, and molten salt flow data within the tank, with a data acquisition cycle of 1 second. Additionally, operating parameter data is collected from the turbine control system, including turbine speed, power output, extraction steam pressure and temperature at each stage, vibration parameters, and blade angles, with an acquisition frequency of 10 times per second. The intelligent peak-shaving decision-maker uses a recurrent neural network algorithm to calculate the current heat storage capacity of the thermal storage tank. Input parameters include thermal storage tank temperature distribution data, molten salt flow data, and the temperature difference between the inlet and outlet of the heat exchanger. The current heat storage value is calculated using the heat balance equation. The remaining heat storage capacity is calculated using a capacity estimation algorithm. Based on the density, specific heat capacity, and temperature difference of the heat storage medium, the theoretical maximum heat storage capacity is calculated. The remaining heat storage capacity is obtained by subtracting the current heat storage capacity, with a calculation accuracy requirement of less than 3%.

[0042] The specific implementation of step S03 involves constructing a two-layer game optimization model. The upper-layer model aims to maximize the peak-shaving benefits of the power grid and establishes a multi-objective optimization function, including a peak-period power compensation benefit term, a thermal storage tank heat storage utilization rate term, an operating cost minimization term, and a power grid load balance constraint term. The peak-period power compensation amount is calculated based on the difference between the peak load and the baseline load in the load forecast data, with a compensation power range of 50MW to 200MW. The current heat storage capacity of the thermal storage tank comes from the real-time calculation results of the intelligent peak-shaving decision-maker, with a heat storage range of 0 to 500MWh. Operating costs include equipment depreciation, maintenance, and energy consumption costs, and cost calculation is based on equipment operating hours and power output. Constraints are set as follows: thermal storage tank capacity constraint is that the heat storage capacity does not exceed 95% of the design capacity; power grid load balance constraint requires that the deviation between power generation and load demand be less than 2%; and steam extraction ratio range constraint is 5% to 25%. The lower-layer model aims to maximize the operating safety of the steam turbine and establishes a safety evaluation function, including steam turbine operating stability indicators, steam parameter safety indicators, vibration level control indicators, and blade stress safety indicators. The turbine's operational stability is comprehensively evaluated based on speed fluctuation rate, power output stability, and pressure stability at various levels, with a stability index requirement greater than 0.9. Steam temperature safety constraints are set at 540℃ to 580℃, and steam pressure safety constraints at 1.2MPa to 2.2MPa. Turbine vibration power is controlled below 80% of the design value, and blade stress constraints require stress values ​​to be less than 70% of the material's allowable stress. Two-layer models are coupled through the optimal steam extraction ratio, and an iterative solution algorithm is used to find the optimal solution. The iteration convergence condition is set to the objective function change being less than 0.001.

[0043] The specific implementation of step S04 involves activating the thermal storage mode during periods of low grid load. The criteria for determining low grid load are that the current load is less than 75% of the daily average load and this condition lasts for more than 30 minutes. The opening of the extraction valve is controlled based on the optimal steam extraction ratio calculated using a game theory model. The extraction valve is an electrically controlled regulating valve with a response time of less than 10 seconds and a control accuracy of ±1%. The extracted medium-pressure steam enters the heat exchanger through the extraction pipeline, with the steam flow rate controlled within the range of 50 t / h to 200 t / h according to the extraction ratio. In the heat exchanger, the steam exchanges heat with molten salt, transferring heat to the molten salt and condensing into condensate. During the heat exchange process, the steam inlet temperature is controlled at 560℃ ± 10℃, and the outlet temperature at 180℃ ± 5℃. The molten salt is heated in the heat exchanger, with an inlet temperature of 550℃ and an outlet temperature reaching 620℃. The molten salt flow rate is controlled within the range of 300 t / h to 500 t / h. After collection, the condensate is returned to the turbine feedwater system via a return water pipeline. The return water temperature is 180℃, and the return water flow rate is kept in balance with the extraction steam flow rate. During the thermal storage process, the temperature rise rate of the thermal storage tank is monitored in real time. Under normal thermal storage conditions, the temperature rise rate is 2℃ / min to 5℃ / min. When the stored heat reaches 90% of the design capacity, the extraction ratio is automatically reduced.

[0044] The specific implementation of step S05 involves activating the heat release mode during peak grid load periods. Peak load periods are defined as current load exceeding 120% of the daily average load and a predicted duration exceeding 20 minutes. The heat release system of the thermal storage tank is activated, drawing high-temperature molten salt from the top of the tank via a molten salt circulation pump. The molten salt temperature is between 610°C and 630°C, and the flow rate is controlled within the range of 200 t / h to 400 t / h. The high-temperature molten salt exchanges heat with feedwater in a heat exchanger. The feedwater inlet temperature is 160°C, and after heat exchange, superheated steam at 540°C and a pressure of 1.8 MPa to 2.0 MPa is generated. This superheated steam is then fed into the high-pressure or intermediate-pressure cylinder of the turbine via a steam pipeline, driving the turbine impeller to increase power generation. The power increase is calculated based on the steam flow rate, typically ranging from 80 MW to 150 MW. After releasing heat in the heat exchanger, the molten salt's temperature drops to 580°C and returns to the bottom of the thermal storage tank via a return pipeline, forming a molten salt circulation system. The heat release process is controlled using a proportional-integral-derivative (PID) control algorithm, which dynamically adjusts the molten salt flow rate and steam generation based on grid load demand, with a control response time of less than 30 seconds. The heat release mode automatically stops when the temperature of the thermal storage tank drops to 570℃ or the stored heat capacity falls below 10% of the total capacity, ensuring safe system operation.

[0045] The specific implementation of step S06 involves real-time optimization and adjustment of the steam extraction ratio through an intelligent peak-shaving decision-maker. The decision-maker uses a sliding time window strategy to process historical data, with a time window length of 4 hours and a sliding step of 15 minutes. The upper limit of the extraction ratio is dynamically adjusted based on the volatility of the grid load forecast data: 20% when the load volatility is less than 10%, 15% when the load volatility is between 10% and 20%, and 10% when the load volatility is greater than 20%. The intelligent peak-shaving decision-maker uses a reinforcement learning algorithm for online learning, evaluating the long-term benefits of different extraction ratios through a state-action value function. The learning rate is set to 0.01, and the discount factor is set to 0.95. The decision-maker inputs the current grid load, the status of the thermal storage tank, the turbine operating parameters, and the load forecast for the next 4 hours, and outputs the optimal extraction ratio and thermal storage release strategy. A smooth transition strategy is adopted for the extraction ratio adjustment, with a single adjustment amplitude not exceeding 2% and an adjustment interval of no less than 5 minutes to avoid the impact of frequent adjustments on the turbine's operational stability.

[0046] The specific implementation of step S07 involves establishing a thermal storage efficiency monitoring system. This system uses a distributed sensor network to monitor the temperature distribution inside the thermal storage tank in real time. Eight layers of temperature measurement points are arranged vertically inside the tank, with four radial measurement points in each layer, totaling 32 temperature monitoring points. Temperature distribution monitoring employs wireless sensor network technology. The sensor nodes have data processing and wireless communication capabilities, with a data transmission frequency of once per minute. Pressure change monitoring is achieved using 10 pressure sensors, located at the top, middle, and bottom of the thermal storage tank, to monitor the pressure change trend and pressure distribution uniformity. Heat loss monitoring measures the heat flux density on the outer wall of the thermal storage tank using a heat flux density sensor, calculating the overall heat loss rate. The heat loss rate control target is less than 2% / day. Based on the monitoring results, a fuzzy control algorithm is used to optimize the molten salt circulation flow rate. When the temperature distribution non-uniformity exceeds 5°C, the molten salt circulation flow rate is increased; when the heat loss rate exceeds a set threshold, the operating status of the insulation system is adjusted. The heat exchanger operating parameters are optimized using a particle swarm optimization algorithm. With the goal of maximizing heat exchange efficiency, the flow ratio between the tube side and the shell side of the heat exchanger is optimized. The optimization variables include molten salt flow rate, feed water flow rate, and heat exchanger heat transfer coefficient. The constraints include flow range limitation and temperature difference limitation.

[0047] The detailed structure of the intelligent peak-shaving decision-maker is based on a sequence-to-sequence deep learning architecture, comprising four main components: an input encoding layer, a multi-layer self-attention mechanism, a feedforward network layer, and an output decoding layer. The input encoding layer handles multi-source heterogeneous data, including grid load forecasting data, thermal storage tank temperature and pressure status data, and turbine operating parameter data. An embedding layer maps different data types to a unified feature space, with an embedding dimension of 256. Position encoding is generated using sine and cosine functions to preserve temporal information, and the encoding length is consistent with the input sequence length. The multi-layer self-attention mechanism consists of six self-attention layers, each containing eight attention heads, each with a 32-dimensional dimension. This multi-head attention mechanism captures the correlations and temporal dependencies between different data sources. Self-attention computation employs a scaled dot product attention mechanism, extracting features through linear transformations of the query matrix, key matrix, and value matrix. A dropout rate of 0.1 is set to prevent overfitting. The feedforward network layer employs a two-layer fully connected network. The first layer has 1024 neurons, using the ReLU activation function, while the second layer has 256 neurons, used to extract nonlinear feature patterns and enhance the model's expressive power. Layer normalization is applied to the output of each sub-layer, and residual connections are used to mitigate the vanishing gradient problem. The output decoding layer contains three independent output branches, generating the current heat storage capacity, remaining heat storage capacity, and optimal steam extraction ratio of the thermal storage tank, respectively. Each branch is implemented using a fully connected layer, with the output layer activation functions being a linear function, the ReLU function, and the sigmoid function, respectively.

[0048] The detailed steps for establishing the training dataset for the intelligent peak-shaving decision-maker include four stages: data collection, preprocessing, sample construction, and dataset partitioning. The data collection stage acquires 12 consecutive months of operational data from the power plant's historical operation database, including minute-level grid load data, thermal storage tank temperature and pressure data, turbine operating parameter data, and corresponding thermal storage tag data, totaling approximately 500,000 records. Grid load data includes real-time load values, load forecast values, and load change rates. Thermal storage tank data includes temperature values ​​at 32 locations, pressure values ​​at 10 locations, and molten salt flow data. Turbine data includes speed, power, pressure and temperature at various levels, and vibration parameters. Tag data is obtained through a combination of heat balance calculations and expert annotation. Thermal storage calculations are based on molten salt temperature and mass, and the extracted tag ratios are based on historical best operating experience. The data preprocessing stage first involves data cleaning, removing abnormal and missing data during sensor malfunction periods. Abnormal data is defined as values ​​exceeding three standard deviations from the normal range. Then, data normalization is performed, using a minimum-maximum normalization method to scale all data to the range of 0 to 1, maintaining consistency in the numerical range across different data sources. Temporal alignment ensures consistency of timestamps across all data sources, and linear interpolation is used to fill in a few missing data points. During sample construction, a sliding window method is used to divide the continuous data into training samples. Each sample contains 4 hours of historical data as the input sequence and the target value for the next time point as the output sequence, resulting in a total of 100,000 samples. The input sequence length is set to 240 time steps, with each time step containing an 85-dimensional feature vector. The output sequence contains a 3-dimensional target variable. In the dataset partitioning phase, samples are split chronologically. The first 80% of the data serves as the training set for model parameter learning, the middle 10% as the validation set for hyperparameter tuning and early stopping mechanisms, and the last 10% as the test set for model performance evaluation, ensuring that the temporal distribution of the test data is consistent with real-world application scenarios.

[0049] The key technical concepts of this invention include four aspects: a dynamic peak-shaving game model, an intelligent peak-shaving decision-maker, a sliding time window adaptive strategy, and real-time monitoring and optimization of thermal storage efficiency. The dynamic peak-shaving game model, through a two-layer optimization architecture, simultaneously considers the dual objectives of maximizing grid peak-shaving benefits and maximizing turbine operational safety. Compared to traditional single-objective optimization methods, it achieves higher economic benefits while ensuring equipment safety. The coupled solution of the upper and lower layer models avoids suboptimal solutions caused by objective conflicts, enabling the thermal storage system to find the globally optimal operating strategy under complex constraints. The intelligent peak-shaving decision-maker employs a deep learning sequence-to-sequence architecture to process multi-source time-series data. Compared to traditional rule-based control methods, it has stronger nonlinear mapping capabilities and adaptive learning capabilities, automatically mining complex correlation patterns from historical operating data. Through a multi-head self-attention mechanism, it captures long-term dependencies between different data sources, achieving accurate prediction of the thermal storage system's state and intelligent optimization of the control strategy. The sliding time window adaptive strategy dynamically adjusts the model memory length based on the characteristics of grid load fluctuations. Compared with the traditional method of fixed window length, it can better adapt to the time-varying characteristics of grid load. A short memory window is used during periods of stable load to improve response speed, while a long memory window is used during periods of severe load fluctuations to enhance prediction stability. A smooth transition of the memory length is achieved through graded control of the intensity value. The real-time monitoring and optimization system for thermal storage efficiency achieves refined management of the thermal storage process through a distributed sensor network and intelligent algorithms. Compared with traditional coarse monitoring methods, it can promptly detect problems such as decreased thermal storage efficiency and increased heat loss. Multi-point temperature monitoring identifies temperature stratification within the thermal storage tank, and optimization of molten salt circulation flow and adjustment of heat exchanger parameters based on monitoring results significantly improves the overall efficiency of the thermal storage system. The synergistic effect of these four key technological approaches forms a complete intelligent thermal storage peak-shaving control system. A game theory model provides macroscopic optimization strategies, an intelligent decision-maker achieves precise microscopic control, an adaptive window strategy ensures rapid system response to environmental changes, and real-time monitoring and optimization ensure efficient and stable system operation. Compared with traditional separate control methods, this integrated intelligent control architecture can achieve a comprehensive improvement in the performance of the thermal storage system and a significant enhancement of the grid's peak-shaving capability.

[0050] It should be noted that this invention also solves the following technical problems: This invention addresses the technical problem of the inability to monitor and optimize thermal storage efficiency in real time in traditional thermal storage systems. In existing technologies, thermal storage systems typically lack precise monitoring of the internal temperature distribution, pressure changes, and heat loss of the storage tank, resulting in ineffective control and optimization of thermal storage efficiency. This invention establishes a thermal storage efficiency monitoring system to monitor the temperature distribution, pressure changes, and heat loss of the storage tank in real time, and optimizes the circulation flow rate of the high-temperature thermal storage medium and the operating parameters of the heat exchanger based on the monitoring results, achieving real-time monitoring and dynamic optimization of thermal storage efficiency. This system can promptly detect abnormalities in the thermal storage process, and maximize thermal storage efficiency, reduce heat loss, and improve the overall energy utilization efficiency of the system by adjusting the circulation flow rate of the thermal storage medium and the operating parameters of the heat exchanger. This invention also solves the technical problem that the memory length parameter in intelligent peak-shaving decision-making systems cannot adaptively adjust according to the system's operating status. Traditional intelligent decision-making systems typically use fixed memory length parameters, which cannot be dynamically adjusted according to actual operating conditions such as grid load fluctuations, changes in the temperature of the thermal storage tank, and changes in turbine power, resulting in low decision-making accuracy. This invention designs a proportional regulation function, which calculates the regulation intensity value based on the power grid load fluctuation amplitude, the thermal storage tank temperature change rate, and the turbine power change rate. Different memory length parameters are used according to different regulation intensity value ranges. When the regulation intensity value is in the range of 0 to 0.3, a short memory length of 8 time segments is used; when the regulation intensity value is in the range of 0.3 to 0.7, a medium memory length of 16 time segments is used; and when the regulation intensity value is in the range of 0.7 to 1.0, a long memory length of 32 time segments is used. This realizes the adaptive adjustment of the memory length parameter of the intelligent peak shaving decision-maker, improving the adaptability of the decision-making system to different operating conditions and the decision-making accuracy.

[0051] Specifically, the principle of this invention is as follows: The fundamental reason why this invention can solve the above-mentioned technical problems lies in its establishment of a complete intelligent decision-making system. This system can comprehensively consider three key factors: power grid load forecasting, thermal storage tank status, and turbine operation safety. First, the dynamic peak-shaving game model achieves multi-objective optimization through a two-layer structure. The upper-layer model considers maximizing the peak-shaving benefits of the power grid. The objective function includes the product of the peak-period power compensation and the current heat storage of the thermal storage tank, the reciprocal of the operating cost, the logarithm of the power grid load demand, and the sine function of the optimal steam extraction ratio. The lower-layer model considers maximizing the operation safety of the turbine. The objective function includes the product of the turbine operation stability index and the steam temperature, the reciprocal of the turbine vibration power, the cosine function of the blade angle, and the exponential function of the optimal steam extraction ratio. The two objective functions are coupled through the optimal steam extraction ratio term, ensuring that the decision result can both meet the peak-shaving needs of the power grid and guarantee the safe operation of the turbine. Secondly, the intelligent peak-shaving decision-maker adopts a deep network architecture based on a sequence-to-sequence model. It processes the correlation between grid load forecast data and thermal storage tank temperature and pressure status data through a multi-layer self-attention mechanism. The feedforward network layer extracts nonlinear feature patterns, and the output decoding layer generates the current heat storage capacity, remaining heat storage capacity, and optimal steam extraction ratio of the thermal storage tank, achieving intelligent decision-making for complex multivariate systems. Finally, the sliding time window strategy dynamically adjusts the upper limit of the extraction ratio based on grid load forecast data. The extraction ratio adjustment function calculates the adjustment intensity value based on the grid load fluctuation amplitude, the thermal storage tank temperature change rate, and the turbine power change rate. Different memory length parameters are used for different adjustment intensity values, realizing the adaptive adjustment capability of the decision-making system.

[0052] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0053] In this embodiment, the specific implementation of step S01 is the same as described above, and will not be repeated in detail here.

[0054] The specific implementation of step S02 involves establishing a data acquisition system and calculating the thermal storage parameters using an intelligent peak-shaving decision-maker. The formula for calculating the current heat storage capacity of the thermal storage tank is as follows:

[0055] ;

[0056] In the formula, The current heat storage capacity of the thermal storage tank is expressed in MWh. The density of molten salt is 2100 kg / m³. ; This represents the volume of molten salt, in units of... ; The specific heat capacity of molten salt is 1.5 kJ / (kg·K); The average temperature inside the thermal storage tank, in K; For reference temperature, the value is 823K; is the conversion factor from kJ to MWh.

[0057] in, Calculated from temperature sensor data:

[0058] ;

[0059] In the formula, For the first The measured values ​​of a temperature sensor, This represents the total number of temperature sensors, with a value of 20.

[0060] The formula for calculating the remaining thermal storage capacity is:

[0061] ;

[0062] In the formula, The remaining thermal storage capacity is expressed in MWh. The maximum heat storage capacity designed for the heat storage tank is 500 MWh; This represents heat loss, expressed in MWh.

[0063] The specific implementation of step S03 is to construct a dynamic peak-shaving game model, and the objective function of the upper-level model is expressed as:

[0064] ;

[0065] In the formula, This represents the objective function value of the upper-level model. This represents the power compensation amount during peak power periods, expressed in MW. This represents the current heat storage capacity of the thermal storage tank. Operating costs, expressed in yuan / hour; This represents the grid load demand, expressed in MW. The optimal steam extraction ratio ranges from 0.05 to 0.25. The default value for the reference power compensation is 100MW. For reference heat storage, the default value is 500MWh; For reference operating costs, the default value is 10,000 yuan / hour; For reference load demand, the default value is 1000MW.

[0066] The power compensation amount during peak hours is calculated as follows:

[0067] ;

[0068] In the formula, To predict peak load, This is the baseline load.

[0069] The formula for calculating operating costs is:

[0070] ;

[0071] In the formula, For fuel costs, To maintain costs, This refers to the cost of electricity consumption.

[0072] The objective function of the lower-level model is expressed as:

[0073] ;

[0074] In the formula, The objective function value of the lower-level model; This is an indicator of the operational stability of the steam turbine, with a value range of 0 to 1. This refers to the steam temperature, expressed in Kelvin (K). This refers to the turbine vibration power, measured in kW. The blade angle is expressed in radians. It is a natural constant; : Refer to the stability index, the default value is 0.9; For reference steam temperature, the default value is 833K (corresponding to 560℃); For reference vibration power, the default value is 50kW; For reference blade angle, the default value is 0.5 radians; For reference sampling ratio, the default value is 0.15.

[0075] The turbine's operational stability index is calculated as follows:

[0076] ;

[0077] In the formula, For speed stability indicators, For power stability indicators, As a pressure stability indicator, Let be the weighting coefficient, satisfying .

[0078] The specific implementation of step S04 is to perform thermal storage during the off-peak period of the power grid load. The formula for calculating the steam extraction flow rate is as follows:

[0079] ;

[0080] In the formula, The steam flow rate is expressed in kg / s. This represents the total steam flow rate of the steam turbine, expressed in kg / s.

[0081] The formula for calculating the heat exchange capacity of a heat exchanger is:

[0082] ;

[0083] In the formula, This refers to the heat exchanger's heat exchange capacity, measured in kW. The value is the enthalpy of the imported steam, expressed in kJ / kg. This is the enthalpy value of the outlet condensate, expressed in kJ / kg.

[0084] The specific implementation of step S05 is to release heat during peak grid load periods, and the power compensation amount is calculated using the following formula:

[0085] ;

[0086] In the formula, The increase in power generation is expressed in MW. The turbine efficiency ranges from 0.35 to 0.42. The thermal power released by the thermal storage tank is expressed in MW.

[0087] The heat power released by the thermal storage tank is calculated as follows:

[0088] ;

[0089] in The thermal power released by the thermal storage tank, measured in kW. This is the molten salt mass flow rate, expressed in kg / s. This refers to the enthalpy of molten salt at high temperatures, expressed in kJ / kg. This represents the enthalpy of molten salt at low temperatures, expressed in kJ / kg. The enthalpy of molten salt at high temperatures and low temperatures are calculated using the following formulas:

[0090] ;

[0091] ;

[0092] When the specific heat capacity of molten salt varies little within the operating temperature range, the average specific heat capacity can be used for simplified calculation. In this case, the enthalpy calculation formula is simplified to:

[0093] and ;

[0094] in The enthalpy is the value under reference conditions, usually taken as 0 at the reference temperature, and the unit is kJ / kg. For reference temperatures, the default values ​​are 273.15K or 298.15K. This refers to the temperature of high-temperature molten salt, typically 883-903K, or 610-630℃. This refers to the temperature of low-temperature molten salt, typically 853K, which is equivalent to 580℃. The average isobaric specific heat capacity over the operating temperature range is approximately 1.5 kJ / (kg·K) for a mixed molten salt of 60% sodium nitrate and 40% potassium nitrate.

[0095] Optionally, for more accurate calculations that require consideration of the temperature dependence of specific heat capacity, the temperature relationship of molten salt specific heat capacity can be used. ,in This is the fitting coefficient for the specific heat capacity of molten salt; for nitrate molten salts, the typical value is... , , The corresponding formula for calculating the precise enthalpy is:

[0096] .

[0097] The specific implementation of step S06 is to adjust the extraction ratio in real time through an intelligent peak-shaving decision-maker. The dynamic adjustment formula for the upper limit of the extraction ratio is as follows:

[0098] ;

[0099] In the formula, This is the upper limit of the extraction ratio; The baseline sampling ratio is 0.2. This is an adjustment factor with a value of 0.5; This refers to the power grid load fluctuation rate.

[0100] The formula for calculating power grid load fluctuation rate is:

[0101] ;

[0102] In the formula, For the first The load value at a given moment. This is the average load value. This represents the number of data points within the time window.

[0103] The specific implementation of step S07 is to establish a thermal storage efficiency monitoring system, and the formula for calculating the heat loss rate is:

[0104] ;

[0105] In the formula, This refers to the heat loss rate; The outer surface area of ​​the thermal storage tank is expressed in units of... ; The heat transfer coefficient is expressed in W / ( ·K); The temperature difference between the inside and outside is expressed in Kelvin (K). The time interval is expressed in hours (h).

[0106] The formula for calculating temperature distribution non-uniformity is:

[0107] ;

[0108] In the formula, This represents the standard deviation of the temperature distribution, used to evaluate the uniformity of temperature distribution.

[0109] The molten salt circulation flow rate optimization uses a proportional-integral-derivative (PID) control algorithm, and the control equation is:

[0110] ;

[0111] In the formula, To control the output signal; This is an error signal; These are the proportional, integral, and differential coefficients, respectively.

[0112] The formula for calculating the adjustment intensity value of the extraction proportional adjustment function is as follows:

[0113] ;

[0114] In the formula, To adjust the intensity value, the numerical range is from 0 to 1; Let be the weighting coefficient, satisfying ; The temperature change rate of the thermal storage tank is expressed in K / min. This represents the turbine power change rate, expressed in MW / min.

[0115] The parameter acquisition method is as follows: Obtained through statistical calculation of historical power grid load data; It is obtained by calculating the time derivative of temperature sensor data; Obtained by calculating the time derivative of turbine power output data; weighting coefficient Determined through fitting historical operational data, the typical value is: .

[0116] The principles and effects of each formula are explained below. The current formula for calculating the heat storage capacity of the thermal storage tank is based on the first law of thermodynamics. It calculates the heat storage capacity through the physical properties of the molten salt and temperature changes. This formula considers key physical properties such as molten salt density and specific heat capacity, providing more accurate heat storage calculations compared to traditional empirical estimation methods. It enables real-time and accurate monitoring of the thermal storage system's status, providing a reliable data foundation for subsequent control decisions. The dynamic peak-shaving game model uses a multinomial combination form for its two-layer objective function. The upper-layer model reflects the synergistic effect of heat storage and power compensation through a product term, while the reciprocal term ensures effective control of operating costs. The logarithmic and sine terms handle the nonlinear characteristics of load demand and the periodic constraints of the extraction ratio, respectively. The lower-layer model strengthens the coupling relationship between stability and temperature through a product term, prevents excessive vibration power through a reciprocal term, and handles the angular characteristics of the blade angle and the exponential growth characteristics of the extraction ratio, respectively. Compared to traditional single-objective linear optimization methods, this game model can achieve multi-objective coordinated optimization under complex constraints, significantly improving the system's economy and safety. The dynamic adjustment formula for the upper limit of the extraction ratio achieves adaptive control through linear adjustment of the load fluctuation rate. When the load fluctuates drastically, it automatically lowers the upper limit of the extraction ratio to ensure system stability, and appropriately increases the extraction ratio to enhance the thermal storage effect when the load is stable. Compared with the traditional method of fixed ratio limitation, this dynamic adjustment strategy can optimize system operating parameters in real time according to the grid status, effectively balancing the contradiction between thermal storage efficiency and operational safety. The heat loss rate calculation formula is based on the basic equation of heat transfer. It calculates the heat loss power by multiplying the surface area, heat transfer coefficient, and temperature difference, and then obtains the loss rate by dividing it by the amount of heat stored. This formula accurately reflects the insulation performance and heat loss characteristics of the thermal storage tank, providing more accurate evaluation results compared with simplified heat loss estimation methods, and providing a quantitative analysis basis for optimizing the efficiency of thermal storage systems. The formula for calculating the regulation intensity value adopts a multivariate linear weighted combination form, which comprehensively considers three key factors: power grid load fluctuation, thermal storage tank temperature change, and turbine power change. The importance of different factors is adjusted through weighting coefficients. Compared with the traditional method of single-variable control, this multivariate fusion strategy can more comprehensively reflect the system operating status, realize intelligent adaptive adjustment of memory length parameters, and significantly improve the adaptability and response accuracy of the control system to complex operating conditions.

[0117] It should be noted that the variables involved in this invention are explained in detail in Table 1 below.

[0118] Table 1. Variable Explanation Table

[0119]

[0120] To better understand and implement this invention, a specific application scenario is provided below as Example 2: A technical team implemented a high-temperature thermal energy storage dynamic peak-shaving system upgrade on a 600MW supercritical thermal power unit. The unit's original peak-shaving capacity was limited, making it difficult to meet rapid response requirements under conditions of large grid load fluctuations. Based on the technical solution of this invention, the technical team designed and established a complete thermal energy storage peak-shaving system. The high-temperature thermal energy storage system involved in the thermal power plant is as follows... Figure 3 As shown.

[0121] In the system establishment process of step S01, the technical team first established the structure of the thermal storage tank. The thermal storage tank adopts a cylindrical design with a diameter of 18m, a height of 25m, and an effective volume of 6000 cubic meters. The thermal storage tank has an inner layer of 12mm thick 316L stainless steel, an outer layer of 200mm thick aluminum silicate fiber insulation, and a stainless steel outer protective plate. The tank is designed to operate at a pressure of 2.5MPa and a temperature of 680℃, with a storage capacity of 500MWh. The technical team filled the tank with 3600t of binary molten salt, with a salt composition of 60%. and 40% It has a melting point of 220℃ and an operating temperature range of 290℃ to 565℃.

[0122] The steam extraction pipeline system consists of two parts: a main extraction pipeline and an auxiliary pipeline. The main extraction pipeline connects to the fourth-stage extraction port of the intermediate-pressure cylinder of the steam turbine, with a diameter of 900 mm, a length of 85 m, and a design flow rate of 180 t / h. The auxiliary pipeline connects to the fifth-stage extraction port, with a diameter of 600 mm and a design flow rate of 120 t / h. All extraction pipelines are insulated with a thickness of 150 mm, using P91 steel, and have a design pressure resistance of 3.5 MPa. Twelve electrically operated regulating valves are installed at key locations along the pipeline, with a valve response time of 8 seconds and a control accuracy of ±0.5%.

[0123] The heat exchanger system consists of two parts: a steam condenser and a molten salt heater. The steam condenser adopts a horizontal shell-and-tube structure with a heat exchange area of ​​1800 mm. The tube bundle uses φ25×3mm stainless steel tubes, with a total of 3200 tubes. The molten salt heater adopts a vertical shell-and-tube structure with a heat exchange area of ​​2200... The heat exchange efficiency is designed to be 88%. The heat exchanger shell is made of 16MnR steel, the tube bundle is made of TP347H stainless steel, the design pressure is 2.8MPa, and the design temperature is 650℃.

[0124] The temperature and pressure monitoring sensor network is deployed in a layered manner, with eight monitoring layers vertically arranged within the thermal storage tank. Each layer contains four temperature measuring points and two pressure measuring points, totaling 32 temperature sensors and 16 pressure sensors. The temperature sensors are platinum resistance thermometers, with a measurement range of 200℃ to 700℃ and an accuracy of ±1.5℃. The pressure sensors are diffused silicon sensors, with a measurement range of 0 to 3.0 MPa and an accuracy of ±0.03 MPa. Sensor data is collected to the central control system wirelessly, with a data acquisition frequency of twice per second.

[0125] In step S02, during data acquisition, the technical team established a comprehensive data acquisition platform to obtain load forecast data from the power grid dispatch center. The forecast time span was 6 hours, and the data update interval was 10 minutes. The thermal storage tank status monitoring data included 32 temperature measuring points, 16 pressure measuring points, and molten salt flow data, with an acquisition cycle of 0.5 seconds. Turbine operating parameters included over 200 monitoring parameters such as unit speed (3000 r / min), rated power (600 MW), extraction parameters at each stage, and blade angles. The intelligent peak-shaving decision-maker used a deep learning algorithm to calculate the current heat storage capacity of the thermal storage tank in real time. Under typical operating conditions, the heat storage capacity was 385 MWh, with a remaining heat storage capacity of 115 MWh.

[0126] In step S03, the implementation of the dynamic peak-shaving game model employs a two-layer optimization algorithm. The upper-layer model aims to maximize the grid's peak-shaving revenue, while the lower-layer model is constrained by the safe operation of the steam turbine. During a typical day, the grid load reaches its lowest point of 420MW at 6:00 AM. At this time, the game model calculates the optimal steam extraction ratio to be 18%, corresponding to an extraction steam flow rate of 156t / h. When the load reaches its peak of 580MW at 2:00 PM, the model determines the optimal thermal energy storage release strategy: a molten salt flow rate of 280t / h, generating 130t / h of superheated steam and increasing power generation by 95MW.

[0127] The operating parameter monitoring data shown in Table 2 reflects the system's performance under different operating conditions.

[0128] Table 2 System operating parameters under typical operating conditions

[0129]

[0130] In the thermal storage mode implementation of step S04, when the grid load is at its lowest point and the duration exceeds 45 minutes, the system automatically activates the thermal storage mode. The steam extraction valve adjusts its opening to 72% based on the game theory model calculations, extracting steam at a temperature of 545℃, a pressure of 1.85MPa, and a stable flow rate of 156t / h. The extracted steam exchanges heat with molten salt in the heat exchanger. After condensation, the steam temperature drops to 185℃, and the condensate is returned to the unit via the return water system. During the heat exchange process, the molten salt temperature rises from 485℃ to 550℃, achieving a thermal storage power of 185MW. The thermal storage process lasts 3.5 hours, accumulating a thermal storage capacity of 648MWh.

[0131] Step S05, the heat release mode, is implemented during peak grid load periods, initiated when the load exceeds 120% of the baseline value and the predicted duration is greater than 30 minutes. High-temperature molten salt is drawn from the top of the thermal storage tank via a circulating pump at a flow rate of 280 t / h and a temperature of 548°C. The molten salt exchanges heat with feedwater in a heat exchanger, heating the feedwater temperature from 165°C to 535°C, generating 130 t / h of superheated steam at a pressure of 1.95 MPa. The superheated steam is sent to the intermediate-pressure cylinder of the steam turbine, driving the generator to increase the output power by 95 MW. After heat release, the temperature of the molten salt drops to 465°C and returns to the bottom of the thermal storage tank via a reflux system. The heat release process lasts for 2.8 hours, releasing a cumulative heat of 784 MWh.

[0132] In the implementation of intelligent peak shaving in step S06, the intelligent peak shaving decision-maker processes historical data using a 4-hour sliding time window and dynamically adjusts the upper limit of the extraction ratio based on the characteristics of grid load fluctuations. During a stable period with a load fluctuation rate of 8%, the upper limit of the extraction ratio is set to 20%. When the load fluctuation rate rises to 15%, the upper limit is automatically adjusted to 16%. In the event of severe load fluctuations reaching 25%, the upper limit is reduced to 12% to ensure system stability. The decision-maker updates the optimization strategy every 5 minutes, and the average response time for adjusting the extraction ratio is 35 seconds.

[0133] The thermal storage efficiency monitoring data shown in Table 3 reflects the thermal performance of the system.

[0134] Table 3. Thermal Storage Efficiency Monitoring Data

[0135]

[0136] In the implementation of the thermal storage efficiency monitoring system in step S07, the technical team monitored the temperature distribution inside the thermal storage tank in real time through 32 temperature monitoring points. They found significant temperature stratification, with the upper part of the tank being 12°C higher than the lower part. Pressure monitoring showed uniform pressure distribution inside the tank, with a maximum pressure difference of 0.08 MPa. Heat loss monitoring was achieved by measuring the heat flux density at 16 points on the tank wall, and the calculated daily heat loss rate was 1.8%, meeting the design requirements. Based on the monitoring results, the system used a fuzzy control algorithm to optimize the molten salt circulation flow rate. When the temperature non-uniformity exceeded 4°C, the circulation flow rate was increased by 15%, effectively improving the temperature distribution uniformity.

[0137] like Figure 4 As shown, the temperature distribution inside the thermal storage tank exhibits a clear stratification characteristic, with a higher temperature at the top and a relatively lower temperature at the bottom, and the temperature gradient is most pronounced in the middle region. Figure 5 As shown, the system's power regulation curve during the 24-hour operating cycle demonstrates a good peak-shaving effect, effectively storing heat during off-peak periods and releasing energy promptly during peak periods. Figure 6 As shown, the temperature change of molten salt over time reflects the dynamic characteristics of the heat storage and release process, and the temperature change is stable and controllable.

[0138] The decimation proportional control function exhibits good adaptive performance in actual operation. The regulation intensity value is calculated based on the grid load fluctuation amplitude, the temperature change rate of the thermal storage tank, and the power change rate of the turbine. Under typical operating conditions, the load fluctuation amplitude is 0.12, the temperature change rate is 2.8℃ / min, and the power change rate is 0.85MW / min. The calculated regulation intensity value is 0.42, corresponding to a control strategy with a medium memory length of 16 time segments.

[0139] Operational test results show that during 30 consecutive days of operation, the thermal storage peak-shaving system achieved an average thermal storage efficiency of 91.6%, an average heat release efficiency of 88.2%, and a system availability of 98.5%. The unit's peak-shaving response time was reduced from the traditional 45 minutes to 25 minutes, and the peak-shaving capacity increased from 120MW to 195MW. Under conditions of rapid changes in grid load, the system can switch from thermal storage mode to heat release mode within 30 minutes, effectively meeting the grid's peak-shaving needs.

[0140] Compared to traditional thermal power peak shaving methods, the technical solution of this invention brings significant technological advancements. Traditional peak shaving mainly relies on changing fuel supply and feedwater flow to adjust unit output, resulting in slow response and significant equipment stress. This invention, however, achieves time-based heat transfer through the physical energy storage of the thermal storage medium, avoiding the thermal shock to the boiler and turbine caused by frequent combustion adjustments. The dual-layer optimization architecture of the dynamic peak shaving game model balances economy and safety. Compared to traditional single-objective control, it can find the global optimal solution under complex constraints, effectively balancing the conflict between peak shaving benefits and equipment safety. The intelligent peak shaving decision-maker, based on deep learning's sequence prediction capabilities, has stronger environmental adaptability than traditional rule-based control methods. It can automatically learn the optimal control strategy from historical data, achieving adaptive adjustment of control parameters. The real-time thermal storage efficiency monitoring system achieves refined management and control of the thermal storage process through a distributed sensor network and intelligent algorithms. Compared to traditional extensive monitoring methods, it can promptly detect and address the problem of declining thermal storage efficiency, significantly improving the overall performance and reliability of the system.

[0141] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage, used for dynamic peak-shaving using a high-temperature thermal energy storage system in the thermal power plant, wherein the high-temperature thermal energy storage system includes a thermal storage tank, a steam extraction pipeline, a heat exchanger, and temperature and pressure monitoring sensors; the thermal storage tank is filled with a high-temperature thermal storage medium; and the steam extraction pipeline connects the intermediate stage steam extraction port of the steam turbine to the inlet of the thermal storage tank; characterized in that... The method includes: real-time acquisition of power grid load forecast data, thermal storage tank temperature and pressure status data, and turbine operating parameter data; and calculation of the current heat storage capacity and remaining heat storage capacity of the thermal storage tank through an intelligent peak-shaving decision-maker; based on the power grid load forecast data and thermal storage tank temperature and pressure status data, determining the optimal steam extraction ratio and thermal storage release strategy through a dynamic peak-shaving game model, which includes an upper-level model aimed at maximizing power grid peak-shaving benefits and a lower-level model aimed at maximizing turbine operating safety; during periods of low power grid load, extracting intermediate-stage steam from the turbine to the thermal storage tank according to the optimal steam extraction ratio; transferring heat from the extracted steam to the high-temperature thermal storage medium through a heat exchanger, and returning the condensate to the turbine circulation system; and during periods of high power grid load, activating the thermal storage tank's thermal system, whereby the high-temperature thermal storage medium heats the feedwater through a heat exchanger to generate steam, which drives the turbine impeller to increase power generation, thereby achieving power compensation during peak power grid load periods. The steps of the dynamic peak-shaving game model are as follows: the objective function of the upper-level model is the product of the power compensation amount during the peak period of the power grid and the current heat storage capacity of the thermal storage tank, plus the inverse term of the operating cost minus the logarithm of the power grid load demand, plus the sine function term of the optimal steam extraction ratio. The constraints include thermal storage tank capacity constraints, power grid load balance constraints, and steam extraction ratio range constraints. The objective function of the lower-level model is the product of the turbine operating stability index and the steam temperature, plus the inverse term of the turbine vibration power minus the cosine function term of the blade angle, plus the exponential function term of the optimal steam extraction ratio. The constraints include turbine operating parameter safety range constraints, steam temperature and pressure constraints, and blade stress constraints.

2. The dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage according to claim 1, characterized in that, The coupling term of the two objective functions is the optimal steam extraction ratio term, which represents the combined impact of the steam extraction ratio on the grid peak-shaving efficiency and the turbine operation safety.

3. The dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage according to claim 2, characterized in that, The intelligent peak-shaving decision-maker is specifically based on a deep network architecture of sequence-to-sequence model, including an input encoding layer, a multi-layer self-attention mechanism, a feedforward network layer, and an output decoding layer. The self-attention mechanism is used to process the correlation between power grid load forecast data and thermal storage tank temperature and pressure status data. The feedforward network layer is used to extract nonlinear feature patterns. The output decoding layer generates the current heat storage capacity, remaining heat storage capacity, and optimal steam extraction ratio of the thermal storage tank.

4. The dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage according to claim 3, characterized in that, The intelligent peak shaving decision-maker adopts a sliding time window strategy to dynamically adjust the upper limit of the extraction ratio based on the power grid load forecast data. The sliding time window strategy is a data processing method that uses a fixed-length time window to analyze historical data and updates the window content over time.

5. The dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage according to claim 4, characterized in that, The training dataset for the intelligent peak-shaving decision-maker is established by collecting historical grid load data, thermal storage tank operating status data, turbine operating parameter data, and corresponding label data for the current heat storage capacity, remaining heat storage capacity, and optimal steam extraction ratio of the thermal storage tank. The collected data is preprocessed, including data cleaning, normalization, and time-series alignment. The preprocessed data is then segmented according to time windows to form training samples.

6. The dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage according to claim 5, characterized in that, The training of the intelligent peak shaving decision-maker specifically involves supervised learning training of the model using training set data, employing a mean squared error loss function and an adaptive moment estimation optimizer for parameter updates, setting the learning rate to 0.001, the batch size to 32, and the number of training epochs to 100. During the training process, a validation set is used to evaluate model performance and an early stopping mechanism is implemented to prevent overfitting.

7. The dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage according to claim 6, characterized in that, The memory length parameter in the intelligent peak shaving decision-maker is determined based on the grid load prediction time window length, the thermal capacity of the thermal storage tank, and the turbine response time. When the grid load prediction time window length is 4 hours, the thermal capacity of the thermal storage tank is 500 MWh, and the turbine response time is 15 minutes, the memory length parameter is set to 16 time segments.

8. The dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage according to claim 7, characterized in that, The extraction proportional control function is specifically calculated based on the grid load fluctuation amplitude, the thermal storage tank temperature change rate, and the turbine power change rate to obtain the control intensity value. When the control intensity value is in the range of 0 to 0.3, a short memory length of 8 time segments is used; when the control intensity value is in the range of 0.3 to 0.7, a medium memory length of 16 time segments is used; and when the control intensity value is in the range of 0.7 to 1.0, a long memory length of 32 time segments is used.

9. The dynamic peak-shaving method for thermal power plants utilizing high-temperature thermal energy storage according to claim 8, characterized in that, The memory length parameter is the time series length considered by the intelligent peak shaving decision-maker when processing historical data; the adjustment intensity value is the value used to adjust the memory length parameter, calculated by the extraction proportional adjustment function; the grid load fluctuation amplitude is the degree of change of grid load within a certain period of time; the thermal storage tank temperature change rate is the rate of change of temperature inside the thermal storage tank over time; the turbine power change rate is the rate of change of turbine output power over time; and the time segment is the basic time unit for data processing by the intelligent peak shaving decision-maker.

Images