A unit coordinated control system under thermal power deep peak shaving

Abstract

The application discloses a kind of thermal power depth peak shaving unit coordination control system, comprising: total load acquisition module, target load acquisition module, parameter acquisition module, load calculation module, load definition module, deviation calculation module, target loss construction module, parameter updating module, for parameter updating to several control parameters, to minimize target loss function;Parameter judging module is used to judge whether target loss function converges;If not convergent, return load calculation module;Otherwise, output current parameter as the optimal execution parameter of the thermal power unit;The application obtains the predicted value of the total load of demand side and new energy output of future anchoring time, calculates the target load of thermal power, realizes the quantitative decomposition of thermal power depth peak shaving task;Effectively solve the problem that control parameter is difficult to dynamically update facing future load, improve the coordination control ability of thermal power unit in high fluctuation power grid.

Description

Technical Field

[0001] This invention relates to a unit coordination control system, specifically a unit coordination control system for deep peak shaving in thermal power plants. Background Technology

[0002] With the large-scale grid connection of new energy sources, the load volatility of the power system has significantly increased, and traditional thermal power units are gradually shifting from supporting basic loads to supporting regulating loads. In deep peak-shaving scenarios, thermal power units need to undertake the tasks of rapid response and frequent adjustment, and their coordination control accuracy and regulation efficiency have become key factors for the stable operation of the power grid. Patent document CN117970808A discloses a deep peak-shaving control method and system for thermal power units, which solves the problems of narrow applicability, poor stability, and poor control accuracy of traditional thermal power unit coordination systems.

[0003] However, while the aforementioned patent documents and existing technologies demonstrate a certain load response capability in thermal power peak-shaving control, they often prioritize the current system state at the coordinated control level, lacking systematic prediction of future load trends. Furthermore, the control parameters of thermal power units are mostly based on single-point feedback, making it difficult to form a refined model of the relationship between adjustable parameters and output response oriented towards future load targets. In other words, current control strategies under deep peak-shaving in thermal power struggle to accurately identify the target load response of the thermal power system at multiple future moments, and cannot minimize the relationship between parameter updates and load deviations, severely restricting the peak-shaving regulation capability of thermal power units in high-fluctuation scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a coordinated control system for thermal power units under deep peak shaving, which solves the technical problems mentioned in the background art through bidirectional control of parameter updates and load deviations.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A coordinated control system for thermal power units under deep peak shaving, the control system comprising:

[0007] The total load acquisition module is used to acquire the total demand-side load at the future anchoring time.

[0008] The target load acquisition module is used to determine the target load of thermal power units at the future anchoring time based on the total demand-side load.

[0009] The target load for thermal power is characterized as the first load deviation between the total demand-side load and the renewable energy load.

[0010] The parameter acquisition module is used to acquire several control parameters of the thermal power unit at the current moment;

[0011] The load calculation module is used to calculate the thermal power output load at the current moment based on several control parameters at the current moment.

[0012] The load definition module is used to define the thermal power output load at the current moment as the predetermined thermal power load at the future anchoring moment;

[0013] The deviation calculation module is used to calculate the second load deviation between the target load of thermal power and the predetermined load of thermal power at the future anchoring time;

[0014] The target loss construction module is used to construct the target loss function based on several control parameters at the current moment and the second load deviation;

[0015] The parameter update module is used to update several control parameters in order to minimize the target loss function;

[0016] The parameter determination module is used to determine whether the target loss function has converged;

[0017] If convergence is not achieved, the process returns to the load calculation module; otherwise, the current parameters are output as the optimal execution parameters for the thermal power unit to control the deep peak shaving process of the thermal power.

[0018] In some specific embodiments, the target load acquisition module is specifically used for:

[0019] Obtain several future anchoring times, and the corresponding meteorological forecast parameters for those future anchoring times;

[0020] The future anchoring time and the corresponding meteorological forecast parameters are used to construct a time series vector of new energy load, which is then input into a pre-trained new energy load forecasting model to output the new energy load at the future anchoring time.

[0021] Calculate the first load deviation between the total demand-side load and the renewable energy load;

[0022] The first load deviation is defined as the target load of the thermal power unit at the future anchoring time.

[0023] In some specific embodiments, the pre-construction step of the new energy load forecasting model includes:

[0024] H1. Collect meteorological parameters corresponding to several historical time points in the target area;

[0025] H2. Perform feature preprocessing on meteorological parameters to generate preprocessed meteorological parameter features;

[0026] H3. Use historical time points as time indices and associate them with the meteorological parameter characteristics to generate time-series input variables;

[0027] H4. Obtain a predefined lag time window, and collect the new energy load corresponding to several historical time points within the lag time window;

[0028] H5. Define the renewable energy load corresponding to historical time points as time-series target variables;

[0029] H6. Based on the aforementioned time-series input variables and time-series target variables, construct a time-series sample of new energy loads for supervised learning;

[0030] H7. Input the time series samples of new energy load into a general time series regression model for supervised training. After iteration, a new energy load prediction model is generated.

[0031] In some specific embodiments, the load calculation module is specifically used for:

[0032] Obtain several control parameters of the thermal power unit at the current moment;

[0033] Several control parameters include: boiler fuel input rate, air supply volume, main steam valve opening, and turbine governor setpoint;

[0034] Based on several control parameters at the current moment, determine the thermal power output load of the thermal power unit at the future anchoring moment;

[0035] In some specific embodiments, the deviation calculation module is specifically used for:

[0036] Obtain the target load and the established load of thermal power at the future anchoring time;

[0037] Calculate the absolute error between the target load of thermal power and the predetermined load of thermal power;

[0038] The absolute error is squared twice to determine the second load deviation at the future anchoring time.

[0039] In some specific embodiments, the target loss construction module is specifically used for:

[0040] Several pre-defined pre-adjustment parameters for future anchoring times;

[0041] Calculate the parameter deviation between the pre-adjustment parameter and the control parameter;

[0042] Regularization is applied to the parameter deviation to construct a parameter change penalty term;

[0043] The target loss function is constructed by taking the second load deviation at the future anchoring time as the first loss term and the parameter change penalty term as the second loss term.

[0044] In some specific embodiments, the expression for the target loss function is:

[0045]

[0046] Where ε is the target loss, n is the preset adjustment quantity for future anchoring times, and i is the time index. This represents the target load of thermal power at the i-th future anchoring time. Let represent the predetermined load of thermal power determined by control parameters at the i-th future anchoring time, λ be the regularization weighting coefficient used to control the penalty degree for parameter deviations between adjacent times, m be the number of control parameters, j be the parameter index of the controllable parameter, and θ be the parameter index. j Let θ represent the j-th control parameter. j-1 This represents the j-th control variable parameter in the previous control step.

[0047] In some specific embodiments, the parameter update module is specifically used for:

[0048] Use the current control parameters as initialization parameters, and set the learning rate and maximum number of iterations;

[0049] Based on several control parameters at the current moment, determine the real-time updated loss of the target loss function;

[0050] Calculate the gradient vector of the loss relative to the control parameters in real time;

[0051] Based on the gradient vector and the set learning rate, several control parameters at the current time are updated to generate the parameters for the next iteration, until the target loss function converges.

[0052] This invention provides a coordinated control system for thermal power units under deep peak shaving, which has the following beneficial effects:

[0053] This invention calculates the target load for thermal power by obtaining the total demand-side load and predicted renewable energy output at a future anchoring time, thus achieving a quantitative decomposition of the deep peak-shaving task for thermal power. Furthermore, it estimates the predetermined load for thermal power by fitting the response between control parameters and thermal power units based on historical parameters. Then, it jointly models load deviation and parameter changes, iteratively optimizing the control parameters through a parameter update strategy, enabling thermal power units to maintain the stability of the parameter adjustment path while meeting the target load. This invention effectively solves the problem of difficulty in dynamically updating control parameters when facing future loads, and improves the coordinated control capability of thermal power units in highly volatile power grids. Attached Figure Description

[0054] Figure 1 This is a structural block diagram of a coordinated control system for thermal power units under deep peak shaving, according to the present invention.

[0055] Figure 2This is a schematic diagram of a coordinated control process for thermal power units under deep peak shaving according to the present invention;

[0056] Figure 3 This is a schematic diagram illustrating the construction process of the target loss function described in this invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Example 1: Please refer to Figure 1 This invention provides a coordinated control system for thermal power units under deep peak shaving, the control system comprising the following modules:

[0059] The total load acquisition module is used to acquire the total demand-side load at the future anchoring time.

[0060] The deep peak-shaving scenario refers to an operating state where the output ratio of new energy sources such as wind power and photovoltaics is relatively high, the system load fluctuates drastically, and there is a risk of dynamic imbalance between power supply and demand. This typically occurs during the grid operation phase when intermittent new energy sources such as wind power and photovoltaics are connected to the grid on a large scale. In this scenario, to maintain power system balance, thermal power units are required to have strong capabilities for rapid load increases and decreases and wide-range load response.

[0061] Furthermore, the total demand-side load represents the actual electricity load demand generated by power users at a future anchor time, and serves as the fundamental reference data for power grid operation and dispatch. Generally, this total demand-side load is determined by the power grid dispatch center based on the results generated by the load forecasting model, and serves as a core input item in the system dispatch plan. Its sources may include short-term load forecast data provided by the State Grid, local power grid companies, or regional energy dispatch platforms. This type of data typically has hourly granularity, with forecast periods covering the next 1 to 72 hours.

[0062] The target load acquisition module is used to determine the target load of thermal power units at the future anchoring time based on the total demand-side load.

[0063] The target load for thermal power is characterized as the first load deviation between the total demand-side load and the renewable energy load.

[0064] The parameter acquisition module is used to acquire several control parameters of the thermal power unit at the current moment;

[0065] The load calculation module is used to calculate the thermal power output load at the current moment based on several control parameters at the current moment.

[0066] The load definition module is used to define the thermal power output load at the current moment as the predetermined thermal power load at the future anchoring moment;

[0067] The deviation calculation module is used to calculate the second load deviation between the target load of thermal power and the predetermined load of thermal power at the future anchoring time;

[0068] The target loss construction module is used to construct the target loss function based on several control parameters at the current moment and the second load deviation;

[0069] The parameter update module is used to update several control parameters in order to minimize the target loss function;

[0070] The parameter judgment module is used to determine whether the target loss function has converged. If it has not converged, it returns to the load calculation module; otherwise, it outputs the current parameters as the optimal execution parameters of the thermal power unit. The optimal execution parameters are used to control the thermal power unit to output the target load at the future anchoring time.

[0071] Specifically, convergence conditions may include: the change in the target loss is less than a preset threshold, the step size of each parameter change is lower than a set lower limit, or the maximum number of iterations is reached.

[0072] In this embodiment, by connecting the total load acquisition module and the target load acquisition module in series, the system can accurately perceive the total demand-side load and new energy load at multiple future times, thereby generating the target thermal power load that the thermal power units need to bear. Combining the parameter acquisition module and the load calculation module, the system can calculate the thermal power output load based on the current controllable parameters, realizing the prediction and estimation of the predetermined thermal power load at the future anchoring time. Furthermore, through the joint execution of the deviation calculation module and the target loss construction module, a target loss function based on the target deviation can be formed within the system. Finally, based on the target loss function, the optimal execution parameters are determined to control the thermal power units to output the target thermal power load at the future anchoring time, thereby balancing the total demand-side load under the deep peak shaving scenario.

[0073] Example 2: See Figures 2 to 3 The technical solution that differs from that of Embodiment 1 in that Embodiment 2 discloses the specific application of each module of the system in Embodiment 1.

[0074] For example, in this embodiment, the target load acquisition module is specifically used for:

[0075] S1-1. Obtain several future anchoring times and the corresponding meteorological forecast parameters for those future anchoring times;

[0076] Specifically, the meteorological forecast parameters can be obtained from authoritative meteorological service platforms, such as hourly forecast data provided by the National Meteorological Information Center or third-party professional meteorological parameter providers.

[0077] S1-2. Construct a new energy load time series vector by combining the future anchoring time with the corresponding meteorological forecast parameters, and input it into the pre-trained new energy load forecasting model to output the new energy load at the future anchoring time.

[0078] Specifically, the new energy load refers to the predicted output load of new energy sources such as wind power and photovoltaics at the future anchoring time, which is essentially a time-series prediction result of new energy load.

[0079] It should be noted that renewable energy load forecasting technology is relatively mature in this field. Existing methods generally use time series modeling based on historical power output and meteorological factors for forecasting, which is a conventional approach. For example, a renewable energy load forecasting model based on time series forecasting can be constructed, and the renewable energy load corresponding to multiple future anchor times can be output based on this model.

[0080] S1-3, Calculate the first load deviation between the total demand-side load and the renewable energy load;

[0081] S1-4. Define the first load deviation as the target load of the thermal power unit at the future anchoring time.

[0082] In this embodiment, the target load acquisition module performs structured processing on the meteorological forecast parameters for future anchoring times, and combines this with a pre-trained renewable energy load forecasting model to effectively obtain the renewable energy load forecasting results for each anchoring time. Furthermore, based on the difference between this forecast value and the total demand-side load, the system accurately calculates the target thermal power load that the thermal power units should bear at each future time.

[0083] For example, in this embodiment, the pre-construction step of the new energy load forecasting model includes:

[0084] H1. Collect meteorological parameters corresponding to several historical time points in the target area;

[0085] The meteorological parameters include, but are not limited to, meteorological elements in multiple dimensions such as wind speed, wind direction, and light intensity;

[0086] H2. Perform feature preprocessing on meteorological parameters to generate preprocessed meteorological parameter features;

[0087] Specifically, data preprocessing includes missing value imputation, outlier removal, and normalization to construct the input variables in the training samples;

[0088] H3. Use historical time points as time indices and associate them with the meteorological parameter characteristics to generate time-series input variables;

[0089] In other words, the input variables consist of historical time indices and corresponding meteorological parameter features, forming a data structure with timestamp identifiers and multidimensional meteorological features, used to characterize time-series change trends;

[0090] H4. Obtain a predefined lag time window, and collect the new energy load corresponding to several historical time points within the lag time window;

[0091] Specifically, the lag time window represents the time interval used to review the impact of historical states on the output at the current or future moment. It is usually set in hours or days and is used to construct a time-series sample structure with memory.

[0092] H5. Define the renewable energy load corresponding to historical time points as time-series target variables;

[0093] H6. Based on the aforementioned time-series input variables and time-series target variables, construct a time-series sample of new energy loads for supervised learning;

[0094] H7. Input the time series samples of new energy load into a general time series regression model for supervised training. After iteration, a new energy load prediction model is generated.

[0095] The general time-series regression model can employ Long Short-Term Memory (LSTM) networks or other supervised regression structures suitable for time series modeling; it learns and models the mapping relationship between meteorological parameters and renewable energy loads within a lag time window to achieve hourly prediction of renewable energy loads at future moments.

[0096] In this embodiment, by constructing a time-series sample structure based on meteorological parameters and historical renewable energy load data, and introducing a regression prediction model with time-series modeling capabilities, the system achieves high-precision hourly prediction of renewable energy load. Specifically, by preprocessing historical meteorological data and constructing lag windows, the model can effectively extract the temporal correlation between meteorological changes and power output; combined with a supervised learning training mechanism, the constructed renewable energy load prediction model has good trend capture and nonlinear fitting capabilities.

[0097] For example, in this embodiment, the load calculation module is specifically used for:

[0098] S4-1. Obtain several control parameters of the thermal power unit at the current moment;

[0099] Several control parameters include: boiler fuel input rate, air supply volume, main steam valve opening, and turbine governor setpoint;

[0100] Specifically, the boiler fuel input rate is used to regulate the main steam generation rate, the air supply affects combustion efficiency and furnace temperature, the main steam valve opening determines the steam flow rate entering the turbine, and the turbine governor setpoint is used to control the turbine target speed, thereby regulating the power generation response. In other words, several control parameters in this embodiment are actively controllable variables in the thermal power unit, directly regulating energy conversion in the system and serving as core parameters for load response control.

[0101] S4-2. Based on several control parameters at the current moment, determine the thermal power output load of the thermal power unit at the future anchoring moment;

[0102] In one embodiment of the present invention, the thermal power output load of the thermal power unit at the future anchoring time is determined based on the fitting relationship between several control parameters at the current time and historical operating data. Specifically, the control parameters include boiler fuel input rate, air supply volume, main steam valve opening degree, and turbine governor setpoint. The historical operating data records the combined states of the above parameters at different times and their corresponding actual output power of the thermal power unit.

[0103] In this implementation, by modeling the mapping relationship between controllable variables and actual load in historical sample data, a thermal power load estimation function can be constructed to describe the output trend corresponding to different parameter combinations.

[0104] In actual execution, the system uses the control parameters acquired at the current moment as input, substitutes them into the estimation function, and thus determines the predetermined load of the thermal power unit at the future anchoring time. This predetermined load represents the power output level that the thermal power system can achieve under natural operation, provided that the current control parameters remain unchanged. Furthermore, the estimation function can be expressed as follows:

[0105]

[0106] Among them, P pred q represents the output load of thermal power plants. f This indicates the boiler fuel input rate, and CV represents the lower heating value of the fuel. This represents the air supply volume under the adjustable coefficient γ, characterizing the nonlinear relationship between the air supply volume and the thermal power output load; a v This indicates the main air valve opening, k1 is the valve opening adjustment coefficient, and v s η represents the turbine governor setting value, k2 is the adjustable coefficient of the governor setting value, and η represents the system conversion efficiency factor, taking into account the overall efficiency of the boiler-turbine-generator.

[0107] Furthermore, this estimation function avoids the boiler fuel input rate q by introducing a logarithmic term log.f and air volume When the value is large, the thermal power output load increases infinitely; after adding the logarithm, it indicates that when both fuel and air supply are very high, the increase in the thermal power output load of the system tends to saturate, reflecting the "diminishing marginal combustion benefits"; the control significance of the logarithmic term in this estimation function is: increasing fuel and air volume results in rapid output growth in the initial stage; as fuel and air supply continue to increase, the output growth slows down, which helps to prevent overheating or overload.

[0108] Furthermore, this estimation function introduces an exponential term exp(-k1·a) v This is used to fit the nonlinear response of the main steam valve opening, expressing that as the main steam valve opening increases, the steam release increases rapidly, but gradually saturates to a certain extent; in real thermal power units, valve opening regulation exhibits an S-shaped response characteristic (i.e., output change is not significant at small openings, the response is fast in the middle section, and saturation is approached at large openings); while The sigmoid function controls the denominator and determines the "release threshold" of the steam action.

[0109] Furthermore, this estimation function introduces the tanh function to reflect the nonlinear effect of the turbine governor setpoint on the thermal power output load. The output range of tanh is (-1, 1), multiplied by a constant and then added by 1, the range is (0, 2), indicating that the larger the governor setpoint, the more significant the output gain, but also the more it tends to saturate, preventing the system from becoming unstable due to excessively high setpoints, while also softly limiting the speed regulation performance.

[0110] For example, in this embodiment, the deviation calculation module is specifically used for:

[0111] S6-1. Obtain the target load and predetermined load of thermal power at the future anchoring time;

[0112] S6-2. Calculate the absolute error between the target load of thermal power and the predetermined load of thermal power;

[0113] S6-3. Take the second square of the absolute error to determine the second load deviation at the future anchoring time.

[0114] In this embodiment, the deviation calculation module compares the target load of thermal power at the future anchoring time with the predetermined load of thermal power point by point, and quantifies it in the form of mean square error, so that the larger deviation contributes more significantly to the overall loss.

[0115] For example, in this embodiment, the target loss construction module is specifically used for:

[0116] S7-1, Predefine several pre-adjustment parameters for future anchoring times;

[0117] The pre-adjustment parameter is used to represent the candidate adjustment values ​​of the control parameters of the thermal power unit at the future anchoring time.

[0118] S7-2. Calculate the parameter deviation between the pre-adjustment parameter and the control parameter;

[0119] The parameter deviation is used to reflect the magnitude of change in real-time control behavior relative to the original state.

[0120] S7-3. Perform regularization on the parameter deviation and construct a parameter change penalty term;

[0121] The regularization process improves the system's stability by introducing adjustment weights to limit the drastic changes in control parameters between consecutive time points.

[0122] S7-4. Take the second load deviation at the future anchoring time as the first loss sub-item and the parameter change penalty term as the second loss sub-item, and jointly construct the target loss function.

[0123] In this embodiment, by introducing a deviation metric between the pre-adjustment parameters and the current control parameters, and combining it with deviation regularization, a unified modeling of control accuracy and adjustment stability is achieved. Specifically, this module constructs a dual loss structure that includes load response error and parameter change penalty terms, enabling the optimization process to suppress drastic parameter jumps while pursuing target load fitting, thereby improving the convergence stability of the control parameter iteration process.

[0124] For example, in this embodiment, the expression for the target loss function is:

[0125]

[0126] Where ε is the target loss, n is the preset adjustment quantity for future anchoring times, and i is the time index. This represents the target load of thermal power at the i-th future anchoring time. Let represent the predetermined load of thermal power determined by control parameters at the i-th future anchoring time, λ be the regularization weighting coefficient used to control the penalty degree for parameter deviations between adjacent times, m be the number of control parameters, j be the parameter index of the controllable parameter, and θ be the parameter index. j Let θ represent the j-th control parameter. j-1 This represents the j-th control variable parameter in the previous control step.

[0127] Specifically, the second load deviation uses mean square error (MSE), which measures the difference between the target load and the predetermined load of the thermal power plant at each anchoring time and is the main driving factor of the control system. Furthermore, MSE is more sensitive to large errors, which is beneficial for parameter convergence optimization. Meanwhile, the parameter change penalty term is constructed by introducing the change between the pre-adjustment parameter and the control parameter in the previous control step to create a regularization penalty term. Therefore, the target loss function is constructed using the above two loss sub-terms to balance output load accuracy and command smoothness.

[0128] For example, in this embodiment, the parameter update module is specifically used for:

[0129] S8-1. Use several control parameters at the current moment as initialization parameters, and set the learning rate and maximum number of iterations;

[0130] S8-2. Based on several control parameters at the current moment, determine the real-time update loss of the target loss function;

[0131] S8-3. Calculate the gradient vector of the real-time update loss relative to the control parameters;

[0132] Specifically, the chain rule or automatic differentiation method can be used to obtain the partial derivative of the loss function with respect to each control parameter, which can be used to guide the direction of parameter updates.

[0133] S8-4. Based on the gradient vector and the set learning rate, update several control parameters at the current time to generate the next round of iteration parameters until the target loss function converges.

[0134] In this embodiment, the parameter update module achieves efficient convergence of the control parameters by introducing a gradient-based iterative optimization strategy. Specifically, by initializing the current control parameters and setting limits on the learning rate and the number of iterations, the module can calculate the gradient vector of the loss function with respect to each control parameter under the guidance of the structured loss function, and perform directional updates accordingly until the target loss function converges.

[0135] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means.

[0136] The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives (SSDs).

[0137] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Furthermore, the mutual couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0138] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application.

Claims

1. A coordinated control system for thermal power units under deep peak shaving, characterized in that, include: The total load acquisition module is used to acquire the total demand-side load at the future anchoring time. The target load acquisition module is used to determine the target load of thermal power units at the future anchoring time based on the total demand-side load. The target load for thermal power is characterized as the first load deviation between the total demand-side load and the renewable energy load. The parameter acquisition module is used to acquire several control parameters of the thermal power unit at the current moment; The load calculation module is used to calculate the thermal power output load at the current moment based on several control parameters at the current moment. The load definition module is used to define the thermal power output load at the current moment as the predetermined thermal power load at the future anchoring moment; The deviation calculation module is used to calculate the second load deviation between the target load of thermal power and the predetermined load of thermal power at the future anchoring time; The target loss construction module is used to construct the target loss function based on several control parameters at the current moment and the second load deviation; The parameter update module is used to update several control parameters in order to minimize the target loss function; The parameter determination module is used to determine whether the target loss function has converged; If convergence is not achieved, return to the load calculation module; otherwise, output the current parameters as the optimal execution parameters for the thermal power unit. The target loss construction module is specifically used for: Several pre-defined pre-adjustment parameters for future anchoring times; Calculate the parameter deviation between the pre-adjustment parameter and the control parameter; Regularization is applied to the parameter deviation to construct a parameter change penalty term; The second load deviation at the future anchoring time is taken as the first loss term, and the parameter change penalty term is taken as the second loss term. The target loss function is constructed by combining them. The expression for the target loss function is: ； in, Let n be the target loss, n be the preset adjustment quantity for future anchoring times, and i be the time index. This represents the target load of thermal power at the i-th future anchoring time. This represents the predetermined load of the thermal power plant determined by the control parameters at the i-th future anchoring time. is the regularization weight coefficient, used to control the degree of penalty for parameter deviation between adjacent time steps, m is the number of control parameters, and j represents the parameter index of the controllable parameter. This represents the j-th control parameter. This indicates the parameter of the j-th control variable in the previous control step.

2. The coordinated control system for thermal power units under deep peak shaving as described in claim 1, characterized in that, The target load acquisition module is specifically used for: Obtain several future anchoring times, and the corresponding meteorological forecast parameters for those future anchoring times; The future anchoring time and the corresponding meteorological forecast parameters are used to construct a time series vector of new energy load, which is then input into a pre-trained new energy load forecasting model to output the new energy load at the future anchoring time. Calculate the first load deviation between the total demand-side load and the renewable energy load; The first load deviation is defined as the target load of the thermal power unit at the future anchoring time.

3. The coordinated control system for thermal power units under deep peak shaving as described in claim 2, characterized in that, The pre-construction steps of the new energy load forecasting model include: H1. Collect meteorological parameters corresponding to several historical time points in the target area; H2. Perform feature preprocessing on meteorological parameters to generate preprocessed meteorological parameter features; H3. Use historical time points as time indices and associate them with the meteorological parameter characteristics to generate time-series input variables; H4. Obtain a predefined lag time window, and collect the new energy load corresponding to several historical time points within the lag time window; H5. Define the renewable energy load corresponding to historical time points as time-series target variables; H6. Based on the aforementioned time-series input variables and time-series target variables, construct a time-series sample of new energy loads for supervised learning; H7. Input the time series samples of new energy load into a general time series regression model for supervised training. After iteration, a new energy load prediction model is generated.

4. The coordinated control system for thermal power units under deep peak shaving as described in claim 1, characterized in that, The load calculation module is specifically used for: Obtain several control parameters of the thermal power unit at the current moment; Several control parameters include: boiler fuel input rate, air supply volume, main steam valve opening, and turbine governor setpoint; Based on several control parameters at the current moment, determine the thermal power output load of the thermal power unit at the future anchoring moment.

5. A coordinated control system for thermal power units under deep peak shaving as described in claim 1, characterized in that, The deviation calculation module is specifically used for: Obtain the target load and the established load of thermal power at the future anchoring time; Calculate the absolute error between the target load of thermal power and the predetermined load of thermal power; The absolute error is squared twice to determine the second load deviation at the future anchoring time.

6. The coordinated control system for thermal power units under deep peak shaving as described in claim 1, characterized in that, The parameter update module is specifically used for: Use the current control parameters as initialization parameters, and set the learning rate and maximum number of iterations; Based on several control parameters at the current moment, determine the real-time updated loss of the target loss function; Calculate the gradient vector of the loss relative to the control parameters in real time; Based on the gradient vector and the set learning rate, several control parameters at the current time are updated to generate the parameters for the next iteration, until the target loss function converges.

Images