An intelligent control method for flexibility and deep peak regulation of a thermal power unit

Abstract

This invention relates to the field of automatic control technology and discloses an intelligent control method for the flexibility and deep peak shaving of thermal power units. Within M time periods in a first preset time period, the following are acquired at fixed time intervals: the output power of N clean energy generating units, the power demand of the power grid, first environmental data of the areas where the N clean energy generating units are located, and second environmental data of the area covered by the power grid. Within a second preset time period, based on a prediction model, a demand prediction model, and an environmental prediction model, the target power of the thermal power units at each moment is determined in the m-th time period. Based on the target power, the thermal power units are automatically controlled to perform deep peak shaving to minimize mechanical damage to the thermal power units.

Description

Technical Field

[0001] This invention relates to the field of automatic control technology, and more specifically, to an intelligent control method for the flexibility and deep peak shaving of thermal power units. Background Technology

[0002] The output of clean energy is highly volatile, intermittent, and uncontrollable, severely impacting the stable operation of the power grid and the regulation capacity of the power system. In the existing power system, thermal power generation, based on its stability, undertakes critical tasks such as base load, peak shaving, frequency regulation, and reserve capacity. Due to the time-sensitive and priority grid connection characteristics of clean energy output, thermal power generating units typically need to dynamically adjust their output power according to changes in clean energy output to achieve grid supply and demand balance. Existing parallel schemes for clean energy and thermal power generation lack systematic control measures regarding the coupling relationship between the operational stability and mechanical losses of thermal power units, making it even more difficult to achieve long-term, flexible, and low-loss operation of thermal power units in a clean energy grid connection environment.

[0003] A method for intelligent control of flexibility and deep peak shaving in thermal power units, disclosed in publication number CN115327910A, includes the following steps: conducting disturbance tests at typical operating points; establishing a gain scheduling model of the coordinated system through fitting and identification methods; establishing a fuzzy rule base; inferring the dynamic mathematical model of the coordinated system corresponding to the grid load command in real time and using it as a prediction model; and using a predictive controller to predict the output trajectory based on the prediction equation, a Kalman filter to estimate the system state, and performance indicators to calculate the optimal control quantity. This method is applicable to the coordinated control system of thermal power units with frequently changing operating conditions.

[0004] This application uses fuzzy control to achieve optimal output of thermal power units through automatic control. However, this method does not consider the operating conditions of the thermal power units, leading to a reduction in their service life. Summary of the Invention

[0005] This invention provides an intelligent control method for the flexibility and deep peak shaving of thermal power units, solving the technical problems mentioned in the background art.

[0006] This invention provides a method for intelligent control of the flexibility and deep peak shaving of thermal power units, including:

[0007] Step 1: Within M time periods of the first preset time period, obtain the following at fixed time intervals:

[0008] The output power of N clean generator sets, the power demand of the power grid, the first environmental data of the area where the N clean generator sets are located, and the second environmental data of the area covered by the power grid.

[0009] Environmental data include: wind speed, wind direction, temperature, air pressure, humidity, and irradiance;

[0010] Step 2: Based on the output power, demand power, first environmental data, and second environmental data, construct the output prediction model, demand prediction model, environmental prediction model, and correlation verification model, respectively.

[0011] Step 3: Within the second preset time period, based on the prediction model, demand prediction model, and environmental prediction model, determine the time period at each moment in the m-th time cycle:

[0012] The predicted output power of N clean generator sets, the predicted power demand of the power grid, the predicted first environmental data of the area where the N clean generator sets are located, and the predicted second environmental data of the area covered by the power grid.

[0013] Step 4: In the m-th time period, determine the target power of the thermal power unit at each moment; based on the target power, automatically control the thermal power unit to perform deep peak shaving to minimize the mechanical damage of the thermal power unit; where 1≤m≤M, and m is a positive integer.

[0014] Furthermore, the output includes a prediction model, a demand prediction model, an environmental prediction model, and a correlation verification model.

[0015] The output prediction model consists of N output prediction units.

[0016] The nth output prediction unit is trained based on the output power of the nth clean generator set in time sequence; where 1≤n≤N, and n is a positive integer.

[0017] The demand forecasting model is trained based on time-series demand power.

[0018] An environmental prediction model, comprising a first prediction unit and a second prediction unit;

[0019] The first prediction unit is trained based on the time-series first environmental data; the second prediction unit is trained based on the time-series second environmental data.

[0020] The correlation verification model includes a first verification unit and a second verification unit:

[0021] Specifically, a first verification unit is trained based on the first environmental data of the areas where N clean generator sets are located and the output power of the N clean generator sets; a second verification unit is trained based on the second environmental data of the power grid coverage area and the power demand of the power grid.

[0022] Furthermore, within each moment of the m-th time period, determine: the predicted output power of the N clean-generating units, the predicted power demand of the power grid, the predicted first environmental data of the region where the N clean-generating units are located, and the predicted second environmental data of the power grid coverage area, including:

[0023] Based on the N output prediction units of the output prediction model, the predicted output power of the N clean generator sets at the i-th moment in the m-th time period is determined respectively.

[0024] Based on the demand forecasting model, the predicted power demand of the power grid at the i-th moment in the m-th time period is determined;

[0025] The first and second prediction units based on the environmental prediction model determine the predicted first environmental data of the area to which N clean generator sets belong and the predicted second environmental data of the power grid coverage area at the i-th time in the m-th time period, respectively.

[0026] The predicted output power of N clean energy generator sets at the i-th time in the m-th time period and the predicted first environmental data of the area where the N clean energy generator sets belong are used to perform the first verification based on the first verification unit of the correlation verification model, specifically including:

[0027] The predicted first environmental data of the area to which the nth clean generator unit belongs is input into the first verification unit to obtain the first verification value; the difference between the first verification value and the predicted output power of the nth clean generator unit is calculated; if the difference is less than or equal to the first verification threshold, the first verification is obtained; otherwise, the hyperparameters of the nth output prediction unit in the output prediction model and the first prediction unit in the environmental prediction model are updated synchronously and backward based on the gradient descent of the difference.

[0028] The predicted power demand of the power grid at time i in the m-th time period and the predicted second environmental data of the power grid coverage area are used for a second verification based on the second verification unit of the correlation verification model. Specifically, this includes:

[0029] The predicted second environmental data of the power grid coverage area is input into the second verification unit to obtain the second verification value; the difference between the second verification value and the predicted power demand of the power grid is calculated; if the difference is less than or equal to the second verification threshold, the second verification is obtained; otherwise, the demand prediction model and the hyperparameters of the second prediction unit in the environmental prediction model are updated synchronously and in reverse based on the gradient descent of the difference.

[0030] Furthermore, the target power of the thermal power unit at each moment is determined, including:

[0031] Calculate the difference between the predicted output power of N thermal power units and the predicted power demand of the power grid at the i-th time point within the m-th time period of the second preset time period; use the difference power at the i-th time point as the target power of the thermal power units at the i-th time point.

[0032] Furthermore, based on the target power, the thermal power units are automatically controlled to perform deep peak shaving to minimize mechanical damage to the thermal power units, including:

[0033] Step 41: Initialize and generate R deep peak shaving schemes that meet the constraints; wherein each deep peak shaving scheme includes a baseline cruise power and a planned power at each time step;

[0034] The constraints include:

[0035] The baseline cruise power is within the power generation range of thermal power units, and the baseline cruise power of any two deep peak shaving schemes is different.

[0036] Thermal power units perform peak shaving tasks at a constant speed or constant acceleration, between the baseline cruise power and the target power.

[0037] At any given moment, the planned power of the thermal power unit is greater than or equal to the target power;

[0038] Step 42: Obtain the objective function value of each deep peak-shaving scheme, and sort them from largest to smallest based on the objective function value to obtain the feature ranking;

[0039] Step 43: Retain a preset number of deep peak-shaving schemes from the feature sorting from front to back, and randomly perform mutation update processing on the remaining deep peak-shaving schemes.

[0040] Step 44: Repeat steps 42 and 43 a preset number of times to obtain the final feature ranking, and obtain the deep peak shaving scheme with the corresponding feature ranking as 1 to control the thermal power unit to perform deep peak shaving.

[0041] Furthermore, the objective function value for each deep peak-shaving scheme is obtained, including:

[0042] Obtain the planned power at each time step in the deep peak shaving scheme and convert it into a two-dimensional graph; where the x-th row and y-th column of the two-dimensional graph represents the power y corresponding to time x.

[0043] Divide the two-dimensional diagram into horizontal and inclined segments, and obtain the inclination rate of each inclined segment;

[0044] Based on the baseline cruise power, the horizontal segment is assigned a first stability weight to calculate the first index;

[0045] Based on the slope rate of the inclined segment, the corresponding inclined segment is assigned a second stable weight to calculate the second index.

[0046] Based on the difference between the planned power and the target power at each time step, a third stable weight is assigned to the difference to calculate the third index.

[0047] The objective function value of the deep peak shaving scheme is calculated based on the first, second, and third indicators.

[0048] Furthermore, the first indicator includes:

[0049] Determine the number of time points N1 corresponding to the horizontal segment, the reference cruise power S, and the first stability weight W1;

[0050] The first indicator includes:

[0051]

[0052] Where F1 represents the first indicator, and ∈ indicates measures to prevent loss. p A constant that is zero, where 0 < ∈ < 1.

[0053] Furthermore, the second indicator includes:

[0054] Determine the vertical direction in the two-dimensional diagram as the reference direction;

[0055] The inclination rate θ of the inclined segment d is determined based on the angle between the inclined segment and the reference direction. d And obtain the number of time points dN2 corresponding to the inclined segment d;

[0056] Based on the slope, the second stable weight dW2 of the slope segment d is determined as follows:

[0057]

[0058] The calculation of the second indicator includes:

[0059]

[0060] Where F2 represents the second index, D represents the number of sloping segments, and d represents the index of D.

[0061] Furthermore, the third indicator includes:

[0062]

[0063] Where F3 represents the third indicator, P represents the number of moments in each time period, p represents the index of P, and Loss... p W3 represents the normalized value of the difference between the planned power and the target power at time p, and W3 represents the third stable weight.

[0064] Furthermore, the objective function value of the deep peak-shaving scheme is calculated based on the first, second, and third indicators, including:

[0065] Fitness = F1 + F2 + F3

[0066] Where Fitness represents the objective function value of the deep peak-shaving scheme.

[0067] The beneficial effects of this invention are as follows: under the premise of ensuring the balance between power grid supply and demand, by conducting multi-model prediction and correlation verification of clean energy output, environmental data and power grid demand, the peak-shaving scheme of thermal power units can be effectively planned, thereby significantly reducing the mechanical wear and energy consumption caused by frequent adjustments of thermal power units, improving the overall flexibility and stability of system operation, extending equipment life, and improving the economic benefits and reliability of the peak-shaving process. Attached Figure Description

[0068] Figure 1 This is a flowchart of an intelligent control method for the flexibility and deep peak shaving of thermal power units according to the present invention. Detailed Implementation

[0069] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0070] like Figure 1 As shown, a method for intelligent control of flexibility and deep peak shaving of thermal power units includes:

[0071] Step 1: Within M time periods of the first preset time period, obtain the following at fixed time intervals:

[0072] The output power of N clean generator sets, the power demand of the power grid, the first environmental data of the area where the N clean generator sets are located, and the second environmental data of the area covered by the power grid.

[0073] Environmental data include: wind speed, wind direction, temperature, air pressure, humidity, and irradiance;

[0074] Step 2: Based on the output power, demand power, first environmental data, and second environmental data, construct the output prediction model, demand prediction model, environmental prediction model, and correlation verification model, respectively.

[0075] Step 3: Within the second preset time period, based on the prediction model, demand prediction model, and environmental prediction model, determine the time period at each moment in the m-th time cycle:

[0076] The predicted output power of N clean generator sets, the predicted power demand of the power grid, the predicted first environmental data of the area where the N clean generator sets are located, and the predicted second environmental data of the area covered by the power grid.

[0077] Step 4: In the m-th time period, determine the target power of the thermal power unit at each moment; based on the target power, automatically control the thermal power unit to perform deep peak shaving to minimize the mechanical damage of the thermal power unit; where 1≤m≤M, and m is a positive integer.

[0078] It should be noted that existing deep peak-shaving schemes plan the output power of thermal power units at each moment to prioritize the use of clean electricity at any given time. However, this leads to erratic changes in the output power of thermal power units, resulting in excessively uneven power output and consequently reducing the lifespan of the units.

[0079] In one embodiment of the present invention, wind speed, wind direction, air temperature, air pressure, humidity, and irradiance are all normalized data. For example, the historical data shows a maximum wind speed of 5 m / s and a minimum wind speed of 1 m / s. When the measured wind speed is 3 m / s, it is normalized to 0.5 based on the maximum and minimum values. Similarly, wind direction, air temperature, air pressure, humidity, and irradiance are processed in the same manner as described above.

[0080] In one embodiment of the present invention, a prediction model, a demand prediction model, an environmental prediction model, and a correlation verification model are output;

[0081] The output prediction model consists of N output prediction units.

[0082] The nth output prediction unit is trained based on the output power of the nth clean generator set in time sequence; where 1≤n≤N, and n is a positive integer.

[0083] The demand forecasting model is trained based on time-series demand power.

[0084] An environmental prediction model, comprising a first prediction unit and a second prediction unit;

[0085] The first prediction unit is trained based on the time-series first environmental data; the second prediction unit is trained based on the time-series second environmental data.

[0086] The correlation verification model includes a first verification unit and a second verification unit:

[0087] Specifically, a first verification unit is trained based on the first environmental data of the areas where N clean generator sets are located and the output power of the N clean generator sets; a second verification unit is trained based on the second environmental data of the power grid coverage area and the power demand of the power grid.

[0088] It should be noted that the architecture and training methods of the output prediction model, demand prediction model, and environmental prediction model are as follows: The output prediction model contains N independent output prediction units, each corresponding to a clean energy generator unit. Based on the time-series output power data of this unit, it is trained using a Long Short-Time Neural Network (LSTM) combined with a sliding window technique. A sliding window of length H extracts the normalized output power of the first H-1 time steps as the input feature vector, and the normalized power at the Hth time step as the sample label. Deep learning is used to capture the time-series dependence features of the unit's output. The demand prediction model is trained using the same LSTM and sliding window method on the time-series power demand data of the power grid. The normalized demand power of the first H-1 time steps is used as the input, and the normalized value at the Hth time step is used as the label to learn the time-series variation pattern of electricity load. The environmental prediction model is divided into two parts: the first prediction unit is trained based on the time-series first environmental data of the region where the clean energy generator unit is located, and the second prediction unit is trained based on the time-series second environmental data of the power grid coverage area. Both use LSTM to model the time-series features of environmental parameters to output predicted values. This modeling method achieves accurate modeling of clean power output, grid demand, and environmental factors through modular design and temporal feature capture.

[0089] In one embodiment of the present invention, the output prediction model, the demand prediction model, and the environment prediction model are all updated in reverse based on the mean squared error loss function to optimize the hyperparameters.

[0090] It should be noted that the correlation verification model is the core module used to explore the nonlinear correlation between environmental data and power data. Built on a one-dimensional convolutional neural network, it consists of a first verification unit and a second verification unit. The first verification unit targets environmental data from the region where the clean-generator units are located. It collects historical synchronous environmental data and corresponding clean-electricity output power as training samples, uses a 1D-CNN to capture local features and nonlinear mappings in the environmental parameter sequence, establishes a theoretical mapping model from the environment to clean-electricity output, and outputs a theoretical verification value of clean-electricity based on environmental data to verify the rationality of the predicted clean-electricity output. The second verification unit targets environmental data from the grid coverage area. Using historical synchronous environmental data and grid demand power as training samples, it analyzes the nonlinear impact of the environment on electricity load using a 1D-CNN to establish a mapping model from the environment to demand power, and outputs a theoretical grid demand verification value to verify the output of the demand prediction model. This model efficiently extracts the temporal features of environmental data through convolutional operations, forming a deviation verification mechanism between the predicted value and the theoretical verification value. When the deviation between the predicted power and the theoretical power driven by the environment exceeds the limit, it triggers the back-optimization of the parameters of the corresponding prediction model.

[0091] In one embodiment of the present invention, determining the predicted output power of N clean-generator units, the predicted power demand of the power grid, the predicted first environmental data of the area where the N clean-generator units are located, and the predicted second environmental data of the area covered by the power grid at each moment in the m-th time period includes:

[0092] Based on the N output prediction units of the output prediction model, the predicted output power of the N clean generator sets at the i-th moment in the m-th time period is determined respectively.

[0093] Based on the demand forecasting model, the predicted power demand of the power grid at the i-th moment in the m-th time period is determined;

[0094] The first and second prediction units based on the environmental prediction model determine the predicted first environmental data of the area to which N clean generator sets belong and the predicted second environmental data of the power grid coverage area at the i-th time in the m-th time period, respectively.

[0095] The predicted output power of N clean energy generator sets at the i-th time in the m-th time period and the predicted first environmental data of the area where the N clean energy generator sets belong are used to perform the first verification based on the first verification unit of the correlation verification model, specifically including:

[0096] The predicted first environmental data of the area to which the nth clean generator unit belongs is input into the first verification unit to obtain the first verification value; the difference between the first verification value and the predicted output power of the nth clean generator unit is calculated; if the difference is less than or equal to the first verification threshold, the first verification is obtained; otherwise, the hyperparameters of the nth output prediction unit in the output prediction model and the first prediction unit in the environmental prediction model are updated synchronously and backward based on the gradient descent of the difference.

[0097] The predicted power demand of the power grid at time i in the m-th time period and the predicted second environmental data of the power grid coverage area are used for a second verification based on the second verification unit of the correlation verification model. Specifically, this includes:

[0098] The predicted second environmental data of the power grid coverage area is input into the second verification unit to obtain the second verification value; the difference between the second verification value and the predicted power demand of the power grid is calculated; if the difference is less than or equal to the second verification threshold, the second verification is obtained; otherwise, the demand prediction model and the hyperparameters of the second prediction unit in the environmental prediction model are updated synchronously and in reverse based on the gradient descent of the difference.

[0099] In one embodiment of the present invention, the multi-model prediction and cross-validation process at time i within the m-th time period achieves accurate verification and dynamic optimization of clean power output, grid demand, and environmental data through two core steps: "prediction value generation" and "bidirectional verification correction," as detailed below:

[0100] Based on the output prediction model, N independent output prediction units are input with the historical time-series output power data of the unit, and output the predicted output power at time i. For example, a photovoltaic unit predicts the power generation at the current time based on the irradiance and power generation sequence of previous time periods.

[0101] The demand forecasting model takes the demand power sequence of previous time steps as input and outputs the predicted demand power at time i, reflecting the expected value of the current grid load.

[0102] First prediction unit: Outputs predicted first environmental data for the regions where N generating units are located, reflecting the direct impact of the local environment on power generation.

[0103] The second prediction unit outputs predicted second environmental data for the power grid coverage area, reflecting the indirect influence of the macro environment on electricity load.

[0104] For each clean generator set, verify whether its predicted output meets the environmental constraints of its region: input the predicted first environmental data of the region to which the nth generator set belongs into the first verification unit, and calculate the theoretical output verification value under the environmental conditions.

[0105] If the verification difference |theoretical verification value - predicted output power| ≤ the first verification threshold (preset reasonable error range), the predicted value is considered valid, and the first verification is completed.

[0106] If the difference exceeds the limit, it indicates that the predicted output contradicts the environmental data, triggering gradient descent synchronous reverse update: simultaneously adjusting the hyperparameters of the nth unit and the first prediction unit of the output prediction model to reduce the deviation between them and ensure that the predicted output conforms to the actual environmental physical laws.

[0107] To verify whether the predicted values ​​of the overall power grid demand conform to the environmental impact patterns of the coverage area:

[0108] The predicted second environmental data of the power grid coverage area is input into the second verification unit to calculate the theoretical demand verification value under the environmental conditions.

[0109] If the verification difference |theoretical verification value - predicted demand power| is less than or equal to the second verification threshold, the demand forecast is considered valid, and the second verification is completed. If the difference exceeds the threshold, it indicates that the demand forecast has not fully reflected the environmental impact, triggering gradient descent synchronous reverse update: simultaneously adjusting the hyperparameters of the demand forecast model and the second forecast unit to make the demand forecast consistent with the theoretical value driven by the environment, avoiding supply and demand imbalance caused by load deviation in the peak shaving scheme.

[0110] By combining "historical data-driven prediction" with "environmental law constraint verification", the predicted values ​​are ensured to conform to the time series change trend and satisfy the physical causal relationship. Especially under abnormal working conditions such as extreme weather, it can effectively correct the prediction bias of a single model.

[0111] The first verification was conducted independently for N clean energy generator sets to adapt to the decentralized nature of distributed energy. The second verification started from the overall grid environment, capturing the common impact of the regional environment on the load, taking into account both individual characteristics and overall patterns.

[0112] Synchronous optimization avoids systematic biases: When validation fails, both the prediction model and the environmental model are updated simultaneously, rather than adjusting a single model in isolation, achieving the synergistic evolution of the "prediction-environment" dual systems. For example, if the photovoltaic output prediction is too high and the irradiance prediction is too low, the light response parameters of the power generation model and the irradiance prediction accuracy of the environmental model are corrected simultaneously, eliminating data contradictions at the source and improving long-term prediction accuracy.

[0113] In one embodiment of the present invention, gradient descent based on the validation difference is used to update the hyperparameters of the output prediction model, the demand prediction model, and the environment prediction model in reverse.

[0114] In one embodiment of the present invention, determining the target power of a thermal power unit at each moment includes:

[0115] Calculate the difference between the predicted output power of N thermal power units and the predicted power demand of the power grid at the i-th time point within the m-th time period of the second preset time period; use the difference power at the i-th time point as the target power of the thermal power units at the i-th time point.

[0116] In one embodiment of the present invention, the target power determination logic for thermal power units at time i is based on the core principle of "prioritizing the consumption of clean electricity and dynamically compensating for the deficit with thermal power". N output prediction units from the output prediction model provide the predicted output power of the N clean power generating units at time i within the m-th time period. The demand prediction model provides the predicted grid demand power at time i, reflecting the expected value of the current grid load. The predicted output of the N clean power generating units is summed to obtain the total clean electricity output at time i. Target power definition: The target power of the thermal power units at time i is the difference between the grid's predicted demand and the total clean electricity output. This difference represents the grid's deficit power, i.e., the power gap that the thermal power units need to compensate for in real time. If the clean electricity output is insufficient, the thermal power units need to increase their power to meet the load demand.

[0117] In one embodiment of the present invention, based on a target power, the thermal power unit is automatically controlled to perform deep peak shaving to minimize mechanical damage to the thermal power unit, including:

[0118] Step 41: Initialize and generate R deep peak shaving schemes that meet the constraints; wherein each deep peak shaving scheme includes a baseline cruise power and a planned power at each time step;

[0119] The constraints include:

[0120] The baseline cruise power is within the power generation range of thermal power units, and the baseline cruise power of any two deep peak shaving schemes is different.

[0121] Thermal power units perform peak shaving tasks at a constant speed or constant acceleration, between the baseline cruise power and the target power.

[0122] At any given moment, the planned power of the thermal power unit is greater than or equal to the target power;

[0123] Step 42: Obtain the objective function value of each deep peak-shaving scheme, and sort them from largest to smallest based on the objective function value to obtain the feature ranking;

[0124] Step 43: Retain a preset number of deep peak-shaving schemes from the feature sorting from front to back, and randomly perform mutation update processing on the remaining deep peak-shaving schemes.

[0125] Step 44: Repeat steps 42 and 43 a preset number of times to obtain the final feature ranking, and obtain the deep peak shaving scheme with the corresponding feature ranking as 1 to control the thermal power unit to perform deep peak shaving.

[0126] It should be noted that in the deep peak-shaving scheme of this invention, the planned power of the thermal power unit at each moment can only be selected from the "baseline cruise power" and the "target power" to ensure the regularity of the adjustment process and the stability of equipment operation. The baseline cruise power is the benchmark value for the stable operation of the thermal power unit, which must be within the safe power generation range of the unit, and the benchmark value is different for different schemes; the target power is the real-time difference between the grid demand and the clean power output, representing the power deficit that the thermal power unit needs to compensate. Specifically, when the target power at a certain moment is less than or equal to the baseline cruise power, the thermal power unit can choose to accurately compensate the grid deficit according to the target power, or maintain stable operation at the baseline cruise power (at this time, the planned power is not lower than the target power, meeting the grid demand); while when the target power is higher than the baseline cruise power, in order to ensure the balance of grid supply and demand, the planned power must be selected as the target power, and the unit needs to adjust from the baseline cruise power to the target power. For example, if the target power is 5, 6, and 7 in sequence within a certain period, and the baseline cruise power is set to 6: At the first moment, the target power of 5 is lower than the baseline value of 6, so the planned power can be either 5 (precise compensation) or 6 (stable operation); at the second moment, the target power equals the baseline value, so the planned power can only be 6 (maintaining stability); at the third moment, the target power of 7 is higher than the baseline value, so the planned power can only be 7 (increasing power to meet demand). Different baseline cruise powers will generate multiple peak-shaving schemes. If the baseline value is set to 7 (higher than the target power of the first two moments), then the planned power can be 7 for the first two moments (stable operation), and at the third moment, since the target power equals the baseline value, it remains at 7. The entire process involves no load changes, minimizing changes in mechanical stress. If the baseline value is set to 5 (lower than the target power of the last two moments), then the power needs to be increased to 6 and 7 sequentially for the last two moments, increasing the number of load changes. By limiting the planned power to switch only between the baseline and target values, the unit's peak-shaving path exhibits a regular variation of "horizontal segment" (stable operation) and "tilted segment" (uniform or uniform acceleration adjustment), avoiding the frequent and irregular fluctuations caused by arbitrary power adjustments in traditional peak-shaving, thereby significantly reducing thermal stress shocks and mechanical fatigue losses in high-temperature components. Finally, by generating multiple schemes with different baseline values ​​during initialization, and combining objective function evaluation and iterative optimization, the optimal scheme with the "longest stable segment and fewest variable loads" is selected, achieving a balance between grid demand response and low-loss equipment operation.

[0127] In one embodiment of the present invention, obtaining the objective function value of each depth peaking scheme includes:

[0128] Obtain the planned power at each time step in the deep peak shaving scheme and convert it into a two-dimensional graph; where the x-th row and y-th column of the two-dimensional graph represents the power y corresponding to time x.

[0129] Divide the two-dimensional diagram into horizontal and inclined segments, and obtain the inclination rate of each inclined segment;

[0130] Based on the baseline cruise power, the horizontal segment is assigned a first stability weight to calculate the first index;

[0131] Based on the slope rate of the inclined segment, the corresponding inclined segment is assigned a second stable weight to calculate the second index.

[0132] Based on the difference between the planned power and the target power at each time step, a third stable weight is assigned to the difference to calculate the third index.

[0133] The objective function value of the deep peak shaving scheme is calculated based on the first, second, and third indicators.

[0134] It should be noted that, in this embodiment of the invention, the objective function value of the deep peak shaving scheme is comprehensively evaluated by quantifying three core indicators: "stability," "smoothness," and "demand matching degree," as follows:

[0135] Two-dimensional diagram construction: The planned power (i.e., the power setpoint of thermal power units at each time) in the peak-shaving scheme is converted into a two-dimensional diagram. The horizontal axis represents time (time x), and the vertical axis represents power (y). Each coordinate point (x, y) represents the planned power y at time x. For example, if the planned power of a certain scheme at times 1-3 is 6, 6, and 7 respectively, then the first two points in the diagram form a horizontal line segment (power remains unchanged), and the third point forms a sloping line segment (power increases).

[0136] In one embodiment of the present invention, the two-dimensional illustration can be constructed based on Excel.

[0137] Horizontal segment: A line segment with a constant planned power over multiple consecutive time periods (such as power stabilizing at the baseline cruise power), reflecting the stable operating state of the unit, with no power changes and minimal mechanical damage.

[0138] Inclined segment: The line segment in which the planned power changes at a constant or constant acceleration over time (such as from the base power 6 to the target power 7), reflecting the unit's load adjustment process. The rate of power change (inclination rate) directly affects the magnitude of mechanical stress.

[0139] In one embodiment of the present invention, the first indicator includes:

[0140] Determine the number of time points N1 corresponding to the horizontal segment, the reference cruise power S, and the first stability weight W1;

[0141] The first indicator includes:

[0142]

[0143] Where F1 represents the first indicator, and ∈ indicates measures to prevent loss.p A constant that is zero, where 0 < ∈ < 1.

[0144] It should be noted that, in this embodiment, the first index is used to quantify the inhibitory effect of the horizontal segment on mechanical damage in the deep peak-shaving scheme, including:

[0145] N1: Represents the duration of the horizontal segment. The larger the N1 value, the longer the thermal power unit can operate stably and the lower the risk of mechanical damage. Therefore, N1 is positively correlated.

[0146] W1 is a manually set weighting coefficient used to reflect the importance of the stable operation of the horizontal segment in the overall objective function. The larger W1 is, the greater the importance attached to the stability of the horizontal segment.

[0147] S: This is the power baseline value for stable operation of the thermal power unit. The smaller S is (within the safe operating range of the unit), The larger the value, the larger F1 is, to encourage thermal power units to operate stably at lower base power, because stable operation at low loads can reduce the risk of power fluctuations at high loads and further reduce mechanical damage.

[0148] ∈: is a constant between 0 and 1. When S approaches 0, ∈ can prevent the formula from failing due to a zero denominator, thus ensuring the mathematical validity of the formula.

[0149] For example, if N1 = 5, W1 = 0.5, S = 10, and ∈ = 0.1, then F1 = 0.248; if S decreases to 5, then F1 = 0.49. It can be seen that, with other conditions remaining constant, the smaller S is, the larger F1 is, reflecting a preference for stable operation at low reference power.

[0150] In one embodiment of the present invention, the second indicator includes:

[0151] Determine the vertical direction in the two-dimensional diagram as the reference direction;

[0152] The inclination rate θ of the inclined segment d is determined based on the angle between the inclined segment and the reference direction. d And obtain the number of time points dN2 corresponding to the inclined segment d;

[0153] Based on the slope, the second stable weight dW2 of the slope segment d is determined as follows:

[0154]

[0155] The calculation of the second indicator includes:

[0156]

[0157] Where F2 represents the second index, D represents the number of sloping segments, and d represents the index of D.

[0158] It should be noted that the vertical direction in the two-dimensional diagram is set as the reference direction, and the angle of the inclined segment is measured with reference to lay the foundation for subsequent analysis of the power change rate.

[0159] θ d The angle is determined by the angle between the inclined segment d and the reference direction (vertical direction). The larger the angle, the faster the power change rate, and the greater the potential stress impact on the mechanical components.

[0160] dN2: Represents the number of moments corresponding to the inclined segment d, that is, the duration of the power change process.

[0161] Calculation of the second stable weight dW2: Formula Used to quantify the relative inclination of the inclined segment.

[0162] When θ d When dW2 = 0 (coinciding with the reference direction, i.e., the horizontal segment), it indicates that there is no power change.

[0163] When θ d =90 (perpendicular to the reference direction, theoretically the power changes instantaneously), dW2=1, but in actual peak shaving, it is constrained (uniform speed or uniform acceleration), and this value is between 0 and 1.

[0164] Second indicator F2 synthesis: Formula This indicates that the effects of all inclined segments are summed up.

[0165] D is the total number of sloping segments, and d is the index (from 1 to D).

[0166] If a certain inclined segment θ d If the load is large (power changes rapidly) and the dN2 is long (duration is long), its contribution to F2 is greater, reflecting that the impact of such "rapid and long-term load change" processes on mechanical damage is more significant in this peak-shaving scheme.

[0167] In one embodiment of the present invention, the third indicator includes:

[0168]

[0169] Where F3 represents the third indicator, P represents the number of moments in each time period, p represents the index of P, and Loss... p W3 represents the normalized value of the difference between the planned power and the target power at time p, and W3 represents the third stable weight.

[0170] It should be noted that the third indicator, F3, is used to evaluate the degree of matching between the planned power and the target power in the deep peak shaving scheme.

[0171] F3: Represents the third indicator. The larger the value, the higher the overall matching degree between the planned power and the target power.

[0172] P: The total number of moments in each time period, where p is the moment index (from 1 to P).

[0173] Loss p : The normalized value of the difference between the planned power and the target power at time p, which reflects the degree of deviation of the power at that time. The smaller the value, the smaller the deviation.

[0174] ∈: A very small constant to prevent loss p When the denominator is zero, the formula ensures the mathematical validity of the calculation.

[0175] For each time p, calculate Then sum up the results from all time points. The smaller the difference between the planned power and the target power (i.e., the lower the loss), the better. p The smaller), The larger the value of F3, the larger the accumulated F3. This means that the larger F3 is, the closer the planned power of the peak-shaving scheme is to the target power at each moment, the more accurate the matching of grid demand, and the less extra losses caused by over-regulation. The third indicator F3 quantitatively evaluates the peak-shaving scheme from the perspective of "demand matching," prompting the planned power to closely follow the target power, reducing ineffective power fluctuations, improving the economy and reliability of peak shaving, and reducing mechanical losses caused by unnecessary regulation. In summary, the third indicator F3, by quantifying the deviation between the planned power and the target power, ensures that the peak-shaving process meets grid demand while minimizing additional mechanical losses.

[0176] In one embodiment of the present invention, calculating the objective function value of the deep peak-shaving scheme based on a first indicator, a second indicator, and a third indicator includes:

[0177] Fitness = F1 + F2 + F3

[0178] Where Fitness represents the objective function value of the deep peak-shaving scheme.

[0179] It should be noted that the objective function value of the deep peak-shaving scheme is calculated by adding three key indicators to minimize mechanical damage and maximize operational efficiency of thermal power units during deep peak-shaving. The first indicator, F1, primarily measures the effect of the horizontal segment (power stability segment) on reducing mechanical damage. The longer the duration of the horizontal segment and the lower the baseline cruise power within a reasonable range, the larger the F1 value, indicating better unit stability and lower risk of mechanical damage. The second indicator, F2, focuses on the tilt segment (power change segment). The smaller the angle between the tilt segment and the baseline direction (vertical direction) (i.e., the smoother the power change), and the more suitable the duration of the tilt segment, the larger the F2 value, representing less stress impact on mechanical components during power regulation. The third indicator, F3, is used to evaluate the matching degree between planned power and target power. The smaller the difference between the planned power and target power at each moment (after normalization, the lower the loss), the better. p The smaller the value, the larger the F3 value, indicating that the peak-shaving scheme responds more accurately to the grid demand and can avoid additional losses caused by excessive or insufficient regulation.

[0180] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A method for intelligent control of flexibility and deep peak shaving in thermal power units, characterized in that, include: Step 1: Within M time periods of the first preset time period, obtain the following at fixed time intervals: The output power of N clean generator sets, the power demand of the power grid, the first environmental data of the area where the N clean generator sets are located, and the second environmental data of the area covered by the power grid. Environmental data include: wind speed, wind direction, temperature, air pressure, humidity, and irradiance; Step 2: Based on the output power, demand power, first environmental data, and second environmental data, construct the output prediction model, demand prediction model, environmental prediction model, and correlation verification model, respectively. Step 3: Within the second preset time period, based on the output prediction model, demand prediction model, and environmental prediction model, determine the time period at each moment in the m-th time cycle: The predicted output power of N clean-generating units, the predicted power demand of the power grid, the predicted first environmental data of the area where the N clean-generating units are located, and the predicted second environmental data of the area covered by the power grid, including: Based on the N output prediction units of the output prediction model, the predicted output power of the N clean generator sets at the i-th moment in the m-th time period is determined respectively. Based on the demand forecasting model, the predicted power demand of the power grid at the i-th moment in the m-th time period is determined; The first and second prediction units based on the environmental prediction model determine the predicted first environmental data of the area to which N clean generator sets belong and the predicted second environmental data of the power grid coverage area at the i-th time in the m-th time period, respectively. The predicted output power of N clean energy generator sets at the i-th time in the m-th time period and the predicted first environmental data of the region to which the N clean energy generator sets belong are used for the first verification based on the first verification unit of the correlation verification model, specifically including: The predicted first environmental data of the area to which the nth clean generator unit belongs is input into the first verification unit to obtain the first verification value; the difference between the first verification value and the predicted output power of the nth clean generator unit is calculated; if the difference is less than or equal to the first verification threshold, the first verification is obtained; otherwise, the hyperparameters of the nth output prediction unit in the output prediction model and the first prediction unit in the environmental prediction model are updated synchronously and backward based on the gradient descent of the difference. The predicted power demand of the power grid at time i in the m-th time period and the predicted second environmental data of the power grid coverage area are used for a second verification based on the second verification unit of the correlation verification model. Specifically, this includes: The predicted second environmental data of the power grid coverage area is input into the second verification unit to obtain the second verification value; the difference between the second verification value and the predicted power demand of the power grid is calculated; if the difference is less than or equal to the second verification threshold, the second verification is obtained; otherwise, the demand prediction model and the hyperparameters of the second prediction unit in the environmental prediction model are updated synchronously and in reverse based on the gradient descent of the difference. Step 4: In the m-th time period, determine the target power of the thermal power unit at each moment; based on the target power, automatically control the thermal power unit to perform deep peak shaving to minimize the mechanical damage of the thermal power unit; where 1≤m≤M, and m is a positive integer.

2. The intelligent control method for flexibility and deep peak shaving of thermal power units according to claim 1, characterized in that, Output prediction models, demand prediction models, environmental prediction models, and correlation verification models; The output prediction model consists of N output prediction units. The nth output prediction unit is trained based on the output power of the nth clean generator set in time sequence; where 1≤n≤N, and n is a positive integer. The demand forecasting model is trained based on time-series demand power. An environmental prediction model, comprising a first prediction unit and a second prediction unit; The first prediction unit is trained based on the time-series first environmental data; the second prediction unit is trained based on the time-series second environmental data. The correlation verification model includes a first verification unit and a second verification unit: Specifically, a first verification unit is trained based on the first environmental data of the areas where N clean generator sets are located and the output power of the N clean generator sets; a second verification unit is trained based on the second environmental data of the power grid coverage area and the power demand of the power grid.

3. The intelligent control method for flexibility and deep peak shaving of thermal power units according to claim 2, characterized in that, Determine the target power of the thermal power unit at each moment, including: Calculate the difference between the predicted output power of N thermal power units and the predicted power demand of the power grid at the i-th time point within the m-th time period of the second preset time period; use the difference power at the i-th time point as the target power of the thermal power units at the i-th time point.

4. The intelligent control method for flexibility and deep peak shaving of thermal power units according to claim 3, characterized in that, Based on the target power, the thermal power units are automatically controlled to perform deep peak shaving to minimize mechanical damage to the units, including: Step 41: Initialize and generate R deep peak shaving schemes that meet the constraints; wherein each deep peak shaving scheme includes a baseline cruise power and a planned power at each time step; The constraints include: The baseline cruise power is within the power generation range of thermal power units, and the baseline cruise power of any two deep peak shaving schemes is different. Thermal power units perform peak shaving tasks at a constant speed or constant acceleration, between the baseline cruise power and the target power. At any given moment, the planned power of the thermal power unit is greater than or equal to the target power; Step 42: Obtain the objective function value of each deep peak-shaving scheme, and sort them from largest to smallest based on the objective function value to obtain the feature ranking; Step 43: Retain a preset number of deep peak-shaving schemes from the feature sorting from front to back, and randomly perform mutation update processing on the remaining deep peak-shaving schemes. Step 44: Repeat steps 42 and 43 a preset number of times to obtain the final feature ranking, and obtain the deep peak shaving scheme with the corresponding feature ranking as 1 to control the thermal power unit to perform deep peak shaving.

5. The intelligent control method for flexibility and deep peak shaving of thermal power units according to claim 4, characterized in that, Obtain the objective function value for each deep peak-shaving scheme, including: Obtain the planned power at each time step in the deep peak shaving scheme and convert it into a two-dimensional graph; where the x-th row and y-th column of the two-dimensional graph represents the power y corresponding to time x. Divide the two-dimensional diagram into horizontal and inclined segments, and obtain the inclination rate of each inclined segment; Based on the baseline cruise power, the horizontal segment is assigned a first stability weight to calculate the first index; Based on the slope rate of the inclined segment, the corresponding inclined segment is assigned a second stable weight to calculate the second index. Based on the difference between the planned power and the target power at each time step, a third stable weight is assigned to the difference to calculate the third index. The objective function value of the deep peak shaving scheme is calculated based on the first, second, and third indicators.

6. The intelligent control method for flexibility and deep peak shaving of thermal power units according to claim 5, characterized in that, The first indicator includes: Determine the number of time points N1 corresponding to the horizontal segment, the reference cruise power S, and the first stability weight W1; The first indicator includes: ; F1 represents the first indicator. Indicated for preventing A constant that is 0 .

7. The intelligent control method for flexibility and deep peak shaving of thermal power units according to claim 6, characterized in that, The second indicator includes: Determine the vertical direction in the two-dimensional diagram as the reference direction; The inclination rate of the inclined segment d is determined based on the angle between the inclined segment and the reference direction. Obtain the number of time points corresponding to the inclined segment d. ; Determine the second stable weight of the inclined segment d based on the slope. ,as follows: ; The calculation of the second indicator includes: ； Where F2 represents the second index, D represents the number of sloping segments, and d represents the index of D.

8. The intelligent control method for flexibility and deep peak shaving of thermal power units according to claim 7, characterized in that, The third indicator includes: ； Where F3 represents the third indicator, P represents the number of moments in each time period, and p represents the index of P. W3 represents the normalized value of the difference between the planned power and the target power at time p, and W3 represents the third stable weight.

9. The intelligent control method for flexibility and deep peak shaving of thermal power units according to claim 8, characterized in that, The objective function value of the deep peak-shaving scheme is calculated based on the first, second, and third indicators, including: ； in, This represents the objective function value of the deep peak-shaving scheme.

Images