A primary frequency regulation prediction system for thermal power units based on deep peak shaving

Abstract

This invention discloses a primary frequency regulation prediction system for thermal power units based on deep peak shaving, belonging to the field of power system and automatic control technology. Its key technical points include the following steps: Step 1: Acquiring historical thermal power unit parameters and regulation control commands from the distributed control system (DCS) and safety instrumented system (SIS); Step 2: Preprocessing the acquired historical data to ensure accuracy; Step 3: Extracting primary frequency regulation data segments from the preprocessed data using short-time Fourier transform (STFT); Step 4: Performing characteristic analysis of primary frequency regulation on the obtained data through time series distribution offset normalization. This system has the advantages of real-time acquisition of thermal power unit operating parameters, focusing on analyzing parameters with higher risk identified by the prediction system, enabling flexible coordination between the monitoring and prediction systems, reducing the load on the monitoring system, minimizing thermal power unit operational failures, and significantly improving system stability.

Description

Technical Field

[0001] This invention relates to the field of power system and automatic control technology, specifically to a primary frequency regulation prediction system for thermal power units based on deep peak shaving. Background Technology

[0002] As the installed capacity of new energy sources continues to climb nationwide, the uncertainty of new energy output exacerbates the difficulty of balancing supply and demand during critical periods of power supply security. Thermal power units across the country will play a more crucial role in providing a safety net and regulating the system during the implementation of the new power system, in order to maintain a high proportion of new energy integration. The primary frequency regulation system of thermal power units based on deep peak shaving mainly relies on background technologies such as automatic generation control and power system stability controllers. However, these technologies face limitations in application, including insufficient response speed, increased equipment wear, high costs, and environmental protection pressures.

[0003] Currently, in some regions, the primary frequency regulation capability of certain thermal power units is significantly reduced during periods of deep peak shaving. Existing primary frequency regulation modeling methods are no longer able to accurately describe the primary frequency regulation characteristics of thermal power units under deep peak shaving conditions, and the models cannot accurately assess the primary frequency regulation capability of the units, resulting in a serious threat to grid frequency security. To overcome these challenges, further research and development are needed to improve system response capabilities, optimize control strategies, reduce operating costs, and ensure that environmental protection requirements are met.

[0004] To address the problems existing in the above-mentioned technologies, there is an urgent need to develop a monitoring and prediction system for the primary frequency regulation performance of thermal power units based on deep peak shaving, so as to improve the dynamic response performance, safety performance and economic performance of the primary frequency regulation of the unit under deep peak shaving conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a primary frequency regulation prediction system for thermal power units based on deep peak shaving, so as to solve the problems mentioned in the background.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A primary frequency regulation prediction system for thermal power units based on deep peak shaving includes the following steps:

[0008] Step 1: Collect historical thermal power unit parameters and control commands from the distributed control system (DCS) and safety instrumented system (SIS);

[0009] Step 2: Perform data preprocessing on the collected historical data to ensure data accuracy;

[0010] Step 3: Extract the primary frequency modulation data segment from the preprocessed data based on the short-time Fourier transform (STFT);

[0011] Step 4: Perform primary frequency modulation characteristic analysis on the obtained primary frequency modulation data segment by normalizing the time series distribution offset;

[0012] The extraction of the primary frequency modulation data segment after data preprocessing in step three includes: extracting the primary frequency modulation data segment through short-time Fourier transform (STFT), where the frequency of the primary frequency modulation signal changes with time, and capturing the change of signal frequency over time;

[0013] The primary frequency modulation signal can be represented as:

[0014] ;

[0015] In the formula, A is the amplitude of the primary frequency modulation signal; The carrier frequency is expressed in Hz. This is the frequency deviation, measured in Hz. The frequency of the frequency modulation signal is expressed in Hz. The variable is time, and the unit is seconds (s).

[0016] The short-time Fourier transform is defined as:

[0017] ;

[0018] In the formula, It is a frequency modulation signal in time and frequency Time-frequency distribution at location; Therefore The STFT is a window function centered on the time axis; it divides the signal into several segments along the time axis, and each segment can be independently Fourier transformed to calculate the spectrum of the signal at a certain moment. This is the kernel function for the Fourier transform;

[0019] Since the frequency of the primary frequency-modulated signal changes with time, the output of the STFT will show the frequency change over time. The instantaneous frequency of the frequency-modulated signal can be expressed as:

[0020] .

[0021] As a further aspect of the present invention, the data preprocessing in step two includes noise removal, which includes using a recursive least squares (RLS) algorithm for each new input data point. and output data .

[0022] As a further aspect of the present invention, the steps of the RLS algorithm are as follows:

[0023] Calculate the gain:

[0024] ;

[0025] In the formula, It is the covariance matrix from the previous step; It is the input item at the current moment. It is a positive forgetting factor, located between 0 and 1, which controls the weight of historical data;

[0026] Calculate the estimation error:

[0027] ;

[0028] In the formula, This is the actual output. It is the estimated value from the previous step;

[0029] Calculate the updated parameters:

[0030] ;

[0031] In the formula, This is the estimated value after noise removal.

[0032] As a further aspect of the present invention, the normalization problem of time series distribution shift in step four includes: adopting the time series distribution shift normalization method to eliminate the impact of fluctuations on data analysis and model building by mapping the data to a fixed interval, reducing the distribution shift between the training set and the test set, and making the data distribution concentrated.

[0033] As a further aspect of the present invention, the time series distribution offset normalization method includes:

[0034] Given a set of data To obtain the corresponding output results Assuming , , These represent the number of variables, the length of the input sequence, and the model prediction length, respectively. Input data The mean and variance are respectively:

[0035] ;

[0036] ;

[0037] Next, the input data is normalized using the formula shown below:

[0038] ;

[0039] In the formula, The data is after normalization; It is a positive number that is introduced;

[0040] Because the proportions of data biased towards the maximum and minimum values ​​are not equal across different datasets, scaling is needed to control this bias. The scaling operation is performed using the formula shown below:

[0041] ;

[0042] In the formula, The data values ​​are scaled up. A coefficient to control scaling bias; The minimum value in the normalized data; The maximum value in the original data;

[0043] After time series distribution shift normalization, some stationary information is removed from the data, resulting in a more concentrated distribution and improving model performance and convergence speed. The final prediction model output is:

[0044] .

[0045] As a further aspect of the present invention, in step four, the primary frequency regulation characteristic analysis of the obtained primary frequency regulation data segment is performed through time series distribution offset normalization, including measuring the primary frequency regulation performance indicators of the unit, as follows:

[0046] The performance indicators for primary frequency regulation of the generating unit are as follows. Assuming the evaluation starts at time t0 and ends at time t1, the actual integral power contribution from primary frequency regulation is:

[0047] ;

[0048] In the formula, This is the initial output for primary frequency modulation; The actual power output at time t during the frequency regulation period of the unit;

[0049] The theoretical contribution of the integral charge from primary frequency modulation is:

[0050] ;

[0051] ;

[0052] In the formula, This represents the adjustment amount corresponding to the theoretical output of the unit at time t during the unit's frequency regulation duration. The rated frequency of the unit; This refers to the rated power of the unit. This is the theoretical setting value for the rotational speed inequality.

[0053] For thermal power units, considering the frequency deviation at time t during the unit's frequency regulation duration, taking into account the frequency regulation dead zone:

[0054] when hour, ;

[0055] when hour, ;

[0056] The first frequency modulation performance pass rate is:

[0057] ;

[0058] The average monthly pass rate for primary frequency regulation of grid-connected thermal and gas-fired power units shall not be less than 70%.

[0059] In summary, the embodiments of the present invention have the following beneficial effects compared with the prior art:

[0060] A primary frequency regulation performance monitoring system for thermal power units based on deep peak shaving can monitor the unit's frequency regulation response capability in real time, ensuring that it can quickly and accurately regulate the frequency under deep peak shaving conditions, avoiding grid instability or equipment failure due to excessive frequency deviation. The requirements of deep peak shaving necessitate frequent starts and stops of thermal power units under conditions of large load fluctuations, placing higher demands on the unit's performance and lifespan. Through the primary frequency regulation performance monitoring system, the unit's frequency regulation capability can be evaluated in real time, its response speed and accuracy analyzed, and potential problems in deep peak shaving identified promptly. This improves the unit's frequency regulation efficiency, reduces unnecessary mechanical wear and energy waste, and extends the unit's service life.

[0061] The primary frequency regulation performance monitoring system for thermal power units based on deep peak shaving can perform real-time diagnosis of the unit's frequency regulation performance and issue timely warnings when anomalies occur. For example, if the system detects that the unit's response time is too long or the power adjustment range is insufficient, it can identify potential fault hazards in advance, avoid large-scale shutdowns or grid instability caused by equipment failures, and reduce economic losses and system risks.

[0062] To more clearly illustrate the structural features and effects of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0063] Figure 1 This is a working framework diagram of a monitoring and prediction system for the primary frequency regulation performance of thermal power units based on deep peak shaving.

[0064] Figure 2 A framework diagram for data acquisition, processing, and analysis of a primary frequency regulation performance prediction system for thermal power units with deep peak shaving.

[0065] Figure 3 A flowchart of data acquisition for a primary frequency regulation performance prediction system for thermal power units with deep peak shaving. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0067] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0068] In one embodiment, a primary frequency regulation prediction system for thermal power units based on deep peak shaving is described in [reference needed]. Figures 1-3 This includes the following steps:

[0069] Step 1: Collect historical thermal power unit parameters and control commands from the distributed control system (DCS) and safety instrumented system (SIS);

[0070] Step 2: Perform data preprocessing on the collected historical data to ensure data accuracy;

[0071] Step 3: Extract the primary frequency modulation data segment from the preprocessed data based on the short-time Fourier transform (STFT);

[0072] Step 4: Perform characteristic analysis on the primary frequency modulation using the obtained data through time series distribution offset normalization.

[0073] Further, see Figures 1-3 The data preprocessing in step two includes noise reduction, which involves using a recursive least squares (RLS) algorithm for each new input data point. and output data .

[0074] Further, see Figures 1-3 The steps of the RLS algorithm are as follows:

[0075] Calculate the gain:

[0076] ;

[0077] In the formula, It is the covariance matrix from the previous step; It is the input item at the current moment. It is a positive forgetting factor, located between 0 and 1, which controls the weight of historical data;

[0078] Calculate the estimation error:

[0079] ;

[0080] In the formula, This is the actual output. It is the estimated value from the previous step;

[0081] Calculate the updated parameters:

[0082] ;

[0083] In the formula, This is the estimated value after noise removal.

[0084] Further, see Figures 1-3 The extraction of the primary frequency modulation data segment after data preprocessing in step three includes: extracting the primary frequency modulation data segment through short-time Fourier transform (STFT), where the frequency of the primary frequency modulation signal changes with time, and capturing the change of signal frequency over time.

[0085] Further, see Figures 1-3 The primary frequency modulation signal can be expressed as:

[0086] ;

[0087] In the formula, A is the amplitude of the primary frequency modulation signal; The carrier frequency is expressed in Hz. This is the frequency deviation, measured in Hz. The frequency of the frequency modulation signal is expressed in Hz. The variable is time, and the unit is seconds (s).

[0088] The short-time Fourier transform is defined as:

[0089] ;

[0090] In the formula, It is a frequency modulation signal in time and frequency Time-frequency distribution at location; Therefore The STFT is a window function centered on the time axis; it divides the signal into several segments along the time axis, and each segment can be independently Fourier transformed to calculate the spectrum of the signal at a certain moment. This is the kernel function for the Fourier transform;

[0091] Since the frequency of the primary frequency-modulated signal changes with time, the output of the STFT will show the frequency change over time. The instantaneous frequency of the frequency-modulated signal can be expressed as:

[0092] .

[0093] Further, see Figures 1-3 The normalization problem for handling time series distribution shift in step four includes: using the time series distribution shift normalization method to eliminate the impact of fluctuations on data analysis and model building by mapping the data to a fixed interval, reducing the distribution shift between the training set and the test set, and making the data distribution more concentrated.

[0094] Further, see Figures 1-3 The time series distribution shift normalization method includes:

[0095] Given a set of data To obtain the corresponding output results Assuming , , These represent the number of variables, the length of the input sequence, and the model prediction length, respectively. Input data The mean and variance are respectively:

[0096] ;

[0097] ;

[0098] Next, the input data is normalized using the formula shown below:

[0099] ;

[0100] In the formula, The data is after normalization; It is a positive number that is introduced;

[0101] Because the proportions of data biased towards the maximum and minimum values ​​are not equal across different datasets, scaling is needed to control this bias. The scaling operation is performed using the formula shown below:

[0102] ;

[0103] In the formula, The data values ​​are scaled up. A coefficient to control scaling bias; The minimum value in the normalized data; The maximum value in the original data;

[0104] After time series distribution shift normalization, some stationary information is removed from the data, resulting in a more concentrated distribution and improving model performance and convergence speed. The final prediction model output is:

[0105] .

[0106] Further, see Figures 1-3 As shown in the figure, step four, which measures the primary frequency regulation performance of the unit, includes the following indicators:

[0107] The performance indicators for primary frequency regulation of the generating unit are as follows. Assuming the evaluation starts at time t0 and ends at time t1, the actual integral power contribution from primary frequency regulation is:

[0108] ;

[0109] In the formula, This is the initial output for primary frequency modulation; The actual power output at time t during the frequency regulation period of the unit;

[0110] The theoretical contribution of the integral charge from primary frequency modulation is:

[0111] ;

[0112] ;

[0113] In the formula, This represents the adjustment amount corresponding to the theoretical output of the unit at time t during the unit's frequency regulation duration. The rated frequency of the unit; This refers to the rated power of the unit. This is the theoretical setting value for the rotational speed inequality.

[0114] For thermal power units, considering the frequency deviation at time t during the unit's frequency regulation duration, taking into account the frequency regulation dead zone:

[0115] when hour, ;

[0116] when hour, ;

[0117] The first frequency modulation performance pass rate is:

[0118] ;

[0119] The average monthly pass rate for primary frequency regulation of grid-connected thermal and gas-fired power units shall not be less than 70%.

[0120] In this embodiment, the thermal power unit prediction system analyzes historical data, including unit power. Main steam flow rate Main steam pressure Main steam temperature Steam turbine speed , regulating valve opening The system uses parameters such as adjustment system commands to predict potential faults and risks, identifying key parameters for monitoring. The monitoring system analyzes data in real time and issues timely alarms when indicators deviate from normal ranges. For different situations occurring in thermal power units during primary frequency regulation, the system collects operating parameters in real time, focusing on analyzing parameters identified as high-risk by the prediction system. This allows the monitoring and prediction systems to work flexibly together, reducing the load on the monitoring system and minimizing operational failures in thermal power units.

[0121] With the transformation of the energy structure and the development of smart grids, optimizing deep peak shaving and unit frequency regulation performance has become an urgent issue for the power industry. Data obtained through monitoring systems can provide policymakers and technicians with detailed references, supporting scientific decision-making and technological innovation in power system operation. For example, historical data and trend analysis based on frequency regulation performance can help assess future adjustments to unit investment, operating modes, and dispatching methods.

[0122] A primary frequency regulation performance monitoring system typically includes the following components: data collection, data processing and analysis, and feedback and control.

[0123] Data collection primarily involves gathering real-time data from thermal power units, including various operational data, frequency data, and load data. Based on the prediction results of the forecasting system, the frequency, power, response time, regulation depth, and control signals of the thermal power units are monitored.

[0124] Data processing and analysis include frequency deviation analysis, power change response analysis, and response time assessment.

[0125] Feedback and control mainly include feedback of frequency regulation parameters, maintenance and adjustment of thermal power units.

[0126] The deep peak-shaving thermal power unit primary frequency regulation performance prediction system includes the following steps:

[0127] 1. Collect historical thermal power unit parameters and regulation and control commands from the distributed control system (DCS) and safety instrumented system (SIS).

[0128] 2. Perform data preprocessing on the collected historical data to ensure data accuracy.

[0129] 3. From the preprocessed data, extract the primary frequency modulation data segment based on the short-time Fourier transform (STFT).

[0130] 4. Perform characteristic analysis of primary frequency modulation on the obtained data by normalizing the time series distribution offset.

[0131] To facilitate the setting of deep peak-shaving targets, a thorough analysis of the operating status, unit characteristics, and grid demand of thermal power plants is first required. This involves determining the basic information of the thermal power plant units, such as their type, capacity, and efficiency, as well as their operating characteristics under different loads. Based on historical load data from the grid, peak and off-peak periods are identified, and the grid's peak-shaving demand during different time periods is determined.

[0132] The peak-shaving demand of thermal power units is as follows:

[0133]

[0134] In the formula, The highest point of the grid load, kW;

[0135] The lowest point of the grid load is represented by kW.

[0136] Peak shaving depth of thermal power units is:

[0137]

[0138] In the formula, The maximum power output that the unit can provide, in kW;

[0139] The minimum power output that the unit can provide, in kW.

[0140] When thermal power units are operating stably, the relationship between peak-shaving demand and peak-shaving depth needs to be satisfied is as follows:

[0141]

[0142] The DCS (Distributed Control System) is the core automation system used to control the operation of thermal power units, primarily responsible for the control and monitoring of various processes within the power plant. A key feature of the DCS system is its use of distributed control units for data acquisition, processing, analysis, and control. It collects operational data from various devices within the thermal power unit, including unit power (P) and main steam flow (F), through sensors, actuators, and other equipment. S Main steam pressure p S Main steam temperature T S Steam turbine speed n tThe system includes data such as valve opening (v), load changes, output adjustments, and operational characteristics. It also includes the control commands issued by the DSC system when problems occur in the operation of the thermal power unit. The SIS system is a dedicated system designed to ensure the safe operation of thermal power units. Its core purpose is to execute safety measures such as emergency shutdown, power cut-off, and valve closure in the event of an emergency or abnormal situation, protecting the safety of the unit and personnel. The historical data required by the predictive system is obtained through the DCS system and the SIS system.

[0143] During the data preprocessing stage, the time intervals between the data collection may not be exactly the same. If it is necessary to align these data to the same timestamp, or to supplement data from certain missing time points, interpolation methods can be used to ensure synchronization between the data points and ensure that all data points can be compared and analyzed on the same time base.

[0144] The linear interpolation formula is as follows:

[0145] ;

[0146] In the formula,

[0147] , These are two known data points;

[0148] These are the time points that need interpolation;

[0149] At a certain point in time The estimated value at that location.

[0150] In thermal power units, regression models can be used to handle missing values ​​in historical data. By establishing relationships between data points, regression models can predict missing values. Regression methods can not only predict based on time series data but also fill in missing data by utilizing relationships between multiple related variables. By establishing regression models to handle missing values, the integrity and usability of the data are ensured.

[0151] Noise removal is achieved using the Recursive Least Squares (RLS) algorithm for each new input data point. and output data The steps of the RLS algorithm are as follows:

[0152] Calculate the gain:

[0153] ;

[0154] In the formula, It is the covariance matrix from the previous step; This is the input item at the current moment; It is a positive forgetting factor, usually between 0 and 1, which controls the weight of historical data.

[0155] Calculate the estimation error:

[0156] ;

[0157] in, This is the actual output; It is the estimate from the previous step.

[0158] Calculate the updated parameters:

[0159] ;

[0160] In the formula, This is the estimated value after noise removal.

[0161] After preprocessing the historical data, the primary frequency modulation (FM) data segment is extracted using the Short-Time Fourier Transform (STFT). The frequency of a FM signal is time-dependent, meaning it changes over time. When using the Fourier Transform, the frequency components of the entire signal are averaged, failing to reflect this time-varying frequency information. In contrast, the STFT, by analyzing within different time windows, can capture the frequency changes of the signal over time.

[0162] A single-frequency modulation signal can be represented as:

[0163] ;

[0164] In the formula, A is the amplitude of the primary frequency modulation signal; The carrier frequency is expressed in Hz. This is the frequency deviation, measured in Hz. The frequency of the frequency modulation signal is expressed in Hz. This is a time variable, with the unit being seconds (s).

[0165] The short-time Fourier transform is defined as:

[0166] ;

[0167] In the formula, It is a frequency modulation signal in time and frequency Time-frequency distribution at location; Therefore The STFT is a window function centered on the time axis; it divides the signal into several segments along the time axis, and each segment can be independently Fourier transformed to calculate the spectrum of the signal at a certain moment. This is the kernel function for the Fourier transform.

[0168] Since the frequency of the primary frequency modulation signal changes over time, the output of the STFT will show the frequency change over time.

[0169] The instantaneous frequency of a frequency-modulated signal can be expressed as:

[0170] ;

[0171] Time series distribution shift refers to the phenomenon where the statistical properties or distribution of time series data change over time, causing models or analytical methods to become ineffective in predicting or inferring from new data. In time series analysis, it is often assumed that the statistical properties of the data, such as the mean and variance, remain stable throughout the observation period. However, when these properties shift, distribution shift occurs.

[0172] When analyzing time-series data of thermal power units, data distribution offset may be encountered. Common causes include:

[0173] 1. Factors such as different seasons, weather conditions, and holidays cause changes in data such as load, temperature, and turbine speed.

[0174] 2. Inconsistency in time series data caused by system load fluctuations.

[0175] 3. Differences in the operating characteristics of thermal power units at different time periods.

[0176] Without addressing these distribution shifts, models can be affected by data fluctuations, leading to inaccurate predictions or optimizations. Normalization methods can help solve this problem. This patent uses time series distribution shift normalization, which eliminates the impact of such fluctuations on data analysis and model building by mapping data to a fixed interval. This method reduces the distribution shift between the training and test sets and makes the data distribution more concentrated. At the same time, this method has fewer parameters and a more easily understood principle.

[0177] Given a set of data To obtain the corresponding output results Assuming , , These represent the number of variables, the length of the input sequence, and the length of the model prediction, respectively.

[0178] Input data The mean and variance are respectively:

[0179] ;

[0180] ;

[0181] Next, the input data is normalized using the formula shown below:

[0182] ;

[0183] In the formula: The data is after normalization; It is a small positive number that is introduced to prevent the denominator from being 0 during calculation.

[0184] Because the proportions of data biased towards the maximum and minimum values ​​are not equal across different datasets, scaling is needed to control this bias. The scaling operation is performed using the formula shown below:

[0185] ;

[0186] In the formula, The data values ​​are scaled up. A coefficient to control scaling bias; The minimum value in the normalized data; This represents the maximum value in the original data.

[0187] After time series distribution shift normalization, some stationary information is removed from the data, resulting in a more concentrated distribution and improving model performance and convergence speed. The final prediction model output is:

[0188] ;

[0189] The performance indicators for primary frequency regulation of the generating unit are as follows. Assuming the evaluation starts at time t0 and ends at time t1, the actual integral power contribution from primary frequency regulation is:

[0190] ;

[0191] In the formula, This is the initial output for primary frequency modulation; The actual output at time t during the frequency regulation period of the unit.

[0192] The theoretical contribution of the integral charge from primary frequency modulation is:

[0193] ;

[0194] ;

[0195] In the formula; This represents the adjustment amount corresponding to the theoretical output of the unit at time t during the unit's frequency regulation duration. The rated frequency of the unit; This refers to the rated power of the unit. This is the theoretical setting value for the rotational speed inequality.

[0196] For thermal power units, considering the frequency deviation at time t during the unit's frequency regulation duration, taking into account the frequency regulation dead zone:

[0197] when hour, ;

[0198] when hour, ;

[0199] The first frequency modulation performance pass rate is:

[0200] ;

[0201] The relevant detailed rules require that the monthly average pass rate of primary frequency regulation for grid-connected thermal power and gas-fired power units shall not be less than 70%.

[0202] Frequency stability of the power grid is crucial for ensuring the safe operation of the power system. Primary frequency regulation is the first line of defense against frequency fluctuations in the power grid. When the power grid frequency deviates, thermal power units respond quickly by adjusting their output power, thereby helping the power grid to restore stability.

[0203] In response to different situations that may occur in thermal power units during primary frequency regulation, the operating parameters of the thermal power units are collected in real time. The focus is on analyzing and predicting parameters that are considered to be at higher risk. This allows the monitoring and prediction systems to work flexibly together, reducing the load on the monitoring system, thereby reducing the number of operational failures of the thermal power units and significantly improving the stability of the system.

[0204] like Figure 1 The diagram shows the working framework of the monitoring and prediction system for the primary frequency regulation performance of thermal power units based on deep peak shaving. The prediction system collects historical parameters, analyzes them to determine the most critical parameters for the current stage, transmits the signals to the monitoring system, and establishes equipment files. The monitoring system collects real-time data, monitors the most critical parameters, and evaluates the primary frequency regulation performance of the thermal power unit.

[0205] The data acquisition, processing, and analysis framework of a primary frequency regulation performance prediction system for thermal power units, such as... Figure 2 As shown:

[0206] (1) Collect historical thermal power unit parameters and regulation and control commands from the distributed control system (DCS) and safety instrument system (SIS).

[0207] (2) Perform data preprocessing on the collected historical data to ensure the accuracy of the data.

[0208] (3) In the preprocessed data, the first frequency modulation data segment is extracted according to the short time Fourier transform (STFT).

[0209] (4) Perform characteristic analysis of the primary frequency modulation based on the obtained data by time series distribution offset normalization.

[0210] First, a thorough analysis of the thermal power plant's operating status, unit characteristics, and grid demand is required to set a deep peak-shaving target. This involves determining the basic information of the thermal power plant units, such as their type, capacity, and efficiency, as well as their operating characteristics under different loads, to determine whether deep peak-shaving is necessary. Based on historical grid load data, peak and off-peak periods are identified to determine the grid's peak-shaving demand in different time periods.

[0211] The peak-shaving demand of thermal power units is as follows:

[0212] ;

[0213] Peak shaving depth of thermal power units is:

[0214] ;

[0215] When thermal power units are operating stably, the relationship between peak-shaving demand and peak-shaving depth needs to be satisfied is as follows:

[0216] ;

[0217] The data acquisition flowchart of the primary frequency regulation performance prediction system for deep peak shaving thermal power units is as follows: Figure 3 As shown, the historical data of the thermal power unit at time t is read at the beginning. The primary frequency regulation characteristics of the thermal power unit are analyzed, and the primary frequency regulation margin of a single unit and all thermal power units is calculated. This margin is compared with the calculated primary frequency regulation threshold of the thermal power unit. If the margin is greater than the threshold, it is recorded as a safe operating condition parameter; if the margin is less than or equal to the threshold, an early warning signal is generated, and the self-adjustment time is analyzed. If the self-adjustment time is less than or equal to 60s, it is recorded as a safe operating condition parameter; otherwise, it is recorded as a dangerous operating condition parameter. After outputting the data, the data at time t+1 is analyzed.

[0218] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A primary frequency regulation prediction system for thermal power units based on deep peak shaving, characterized in that, Includes the following steps: Step 1: Collect historical thermal power unit parameters and control commands from the Distributed Control System (DCS) and Safety Instrumented System (SIS); Step 2: Perform data preprocessing on the collected historical data to ensure data accuracy; Step 3: Extract the first frequency modulation data segment from the preprocessed data based on the Short Time Fourier Transform (STFT); Step 4: Perform primary frequency modulation characteristic analysis on the obtained primary frequency modulation data segment by normalizing the time series distribution offset; The extraction of the primary frequency modulation data segment after data preprocessing in step three includes: extracting the primary frequency modulation data segment through short-time Fourier transform (STFT), where the frequency of the primary frequency modulation signal changes with time, and capturing the change of signal frequency over time; The primary frequency modulation signal can be represented as: ； In the formula, A is the amplitude of the primary frequency modulation signal; The carrier frequency is expressed in Hz. This represents the frequency deviation, measured in Hz. The frequency of the frequency modulation signal is expressed in Hz. The variable is time, and the unit is seconds (s). The short-time Fourier transform is defined as: ； In the formula, It is a frequency modulation signal in time and frequency Time-frequency distribution at location; Therefore The STFT is a window function centered on the time axis; it divides the signal into several segments along the time axis, and each segment can be independently Fourier transformed to calculate the spectrum of the signal at a certain moment. This is the kernel function for the Fourier transform; Since the frequency of the primary frequency-modulated signal changes with time, the output of the STFT will show the frequency change over time. The instantaneous frequency of the frequency-modulated signal can be expressed as: 。 2. The primary frequency regulation prediction system for thermal power units based on deep peak shaving according to claim 1, characterized in that, The data preprocessing in step two includes noise removal, which is achieved using the Recursive Least Squares (RLS) algorithm for each new input data point. and output data .

3. The primary frequency regulation prediction system for thermal power units based on deep peak shaving according to claim 2, characterized in that, The steps of the RLS algorithm are as follows: Calculate the gain: ； In the formula, It is the covariance matrix from the previous step; It is the input item at the current moment. It is a positive forgetting factor, located between 0 and 1, which controls the weight of historical data; Calculate the estimation error: ； In the formula, This is the actual output. It is the estimated value from the previous step; Calculate the updated parameters: ； In the formula, This is the estimated value after noise removal.

4. The primary frequency regulation prediction system for thermal power units based on deep peak shaving according to claim 1, characterized in that, The normalization problem for handling time series distribution shift in step four includes: using the time series distribution shift normalization method to eliminate the impact of fluctuations on data analysis and model building by mapping the data to a fixed interval, reducing the distribution shift between the training set and the test set, and making the data distribution more concentrated.

5. The primary frequency regulation prediction system for thermal power units based on deep peak shaving according to claim 4, characterized in that, The time series distribution shift normalization method includes: Given a set of data To obtain the corresponding output results Assuming , , These represent the number of variables, the length of the input sequence, and the model prediction length, respectively. Input data The mean and variance are respectively: ； ； Next, the input data is normalized using the formula shown below: ； In the formula, The data is after normalization; It is a positive number that is introduced; Because the proportions of data biased towards the maximum and minimum values ​​are not equal across different datasets, scaling is needed to control this bias. The scaling operation is performed using the formula shown below: ； In the formula, The data values ​​are scaled up. A coefficient to control scaling bias; The minimum value in the normalized data; The maximum value in the original data; After time series distribution shift normalization, some stationary information is removed from the data, resulting in a more concentrated distribution. This improves the model's performance and convergence speed, and the final predicted model output value is: 。 6. The primary frequency regulation prediction system for thermal power units based on deep peak shaving according to claim 4, characterized in that, In step four, the primary frequency regulation characteristics analysis of the obtained primary frequency regulation data segment is performed through time series distribution offset normalization, including measuring the primary frequency regulation performance indicators of the unit, as follows: The performance indicators for primary frequency regulation of the generating unit are as follows. Assuming the evaluation starts at time t0 and ends at time t1, the actual integral power contribution from primary frequency regulation is: ； In the formula, This is the initial output for primary frequency modulation; The actual power output at time t during the frequency regulation period of the unit; The theoretical contribution of the integral charge from primary frequency modulation is: ； ； In the formula, This represents the adjustment amount corresponding to the theoretical output of the unit at time t during the unit's frequency regulation duration. The rated frequency of the unit; This refers to the rated power of the unit. This is the theoretical setting value for the rotational speed inequality. For thermal power units, considering the frequency deviation at time t during the unit's frequency regulation duration, taking into account the frequency regulation dead zone: when hour, ; when hour, ; The first frequency modulation performance pass rate is: ； The average monthly pass rate for primary frequency regulation of grid-connected thermal and gas-fired power units shall not be less than 70%.

Images