A deep peak load rate optimization method based on a high-alkali coal combustion unit

Abstract

The application discloses a kind of based on the deep peak shaving load rate optimization method of high alkali coal unit combustion, specifically related to power plant peak shaving optimization field, including acquisition high alkali coal unit operation data and historical instruction, excavate the influence law of high alkali coal combustion characteristics to unit dynamic response, dynamic load distribution is carried out in combination with each unit response margin, and according to output deviation generates compensation instruction.A kind of based on the deep peak shaving load rate optimization method of high alkali coal unit combustion by using association rule algorithm quantifies the restriction relationship of high alkali coal combustion characteristics to unit dynamic response, provides data basis and constraint condition for subsequent optimization;Through the dynamic mapping model constructed, accurate prediction basis is provided for subsequent instruction optimization;Through rolling optimization technology, while ensuring tracking grid instruction, precise matching of control instruction and unit actual capacity is realized, effectively solve the response deviation problem, improve grid frequency tracking accuracy and stability.

Description

Technical Field

[0001] This invention relates to the field of power plant peak shaving optimization technology, and more specifically, to a method for optimizing the deep peak shaving load rate based on high-alkali coal-fired power units. Background Technology

[0002] With the large-scale grid connection of new energy sources, frequency stability control, cross-regional load balancing, and optimal allocation of frequency regulation resources on the grid side have become core technical challenges. Coordinated control of grid dispatch and power plant peak-shaving response has become crucial to ensuring power supply reliability. Traditional technologies, based on real-time load gaps and frequency deviations across the entire region, issue unified load regulation commands to various coal-fired power plants. Power plants collect basic operating parameters of their units and feed them back to the dispatching end to meet the basic requirements of grid load balancing. However, there are significant limitations: the dispatch commands do not consider the impact of coal type characteristics and unit response capability differences on the accuracy of grid frequency regulation, and the dispatching logic relies solely on real-time load data for passive response. Especially when high-alkali coal power plants participate in frequency regulation, the problem of unit response lag and mismatch between grid dispatching needs further exacerbates the instability of grid frequency.

[0003] To address the dynamic adaptation challenges of traditional technologies, existing improved technologies incorporate advanced algorithms such as model predictive control and fuzzy control. Simultaneously, they optimize the bidirectional communication architecture between the power grid and power plants, employing a distributed data transmission mode to enhance the real-time interaction of key information such as load commands and frequency deviations. Compared to traditional technologies, these improved technologies overcome the limitations of centralized passive dispatching of the power grid, constructing a semi-closed-loop collaborative system. This significantly improves the power grid's adaptability to new energy fluctuations and reduces the amplitude of power grid frequency fluctuations when conventional coal-fired power plants participate in frequency regulation.

[0004] However, in practical use, it still has some shortcomings. For example, the parameter range and command allocation logic of its scheduling model are not optimized for the special attributes of high-alkali coal power plants. When the power grid issues frequency regulation commands for rapid load increases and decreases, the actual response capability of high-alkali coal power plants deviates from the grid scheduling expectations, resulting in an imbalance between supply and demand of frequency regulation resources in the region and insufficient tracking accuracy and stability of the power grid frequency. Especially in scenarios with high penetration of new energy, the randomness and suddenness of power grid load fluctuations are significantly enhanced, making it difficult to accurately match the dynamic response characteristics of high-alkali coal power plants. Consequently, the accuracy and reliability of the power grid frequency regulation service cannot meet the requirements, which restricts the consumption of new energy and the overall stable operation of the power system. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for optimizing the deep peak load rate of a high-alkali coal-fired power unit, which solves the problems mentioned in the background art through the following scheme.

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

[0007] A method for optimizing the deep peak load rate of high-alkali coal-fired power units includes:

[0008] S1: Synchronously collect real-time operating data of high-alkali coal generating units connected to the power grid during deep peak shaving, as well as historically issued automatic power generation control commands, to generate the first power grid dataset;

[0009] S2: Based on the first power grid dataset, an association rule mining algorithm is used to identify the influence of high-alkali coal combustion characteristics on the dynamic response of the unit, so as to generate a second power grid dataset;

[0010] S3: Based on the second power grid dataset, construct a dynamic mapping model to describe the dynamic mapping relationship between power grid dispatching instructions and the predicted output of high-alkali coal generating units;

[0011] S4: Based on the dynamic mapping model, the automatic power generation control commands are subjected to forward rolling optimization processing to generate a first optimized command sequence for the high-alkali coal generator set;

[0012] S5: Calculate the response margin of each generator unit participating in deep peak shaving in real time, and perform dynamic load distribution among the generator units based on the first optimized instruction sequence.

[0013] S6: Obtain the actual output after executing the first optimized instruction sequence, calculate the deviation between the actual output and the predicted output of the high-alkali coal generating unit, and generate a second optimized instruction sequence to compensate for the deviation.

[0014] Preferably, in S2, the second power grid dataset is used to quantify the constraint relationship between the combustion characteristics of high-alkali coal and the dynamic response capability of the unit, and it includes at least:

[0015] A quantitative correlation function between alkali metal deposition rate and unit load change rate;

[0016] The dynamic response lag time constant matrix is ​​based on a combination of various operating conditions, wherein the dimensions of the operating condition combination include the current load level of the unit and the coal feed quality index.

[0017] The dynamic safety peak-shaving boundary curve is defined in the two-dimensional plane of load-load change rate and is jointly determined by equipment safety and environmental emission constraints.

[0018] Preferably, step S2, obtaining the second power grid dataset through an association rule mining algorithm, specifically includes:

[0019] The first power grid dataset is represented as a set of time-series variables. ,in:

[0020] Represented as The input vector at time t, its elements At least include: the rate of change of unit load setpoint Rate of change of wall temperature of heated surface ;

[0021] Represented as The target vector at time t, its elements A performance metric characterizing dynamic response capability;

[0022] For the set of time series variables Perform association rule mining to generate rules in the form of "previous items". After The strong association rule of "", where " " indicates a logical implication relationship, and must simultaneously satisfy:

[0023] Minimum support requirement: Rule "A" Support for "B" The support Defined as the union of all variables in the rule, i.e., A and B, the number of times it appears (Count(A∪B)) within the total observation period of high-alkali coal combustion, and the total data volume. The ratio, i.e. ,in This is represented as the preset minimum support threshold;

[0024] Minimum confidence requirement: Rule "A" Confidence of "B" The confidence level Defined as Support Support for the preceding term A The ratio, i.e. ,in This is represented as the preset minimum confidence threshold;

[0025] Based on the strong correlation rules, a quantitative correlation function, a dynamic response lag time constant matrix, and a dynamic safety peak-shaving boundary curve are extracted and fitted to quantify the relationship between the combustion characteristics of high-alkali coal and the dynamic response capability of the unit, so as to form the second power grid dataset.

[0026] Preferably, step S3, constructing the dynamic mapping model, specifically includes:

[0027] Construct a basic structure with the second power grid dataset as embedded parameters;

[0028] The basic structure selects a nonlinear autoregressive moving average model with external input, takes automatic power generation control commands as input, and outputs a multi-dimensional vector containing predicted power output, power output change rate, and the actual achievable steady-state power output range.

[0029] Preferably, in step S4, the automatic power generation control commands input to the dynamic mapping model are subjected to forward rolling optimization processing, which is executed cyclically:

[0030] S401: Real-time acquisition of the current total actual output and operating status of the high-alkali coal generating unit, as well as the data loaded from S2 and related to the current operating status. The corresponding second power grid dataset;

[0031] S402: Within a preset prediction time domain, a set of optimal pre-controlled total load command sequences within a control time domain is obtained by minimizing the optimization objective function, wherein the prediction time domain refers to the length of the future time period considered in the optimization calculation, and the control time domain refers to the length of the command sequence to be executed in the optimization solution;

[0032] S403: Output the instruction value corresponding to the current moment in the optimal pre-controlled total load instruction sequence as the first optimized instruction sequence;

[0033] S404: After waiting for a preset control cycle, return to S401 to perform the next round of rolling optimization with the updated state and automatic power generation control command, wherein the control cycle is the time interval between two rolling optimization calculations.

[0034] Preferably, in step S4, the objective function is solved in each control cycle, specifically expressed as:

[0035] ,

[0036] in, This is represented as the prediction time-domain step size. Represented as the index for the prediction time-domain step. This is represented as the time corresponding to the pk-th step. Represented as Pre-control total load command at any time, Represented as Automatic power generation control commands issued at all times , , These represent the weighting coefficients for instruction tracking error, instruction change rate, and constraint violation risk, respectively. , Expressed as the rate of change of the pre-controlled total load command, This is represented as the preset control cycle. Represented as a penalty function, its value varies with the sequence of pre-control instructions. The risk of triggering constraints increases. This is represented as a sequence of pre-controlled total load commands. Represented as at time The real-time operating status vector of the unit;

[0037] At the same time, the following constraints must be met:

[0038] ,

[0039] in, This represents the feasible region of instructions defined by the safety peak-shaving boundary curve. This represents the allowable range of variation rates determined by the alkali metal deposition state.

[0040] Preferably, step S5, which involves dynamic load allocation based on the first optimized instruction sequence and response margin, specifically includes:

[0041] The goal is to minimize the equivalent regulation pressure on each unit caused by the total load change, wherein the regulation pressure is positively correlated with the load change of each unit and negatively correlated with the comprehensive response margin index, thereby preferentially allocating the load change to the unit with the larger comprehensive response margin.

[0042] Preferably, step S6, generating the second optimization instruction sequence for compensating for the deviation, specifically includes:

[0043] On a second-level timescale, the actual output of the high-alkali coal generating unit is acquired in real time and compared with the predicted output output of the dynamic mapping model to calculate the real-time output deviation and the deviation change trend.

[0044] Based on the magnitude and trend of the deviation and the current operating conditions of the unit, a second optimized instruction sequence is generated, which includes millisecond-level fuel correction, second-level air supply coordination, and load rate preset correction values.

[0045] After a preset short period, verify the convergence of the deviation between the actual output and the predicted output;

[0046] If the target interval is not converged, the process continues iteratively until the deviation is eliminated.

[0047] The technical effects and advantages of this invention are as follows:

[0048] 1. This invention uses the S2 association rule algorithm to quantify the constraint relationship between the combustion characteristics of high-alkali coal and the dynamic response of the unit, which solves the core defect that the scheduling model is not optimized for the characteristics of high-alkali coal, identifies the response lag characteristics under different operating conditions, and forms a dynamic safety peak-shaving boundary under the constraints of equipment safety and environmental emission, providing a data foundation and constraints for subsequent optimization.

[0049] 2. The present invention establishes a precise dynamic relationship between power grid dispatching commands and predicted output of high-alkali coal generating units through the dynamic mapping model constructed by S3, which solves the problem that existing models cannot reflect the impact of high-alkali coal characteristics on the response, and provides an accurate prediction basis for subsequent command optimization.

[0050] 3. This invention, through the rolling optimization technology of S4, ensures tracking of power grid commands while fully considering the safety constraints and dynamic characteristics of high-alkali coal units, achieving precise matching between control commands and the actual capacity of the units, effectively solving the response deviation problem, and improving the accuracy and stability of power grid frequency tracking. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating the steps of a deep peak load rate optimization method for a high-alkali coal-fired power unit, according to an embodiment of this application. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0054] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0055] As attached Figure 1The method shown is a deep peak-shaving load rate optimization method based on high-alkali coal-fired power units. It collects operating data and historical commands from these units to explore the influence of high-alkali coal combustion characteristics on the dynamic response of the units. Dynamic load allocation is then performed based on the response margin of each unit, and compensation commands are generated according to output deviations. Specifically, the method includes the following steps:

[0056] S1: Synchronously collect real-time operating data of high-alkali coal generating units connected to the power grid during deep peak shaving, as well as historically issued automatic power generation control commands, to generate the first power grid dataset;

[0057] S2: Based on the first power grid dataset, an association rule mining algorithm is used to identify the influence of high-alkali coal combustion characteristics on the dynamic response of the unit, so as to generate a second power grid dataset;

[0058] S3: Based on the second power grid dataset, construct a dynamic mapping model to describe the dynamic mapping relationship between power grid dispatching instructions and the predicted output of high-alkali coal generating units;

[0059] S4: Based on the dynamic mapping model, the automatic power generation control commands are subjected to forward rolling optimization processing to generate a first optimized command sequence for the high-alkali coal generator set;

[0060] S5: Calculate the response margin of each generator unit participating in deep peak shaving in real time, and perform dynamic load distribution among the generator units based on the first optimized instruction sequence.

[0061] S6: Obtain the actual output after executing the first optimized instruction sequence, calculate the deviation between the actual output and the predicted output of the high-alkali coal generating unit, and generate a second optimized instruction sequence to compensate for the deviation.

[0062] Specifically, in S1, the first power grid dataset is stored in the database in the form of a structured data table with millisecond-level timestamps, which includes at least power grid-side dispatch command signals, basic unit operating parameters, and high-alkali coal characteristic-derived parameters.

[0063] In this embodiment, the grid-side dispatch command signal is mainly an automatic power generation control command sampled at the second level, acquired through a remote control device interface following standard protocols, with a sampling frequency of 2-4 seconds per point. Upon receiving the automatic power generation control command, a unique timestamp is assigned to its command data. The unit's basic operating parameters include second-level operating status data from the boiler, turbine, and generator sides, acquired through a distributed control system interface at a sampling frequency of 1 second per point. This data includes, but is not limited to, the boiler side's main steam pressure, main steam temperature, reheat steam temperature, drum water level, total coal feed, output of each coal mill, total air volume, furnace negative pressure, flue gas temperature, and wall temperature of each heating surface; the turbine side's turbine speed, regulating stage pressure, extraction steam parameters, condenser vacuum, circulating water temperature, and feedwater pump speed; and the generator side's actual unit output, reactive power, terminal voltage, and stator / rotor current. The high-alkali coal characteristic-derived parameters are obtained from additional measuring points and dedicated analyzers added to the distributed control system, with a sampling frequency synchronized with the unit's basic operating parameters.

[0064] It should be noted that the core of the unique time stamp is the master clock that supports IEEE 1588 PTPv2, which is connected to the main controller of each data acquisition server via fiber optic Ethernet to ensure that the clock deviation is less than 1 millisecond.

[0065] Furthermore, to achieve accurate monitoring of the combustion characteristics of high-alkali coal, the high-alkali coal characteristic-derived parameters included in the real-time operating data are used as differentiated measurement points, including at least: parameters reflecting the alkali metal deposition state, in this embodiment, the furnace high-temperature corrosion monitoring data obtained by corrosion probes arranged in the high-temperature area, as well as the wall temperature and its rate of change of each section of the heating surface; parameters reflecting the slagging tendency, in this embodiment, the flue gas temperature difference and working fluid temperature rise rate before and after the convective heating surface, as well as the soot blower activation frequency and effect evaluation signal; and environmentally related constraint parameters, in this embodiment, the flue gas temperature and NOx concentration at the inlet of the selective catalytic reduction denitrification system.

[0066] Furthermore, before generating the first power grid dataset, the collected raw data is preprocessed. The preprocessing includes at least: using digital filters for noise reduction, identifying and correcting outliers based on statistical rules, and aligning data with different sampling frequencies to a unified time series through interpolation.

[0067] In this embodiment, the digital filter is a first-order low-pass Butterworth filter, and the cutoff frequency is set to 0.5Hz according to the signal characteristics. The mean and standard deviation of each signal are calculated based on a moving window with a length of 10 seconds. For outliers that exceed the mean ± 3 times the standard deviation, linear interpolation is used for correction, and the time series is uniformly resampled to a time interval of 100 milliseconds through linear interpolation.

[0068] Specifically, in S2, the second power grid dataset includes at least: a quantitative correlation function between the alkali metal deposition rate and the unit load change rate; a dynamic response lag time constant matrix based on a combination of various operating conditions, wherein the dimensions of the operating condition combination include the current load level of the unit and the coal feed quality index; and a dynamic safety peak-shaving boundary curve defined in the load-load change rate two-dimensional plane and jointly determined by equipment safety and environmental emission constraints.

[0069] In one possible implementation, the relationship between the special combustion properties of high-alkali coal and its dynamic response capability is quantified from the first power grid dataset by an association rule mining algorithm, including: representing the first power grid dataset as a set of time-series variables. ,in: Represented as The input vector at time t, its elements At least include: the rate of change of unit load setpoint Rate of change of wall temperature of heated surface The alkali metal equivalent concentration index, calculated based on real-time / recent laboratory data of the coal fed into the furnace, is used as a coal quality correlation indicator. The main steam temperature rise rate after recent sootblower operation is used as an indicator of slagging and fouling status. SCR inlet smoke temperature ; Represented as The target vector at time t, its elements Performance indicators characterizing dynamic response capability should include at least: the response lag time of the unit's actual output. Steady-state deviation of load tracking For the set of time series variables Perform association rule mining to discover and generate rules in the form of "previous term". After The strong association rule of ", where , At the same time, the generated and filtered strong association rules must simultaneously meet the following requirements: minimum support requirement: rule "A Support for "B" The support Defined as the union of all variables in the rule, i.e., A and B, the number of times it appears (Count(A∪B)) within the total observation period of high-alkali coal combustion, and the total data volume. The ratio, i.e. ,in Represented as the preset minimum support threshold; minimum confidence requirement: rule "A" Confidence of "B" The confidence level Defined as Support Support for the preceding term A The ratio, i.e. ,in The minimum confidence threshold is represented by the preset threshold. Based on the strong correlation rule, a quantitative correlation function, a dynamic response lag time constant matrix, and a dynamic safety peak-shaving boundary curve are extracted and fitted to quantify the relationship between the combustion characteristics of high-alkali coal and the dynamic response capability of the unit, so as to form the second power grid dataset.

[0070] It should be noted that the fitting of the quantitative correlation function includes: screening the preceding terms. Rate of change of load setpoint of medium-sized unit , the following item Includes the rate of change of the heated surface wall temperature Or a strong association rule for the frequency of blowing dust; in the rule The median of the discrete interval and the corresponding Using the midpoint of the interval as the data point, the least squares method is used to fit the data. The linear relationship in the form is used to derive the correlation function characterizing the alkali metal deposition tendency; the construction of the dynamic response hysteresis time constant matrix includes: screening subsequent terms. Includes response lag time The rule, the previous item The index uses different combinations of medium load levels and coal quality indices obtained by discretizing coal quality test data to correspond to the rules. The values ​​are filled into a two-dimensional lookup table to form the response lag time constant matrix; the plotting of the dynamic safety peak-shaving boundary curve includes: collecting all the preceding terms Involves load instructions and subsequent items Rules are set to indicate 'excessive thermal stress' or 'excessive NOx emissions'; in the load-load change rate coordinate system, the corresponding operating points are marked as infeasible points; using the convex hull algorithm, a closed boundary line is drawn that surrounds all infeasible points, and this boundary line is the safe peak-shaving boundary curve.

[0071] In this embodiment, the association rule mining employs the Apriori algorithm, suitable for time-series data mining, which is improved by combining it with a sliding time window, including: processing the input vector... With the target vector The continuous variables in the dataset are divided into several intervals based on process knowledge and converted into discrete itemsets. A time window of length L=500 sampling points is set and slides along the time series data with a step size of S=50 points to ensure that the dynamic process can be captured. Within each time window, the Apriori algorithm is applied to find all items that meet the minimum support threshold through layer-by-layer search iteration. Frequent itemsets; based on frequent itemsets, generate all itemsets with a confidence level of not less than [a certain value]. rules .

[0072] It should be noted that the process is divided into several intervals based on technological knowledge, and this is achieved by presetting key thresholds for each variable, including: the rate of change of the unit load setpoint. It can be divided into: {'drastic change': ≥3%Pe / min, 'abrupt change': [1.5,3)%Pe / min, 'gradual change': <1.5%Pe / min}; Temperature change rate of the heated surface wall. It can be divided into: {'High-speed growth': ≥2℃ / min, 'Medium-speed growth': [0.5℃ / min, 2℃ / min), 'Low-speed growth': <0.5℃ / min}; Coal quality related indicators Based on the test data, the coal is divided into: {'high alkali coal': ≥0.6%, 'medium alkali coal': [0.3%, 0.6%), 'low alkali coal': <0.3%}. The thresholds are determined based on boiler design parameters, operating procedures and historical fault data statistics.

[0073] In this embodiment, a 660MW supercritical high-alkali coal-fired power unit was selected and compared with the traditional general method under the same AGC command sequence simulating high-permeability fluctuating conditions. The test results show that the traditional method, due to the use of general coal-fired power unit model parameters, has a significant deviation between its expected scheduling and the actual dynamic response capability of the high-alkali coal-fired power unit. Specifically, the standard deviation of the regional power grid frequency deviation is 0.08Hz, and the root mean square error between the actual output of the unit and the AGC command reaches 8.2MW. When dealing with sudden load fluctuations, the unit's response lag time is as long as 45 seconds, causing a temporary imbalance between the supply and demand of frequency regulation resources. In contrast, the method of this invention... Subsequently, by loading the dynamic response lag time constant matrix and the safety peak-shaving boundary curve generated based on historical data mining, the scheduling model parameters were optimized. Test results showed that the standard deviation of the regional power grid frequency deviation was significantly reduced to 0.05Hz, and the frequency stability was improved by about 37.5%. The root mean square error of the actual unit output tracking AGC commands was reduced to 4.5MW, and the frequency regulation accuracy was improved by about 45%. More importantly, when the load command changed abruptly, the command was prospectively smoothed according to the preset safety boundary, which shortened the actual response lag time of the unit to 28 seconds, and the lag phenomenon was improved by about 38%, effectively alleviating the short-term supply and demand imbalance caused by untimely response.

[0074] Specifically, in S3, the construction of the dynamic mapping model includes model structure definition, input and output settings, model training and validation.

[0075] In one possible implementation, constructing the dynamic mapping model includes: constructing a basic structure with the second power grid dataset as embedded parameters; the basic structure takes automatic generation control commands as input and outputs a multi-dimensional vector containing predicted output, output change rate, and actual achievable steady-state output range.

[0076] In this embodiment, the dynamic mapping model uses a nonlinear autoregressive moving average model as its basic structure. This basic structure employs the NARX structure, and its mathematical expression can be abstracted as follows:

[0077] ,

[0078] in, The inputs, represented as model inputs, mainly include automatic generation control commands input in sequence. The state inputs to the model should include at least the unit's historical output sequence and main steam pressure, among other variables. This is represented as the model output, which is the predicted value of the actual power output of the unit in the next time moment and in the future time domain. It is represented as a dynamically configured set of parameters, which is queried and configured online from the second power grid dataset generated by S2.

[0079] Furthermore, the model output It also includes the following obtained from the analysis of the predicted output sequence: the predicted delay time, which is the lag time from the point of change of the command to the start of a significant change in output; and the predicted steady-state deviation, which is the possible difference between the final steady-state value of the output predicted by the model and the target value of the command after the command step, so as to directly reflect the dynamic response capability of the high-alkali coal unit under specific operating conditions.

[0080] Furthermore, the training and verification steps for the dynamic mapping model include: using the second power grid dataset generated by S2, using the root mean square error between the model's predicted output and the actual output as the main loss function, and introducing constraint terms to make the delay trend predicted by the model consistent with the statistical lag time pattern in the second power grid dataset; adopting a segmented training strategy: training a basic NARX network using a large dataset; fine-tuning the last layer of the network using a recursive least squares method with a forgetting factor for multiple different high-alkali coal characteristic subsets; and verifying using historical deep peak-shaving data segments that were not involved in the training. In this embodiment, the qualification standard is set as follows: within the entire operating range, the RMSE of the model's predicted output is less than 1.5% of the unit's rated output, and the error of the predicted delay time does not exceed 15 seconds.

[0081] In this embodiment, unlike the general model, unified instructions, and post-event response logic architecture of the prior art, it is improved to a logic architecture of characteristic perception, customized model, and forward optimization; through S1 and S2 to quantify the dynamic characteristics of the high-alkali coal unit, the dynamic mapping model constructed by S3 serves as the digital carrier of its characteristics, generating a physically executable instruction sequence from the source, and shifting the optimization point from passive post-event adjustment to proactive pre-event planning.

[0082] Specifically, in S4, the automatic power generation control commands input to the dynamic mapping model are subjected to forward rolling optimization processing, which is executed cyclically:

[0083] S401: Real-time acquisition of the current total actual output of high-alkali coal generating units, the operating status of the units, and the second power grid dataset loaded from S2 that corresponds to the current operating status;

[0084] It should be noted that the dynamic mapping model can be used as an internal prediction model. In each control cycle, the controller calls its model to simulate and predict the unit's output response trajectory within a preset time period based on the current state, the input instruction sequence, and the high-alkali coal combustion characteristics obtained from S2. The optimization objective function is calculated and evaluated based on the output response trajectory.

[0085] S402: Within a preset prediction time domain, a set of optimal pre-controlled total load command sequences within a control time domain is obtained by minimizing the optimization objective function, wherein the prediction time domain refers to the length of the future time period considered in the optimization calculation, and the control time domain refers to the length of the command sequence to be executed in the optimization solution;

[0086] S403: Output the instruction value corresponding to the current moment in the optimal pre-controlled total load instruction sequence as the current value of the first optimized instruction sequence, and pass the optimal pre-controlled total load instruction sequence to S5; S404: After waiting for a preset control cycle, return to S401 and perform the next round of rolling optimization with the updated state and automatic generation control instruction, wherein the control cycle is the time interval between two rolling optimization calculations.

[0087] In one possible implementation, the objective function is solved in each control cycle, specifically as follows:

[0088] ,

[0089] in, This is represented as the prediction time-domain step size. Represented as the index for the prediction time-domain step. This is represented as the time corresponding to the pk-th step. Represented as Pre-control total load command at any time, Represented as Automatic power generation control commands issued at all times , , These represent the weighting coefficients for instruction tracking error, instruction change rate, and constraint violation risk, respectively. , Expressed as the rate of change of the pre-controlled total load command, This is represented as the preset control cycle. Represented as a penalty function, its value varies with the sequence of pre-control instructions. The risk of triggering constraints increases. This is represented as a sequence of pre-controlled total load commands. Represented as at time The real-time operating status vector of the unit; at the same time, the following constraints must be met:

[0090] ,

[0091] in, This represents the feasible region of instructions defined by the safety peak-shaving boundary curve. This represents the allowable range of variation rates determined by the alkali metal deposition state.

[0092] In this embodiment, the weighting coefficient , , This is a configurable positive number, the specific value of which needs to be determined on-site based on the actual dynamic characteristics of the high-alkali coal-fired power generating unit, the grid frequency regulation performance requirements, and the importance of the operational safety boundary. This will improve the accuracy of tracking power grid commands and increase... This helps to make the changes in unit output more stable, and increases... This will cause the optimization results to tend to deviate more from the safety constraint boundary.

[0093] It should be noted that, Used to minimize the tracking error between the pre-control command sequence and the original automatic generation control command; Used to minimize the rate of change of the pre-controlled instruction sequence; This is used to characterize the risk penalty of pre-control commands triggering power plant-side operational safety and environmental constraints. Its construction is based on the safety peak-shaving boundary curve obtained from S2 and the correlation function of alkali metal deposition; when the pre-control command sequence... When the predicted unit state approaches or exceeds the boundary, The value increases significantly, thus being automatically avoided in optimization; among the constraints, the feasible region of the instruction... The allowable range of change is directly defined by the safety peak-shaving boundary curve. The upper limit is determined by the quantitative correlation function and the stress limit of the unit equipment.

[0094] In this embodiment, the average tracking error of high-alkali coal generating units to minute-level frequency regulation commands under the traditional method is as high as 8%-12%, often accompanied by instantaneous frequency deviations of more than ±0.15Hz. After applying this solution, a pre-control command that matches the unit's capabilities is generated based on a precise dynamic mapping model, reducing the tracking error to within 3%, narrowing the grid frequency deviation amplitude to within ±0.05Hz, and shortening the frequency recovery time by about 40%. In areas where the penetration rate of new energy exceeds 30%, under the traditional dispatch strategy, the grid frequency qualification rate will drop below 95% in the face of minute-level drastic fluctuations in wind power / solar power. After adopting this solution, by using ultra-short-term new energy power forecast as feedforward input, high-alkali coal generating units can adjust their output in advance and smoothly, improving the forward-looking smoothing effect of fluctuations by more than 50%. Traditional commands, because they do not consider the slagging and corrosion characteristics of high-alkali coal, often cause the units to operate near dangerous conditions during deep peak shaving, forcing them to exit frequency regulation and exacerbating resource imbalances. This solution addresses this by using a safety boundary... Risk and penalties By incorporating optimization into the core, the units automatically avoid high-risk operating areas, ensuring a continuous and reliable supply of frequency regulation services. This reduces the risk of unplanned outages or depreciation of high-alkali coal units during peak shaving by 70%, significantly improving the reliability of frequency regulation resources and the overall stable operation of the power system.

[0095] Specifically, in S5, a margin-weighted optimization allocation algorithm is adopted to minimize the equivalent regulation pressure caused by the total load change to each unit. The regulation pressure is positively correlated with the load change of each unit and negatively correlated with the comprehensive response margin index, so that the load change is preferentially allocated to the unit with the larger comprehensive response margin.

[0096] In one possible implementation, the objective function of the margin-weighted optimization allocation algorithm is specifically expressed as:

[0097] ,

[0098] in, Represented as generator unit Distributed load, It is represented as its comprehensive response margin index; the comprehensive response margin index includes at least rate margin, capacity margin and health margin; the rate margin refers to the maximum safe load change rate constrained by the current high-alkali coal characteristics and equipment status; the capacity margin refers to the current adjustable load space constrained by the equipment operating boundary and environmental parameters; the health margin refers to the adjustment risk coefficient based on the predicted equipment loss rate assessment.

[0099] It should be noted that the objective function of the margin-weighted optimization allocation algorithm needs to be solved under the following constraints:

[0100] Total load balance constraints ,in The current time received from S4 The total pre-controlled load value in the first optimized instruction sequence;

[0101] Unit output upper and lower limit constraints: For each unit ,have ,in and These are the lower and upper limits of the adjustable load, determined by the unit's current capacity margin.

[0102] In this embodiment, the rate margin The dynamic mapping model established by S3 is invoked, with the unit as the basis. The current state is the initial condition. Input a positive unit test step command to simulate its output response. Take the slope of the rate of change of its output response curve at the initial moment as the base rate, and then multiply it by the rate discount factor based on the current coal alkali metal content assessment. , get ,Right now , Represented as the basic response slope; the capacity margin According to the unit Based on the current operating status, query the pre-stored safety peak-shaving boundary curve in S2 to obtain the maximum and minimum allowable loads at the current load point. The difference between these and the current load is the upward / downward load. Health margin Based on the predicted changes in turbine rotor thermal stress or boiler critical wall temperature after executing standard test commands using the dynamic mapping model, and referring to the equipment life loss curve, these values ​​are mapped to a coefficient between 0 and 1. The final comprehensive response margin index It is generated by the coupling of three elements, which can be specifically represented as: .

[0103] Specifically, in S6, generating the second optimization instruction sequence for compensating for the deviation includes: acquiring the actual output of the high-alkali coal generating unit in real time on a second-level time scale, comparing it with the predicted output output of the dynamic mapping model, and calculating the real-time output deviation and the deviation change trend; based on the magnitude and trend of the deviation and the current operating conditions of the unit, calling the preset compensation strategy library to generate a second optimization instruction sequence containing millisecond-level fuel correction, second-level air supply coordination, and preset load rate correction values; sending the second optimization instruction sequence to the unit's distributed control system for execution, and verifying the convergence of the deviation between the actual output and the predicted output after a preset short period; if it does not converge to the target interval, iterating until the deviation is eliminated.

[0104] It should be noted that the compensation strategy library is established based on fuzzy control rules. It takes the real-time output deviation and its changing trend as input, and directly maps and outputs the millisecond-level fuel correction, second-level air supply coordination, and load rate preset correction values ​​by querying a fuzzy rule table established based on historical commissioning data and operating experience of high-alkali coal units. In this embodiment, the fuzzy rules at least include: if the deviation is 'positive large' and the changing trend is 'positive fast', then the fuel correction is 'negative large'; the air supply coordination is calculated in real time based on the corrected fuel quantity using a preset air-coal ratio curve; the second optimization instruction... The generation logic of the millisecond-level fuel correction, second-level air supply coordination, and load rate preset correction value in the sequence includes: First, the millisecond-level fuel correction is generated based on the deviation calculation and directly applied to the coal feeder to achieve rapid coarse adjustment of power; Second, the second-level air supply coordination is calculated based on the expected combustion state after fuel correction to ensure the air-coal ratio balance during combustion and prevent fluctuations in environmental parameters caused by rapid fuel adjustment; Finally, based on the convergence of the current deviation, the load rate setpoint is preset and corrected, and fed back to the dynamic mapping model of S3 or the dynamic load allocation of S5 to form a closed loop across time scales.

[0105] In conclusion, 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, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep peak load rate optimization method based on a combustion high-alkali coal unit, characterized in that, include: S1: Synchronously collect real-time operating data of high-alkali coal generating units connected to the power grid during deep peak shaving, as well as historically issued automatic power generation control commands, to generate the first power grid dataset; S2: Based on the first power grid dataset, an association rule mining algorithm is used to identify the influence of high-alkali coal combustion characteristics on the dynamic response of the unit, so as to generate a second power grid dataset. The second power grid dataset is used to quantify the constraint relationship between high-alkali coal combustion characteristics and the dynamic response capability of the unit, which includes at least: a quantitative correlation function between alkali metal deposition rate and unit load change rate; a dynamic response lag time constant matrix based on multiple operating condition combinations, wherein the dimension of the operating condition combination includes the current load level of the unit and the coal feed quality index; and a dynamic safety peak-shaving boundary curve defined in the load-load change rate two-dimensional plane and jointly determined by equipment safety and environmental emission constraints. The acquisition of the second power grid dataset through the association rule mining algorithm specifically includes: representing the first grid dataset as a set of time series variables wherein: represents an input vector at the time point, the dimension m represents the number of elements of the input vector, and the elements of the input vector are At least includes: unit load set value change rate , heating surface wall temperature change rate ;​ denotes the target vector at time instant, with dimension n representing the number of elements of the target vector, whose elements is a performance index representing the dynamic response capability.​ For the set of time series variables Perform association rule mining to generate entries in the form of "previous items". After The strong association rule of ", where " " indicates a logical implication relationship, and must simultaneously satisfy: Minimum support requirement: Rule "A" Support for "B" The support Defined as the union of all variables in the rule, i.e., A and B, the number of times it appears (Count(A∪B)) within the total observation period of high-alkali coal combustion, and the total data volume. The ratio, i.e. ,in This is represented as the preset minimum support threshold; Minimum confidence requirement: Rule "A Confidence of "B" The confidence level Defined as Support Support for the preceding term A The ratio, i.e. ,in This is represented as the preset minimum confidence threshold; Based on the strong correlation rules, a quantitative correlation function, a dynamic response lag time constant matrix, and a dynamic safety peak-shaving boundary curve are extracted and fitted to quantify the relationship between the combustion characteristics of high-alkali coal and the dynamic response capability of the unit, so as to form the second power grid dataset. S3: Based on the second power grid dataset, construct a dynamic mapping model to describe the dynamic mapping relationship between power grid dispatching instructions and the predicted output of high-alkali coal generating units; S4: Based on the dynamic mapping model, the automatic power generation control commands are subjected to forward rolling optimization processing to generate a first optimized command sequence for the high-alkali coal generator set; The forward scrolling optimization process is executed cyclically: S401: Real-time acquisition of the current total actual output of the high-alkali coal-fired power generating unit, the operating state of the unit, and the corresponding second power grid data set loaded from S2 corresponding to the current operating state corresponding to the current operating state S402: Within a preset prediction time domain, a set of optimal pre-controlled total load command sequences within a control time domain is obtained by minimizing the optimization objective function, wherein the prediction time domain refers to the length of the future time period considered in the optimization calculation, and the control time domain refers to the length of the command sequence to be executed in the optimization solution; S403: Output the instruction value corresponding to the current moment in the optimal pre-controlled total load instruction sequence as the first optimized instruction sequence; S404: After waiting for a preset control cycle, return to S401 to perform the next round of rolling optimization with the updated state and automatic power generation control commands. The control cycle is the time interval between two rolling optimization calculations. In each control cycle, solve for the optimization objective function, specifically expressed as: in, This is represented as the prediction time-domain step size. Represented as the index for the prediction time-domain step. This is represented as the time corresponding to the pk-th step. Represented as Pre-control total load command at any time, Represented as Automatic power generation control commands issued at all times , , These represent the weighting coefficients for instruction tracking error, instruction change rate, and constraint violation risk, respectively. , Expressed as the rate of change of the pre-controlled total load command, This is represented as the preset control cycle. Represented as a penalty function, its value varies with the sequence of pre-control instructions. The risk of triggering constraints increases. This is represented as a sequence of pre-controlled total load commands. Represented as at time The real-time operating status vector of the unit; At the same time, the following constraints must be met: wherein, an instruction feasible region determined as a safety peak shaving boundary curve, a range of allowable change rates determined as an alkali metal deposition state; S5: Calculate the response margin of each generator unit participating in deep peak shaving in real time, and perform dynamic load distribution among the generator units based on the first optimized instruction sequence. S6: Obtain the actual output after executing the first optimized instruction sequence, calculate the deviation between the actual output and the predicted output of the high-alkali coal generating unit, and generate a second optimized instruction sequence to compensate for the deviation.

2. The method according to claim 1, wherein the method is characterized in that: S3, constructing the dynamic mapping model, specifically includes: Construct a basic structure with the second power grid dataset as embedded parameters; The basic structure selects a nonlinear autoregressive moving average model with external input, takes automatic power generation control commands as input, and outputs a multi-dimensional vector containing predicted power output, power output change rate, and the actual achievable steady-state power output range.

3. The method according to claim 1, wherein the method is characterized in that: S5, which performs dynamic load allocation based on the first optimized instruction sequence and response margin, specifically includes: The goal is to minimize the equivalent regulation pressure on each unit caused by the total load change, wherein the regulation pressure is positively correlated with the load change of each unit and negatively correlated with the comprehensive response margin index, thereby preferentially allocating the load change to the unit with the larger comprehensive response margin.

4. The method of claim 1, wherein the method is characterized by: Step S6 generates the second optimized instruction sequence for compensating for the deviation, specifically including: On a second-level timescale, the actual output of the high-alkali coal generating unit is acquired in real time and compared with the predicted output output of the dynamic mapping model to calculate the real-time output deviation and the deviation change trend. Based on the magnitude and trend of the deviation and the current operating conditions of the unit, a second optimized instruction sequence is generated, which includes millisecond-level fuel correction, second-level air supply coordination, and load rate preset correction values. After a preset short period, verify the convergence of the deviation between the actual output and the predicted output; If the target interval is not converged, the process continues iteratively until the deviation is eliminated.

Images