A method and system for evaluating the start-stop peak shaving capability of a steam turbine generator unit

Abstract

The application discloses a kind of steam turbine generator unit start-stop peak shaving capacity evaluation method and system, it is related to steam turbine generator unit performance evaluation technical field, including the following steps: based on the time series of multidimensional operating parameter obtains the fatigue damage degree and whole machine failure risk probability of steam turbine generator unit;Fatigue damage degree and whole machine failure risk probability are respectively substituted into corresponding membership function with different evaluation grades, corresponding membership degree vector is obtained;Based on two kinds of membership degree vector constructs fuzzy relation matrix, carries out fuzzy synthesis to fuzzy relation matrix and weight vector, obtains evaluation result vector;The maximum value in evaluation result vector corresponds to the evaluation grade as whole machine evaluation grade, and each component value in the evaluation result vector and the center component value of its corresponding evaluation grade are weighted and summed, to obtain whole machine evaluation score.The application combines two risk indexes of completely different nature, overcomes the one-sidedness of prior art only to evaluate fatigue damage.

Description

Technical Field

[0001] This invention relates to the field of steam turbine generator set performance evaluation technology, and in particular to a method and system for evaluating the start-stop peak-shaving capability of steam turbine generator sets. Background Technology

[0002] With the rapid increase in the penetration rate of new energy sources such as wind and solar power in the power grid, the randomness and volatility of the power system have significantly increased. In order to maintain the power balance and operational safety of the power grid, the functional positioning of traditional steam turbine generator units has been forced to shift from undertaking baseload power generation tasks to undertaking frequent and deep peak-shaving tasks. Among them, start-stop peak shaving, as a key regulation means, is being used more and more widely.

[0003] However, this frequent start-stop operation mode poses a severe off-design challenge to steam turbine generator sets, especially ultra-supercritical units originally designed for base load operation. The start-stop process subjects critical thick-walled pressure-bearing components such as the turbine high / intermediate pressure cylinder rotor, blades, boiler superheater / reheater headers, and main steam pipelines to repeated and drastic temperature and pressure gradient changes, resulting in significant thermal and mechanical stress cycles. Extensive engineering practice and research have confirmed that low-cycle fatigue (LCF) damage caused by this cyclical operating condition is the core cause of life loss, reliability degradation, and potential failure risks in such units, and its severity far exceeds that of high-temperature creep damage under traditional steady-state operation.

[0004] To quantify and manage this risk, various life assessment technologies have been developed in this field. Early solutions include the Creep-FatiguePro system, developed with support from the Electric Power Research Institute (EPRI), which enables online monitoring of the aging of critical components. This type of system collects operating parameters such as temperature and pressure from the unit, utilizes a pre-defined stress transfer function and a linear damage accumulation rule based on ASME standards, to track creep-fatigue damage in boiler and turbine components, and estimate their remaining life.

[0005] Subsequently, as demonstrated by the research of Banaszkiewicz et al., more advanced assessment methods were proposed (Advanced Lifetime Assessment of Steam Turbine Components Based on Long-Term Operating Data). The advancement of these methods lies in moving away from simplified "representative operating conditions" and instead analyzing long-term, real-world operating history data of the unit, combined with accurate thermo-mechanical coupling analysis using the Finite Element Method (FEM). This allows for more accurate calculations of the stress-strain history experienced by components during each start-stop cycle, resulting in more precise fatigue damage assessments.

[0006] Despite the significant progress made in calculating component fatigue damage using the aforementioned existing technologies, they still have the following technical limitations when addressing the current grid's systematic assessment requirements for the overall peak-shaving capacity of generating units: (1) The assessment dimension is singular and limited to physical damage mechanism: The core theory of existing advanced methods is still based on materials mechanics and fracture mechanics, focusing on the accumulation of predictable physical damage (such as fatigue and creep). Its goal is to predict the "remaining life" or "crack initiation time" of a specific component. However, as a complex system, the overall operational reliability of the unit is not only related to this, but also affected by a large number of other factors (such as abnormal status of auxiliary equipment system, control logic protection action, nonlinear coupling of operating parameters, etc.) and random sudden failures.

[0007] (2) The evaluation object is fixed and limited to key components: The existing evaluation methods are based on specific, predetermined high-stress points on key components, such as rotor blade root grooves and pipe welds. This "point-to-surface" evaluation model, although of great value in guiding the inspection and maintenance of components, cannot provide a macro-level, comprehensive evaluation of the entire unit as a system and its ability to adapt to and withstand frequent start-stop operations.

[0008] (3) The assessment results are highly specialized and lack direct decision-making guidance: The results output by existing technologies, such as "damage factor", "remaining operating hours" or "crack propagation length", are highly specialized engineering data. For decision-makers such as power plant operators and grid dispatchers, these data are difficult to directly translate into an intuitive classification of the unit's peak-shaving capacity, nor can they provide a clear and quantitative basis for horizontal comparison in optimizing the allocation of peak-shaving tasks among multiple units, thus lacking direct decision support capabilities. Summary of the Invention

[0009] In view of the deficiencies of the existing technology, the present invention provides a method and system for evaluating the start-up and peak-shaving capacity of steam turbine generator sets, which solves the existing problems.

[0010] The present invention adopts the following technical solution: In a first aspect, the present invention provides a method for evaluating the start-up and shutdown peak-shaving capacity of a steam turbine generator set, comprising the following steps: Real-time acquisition of time series of multi-dimensional operating parameters of steam turbine generator sets; A load cycle spectrum is constructed based on the time series of multidimensional operating parameters. Damage accumulation is calculated on the load cycle spectrum to obtain the fatigue damage degree of the steam turbine generator set. The time series of multidimensional operating parameters is input into a pre-trained fault prediction model to obtain the overall failure risk probability. Substituting fatigue damage degree and overall machine failure risk probability into membership functions corresponding to different evaluation levels yields the corresponding membership vectors. A fuzzy relation matrix is ​​constructed based on these two membership vectors, and the fuzzy relation matrix is ​​fuzzily synthesized with the weight vector to obtain the evaluation result vector. Specifically, mapping relationships between fatigue damage degree, overall machine failure risk probability, and membership degrees for different evaluation levels are established to obtain the membership functions. The weight vector is obtained by analyzing the importance of fatigue damage degree and overall machine failure risk probability in the evaluation. The evaluation level corresponding to the maximum value in the evaluation result vector is taken as the overall evaluation level, and the weighted sum of each component value in the evaluation result vector and its corresponding evaluation level's central score is obtained to get the overall evaluation score.

[0011] Preferably, the evaluation level includes excellent. V 1. Good V 2. Middle V 3 sums and differences V 4. Assign a central score from largest to smallest to Excellent, Good, Average, and Poor. s 1. s 2. s 3 and s 4.

[0012] Preferably, the fuzzy relation matrix is ​​as follows: ; In the formula, R It is a fuzzy relation matrix. r i This represents the membership vector of fatigue damage degree or overall machine failure risk probability to different assessment levels. m ij The membership degree of fatigue damage degree or overall machine failure risk probability to different assessment levels, where, i =1 represents the degree of fatigue damage. i=2 represents the probability of overall machine failure risk. j =1 is the optimal value. j =2 is considered good. j =3 is the middle. j =4 is the difference.

[0013] Preferably, the evaluation result vector is as follows: ; in, B = ( b 1, b 2, b 3, b 4); b j = oh 1 m 1j + oh 2 m 2j ; In the formula, B For the evaluation result vector, b j To determine the overall membership degree corresponding to the assessment level, A For the weight vector, For fuzzy operators, oh 1 represents the fatigue damage weight. oh 2 represents the probability weight of overall machine failure risk; The specific evaluation score for the entire machine is as follows: ; In the formula, "Score" represents the overall system evaluation score. s j The center score for different assessment levels.

[0014] Preferably, the step of constructing a load cycle spectrum based on the time series of multidimensional operating parameters, and performing damage accumulation calculation on the load cycle spectrum to obtain the fatigue damage degree of the steam turbine generator set specifically includes the following steps: Three-dimensional physical modeling of key components in a steam turbine generator set is performed using a finite element analysis model. The time series of multi-dimensional operating parameters are input into the key components for thermo-mechanical coupling calculation to obtain the load spectrum. The load spectrum is processed by a cycle counting method, and the load spectrum is decomposed into multiple stress cycle events. By statistically analyzing the characteristic parameters and the number of occurrences of multiple stress cyclic events, the load cyclic spectrum is obtained. Each stress cycle event in the load cycle spectrum is transformed to obtain the equivalent stress amplitude; Based on the fatigue life curves of the key component materials, determine the allowable number of cycles corresponding to each equivalent stress amplitude. The degree of fatigue damage is calculated based on the number of occurrences of each stress cycle event and the allowable number of cycles.

[0015] Preferably, the cycle counting method includes rainflow counting, which transforms each stress cycle event in the load cycle spectrum using a Goodman, Gerber, or Soderberg model, and the fatigue life curve includes an SN curve and... -N curve.

[0016] Preferably, the fault prediction model employs a deep learning model, and its pre-training includes the following steps: Historical time series of multidimensional operating parameters of steam turbine generator sets are collected, and feature vectors representing the unit status are extracted from the historical time series using a sliding time window method. Set warning time window Δ t If the unit is T If a sudden failure occurs at any time, then the time period [ T - L -Δ t , T -Δ t All feature vectors within the range are labeled as positive samples. L The window length is used; feature vectors of other normally operating data segments far from the time of the failure are labeled as negative samples to construct the training dataset. The fault prediction model is pre-trained using the training dataset.

[0017] Preferably, the step of inputting the time series of multidimensional operating parameters into a pre-trained fault prediction model to obtain the overall machine fault risk probability includes the following steps: Capture the current moment in real time using a sliding time window. t The previous length was L The time series is obtained, and the feature vector within the time window is extracted; The feature vector is input into a pre-trained fault prediction model to obtain the overall fault risk probability.

[0018] Secondly, the present invention provides a system for evaluating the start-up and peak-shaving capacity of a steam turbine generator set, comprising: The data acquisition module is used to collect the time series of multi-dimensional operating parameters of the steam turbine generator set in real time. The acquisition module is used to construct a load cycle spectrum based on the time series of multidimensional operating parameters, perform damage accumulation calculation on the load cycle spectrum, and obtain the fatigue damage degree of the steam turbine generator set; the time series of multidimensional operating parameters is input into a pre-trained fault prediction model to obtain the overall failure risk probability; The calculation module is used to substitute fatigue damage degree and overall machine failure risk probability into the membership functions corresponding to different evaluation levels to obtain the corresponding membership vectors; construct a fuzzy relation matrix based on the two membership vectors, and fuzzily synthesize the fuzzy relation matrix and the weight vector to obtain the evaluation result vector; wherein, the mapping relationship between fatigue damage degree and overall machine failure risk probability and the membership degree of different evaluation levels is established to obtain the membership functions; and the weight vector is obtained by analyzing the importance of fatigue damage degree and overall machine failure risk probability in the evaluation. The evaluation module is used to take the evaluation level corresponding to the maximum value in the evaluation result vector as the overall evaluation level, and to perform a weighted summation of each component value in the evaluation result vector and the central score of its corresponding evaluation level to obtain the overall evaluation score.

[0019] Compared with the prior art, the above-mentioned at least one technical solution adopted by the present invention can achieve the following beneficial effects: This invention first collects time series data of multidimensional operating parameters of a steam turbine generator set, and based on the time series, obtains fatigue damage degree of key components and overall failure risk probability of the generator set. This invention combines fatigue damage degree with the "overall failure risk probability" predicted by a deep learning model—two completely different risk indicators—to construct a two-dimensional evaluation system. This overcomes the limitations of existing technologies that are confined to physical damage mechanisms, achieving comprehensiveness and systematicity in the evaluation dimensions and significantly improving the accuracy of the evaluation results. Then, fatigue damage degree and overall failure risk probability are substituted into membership functions corresponding to different evaluation levels to obtain corresponding membership vectors. A fuzzy relation matrix is ​​constructed based on these two membership vectors, and the fuzzy relation matrix is ​​fuzzily synthesized with the weight vector to obtain the evaluation result vector. Through fuzzy comprehensive evaluation and hierarchical analysis, complex professional data (damage degree, failure risk probability) is transformed into intuitive levels (e.g., "excellent," "good," "medium," "poor") and quantitative scores that managers and dispatchers can directly use for decision-making. This solves the problem that existing technologies are highly specialized but lack strong decision-making guidance. The evaluation object of this invention is the generator set, rather than the remaining life of a single part. Instead, it is a system-level and comprehensive evaluation of the entire unit's ability to adapt to and withstand start-up, shutdown, and peak-shaving tasks, which is more in line with the actual needs of power plant operation and grid dispatch. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the method for evaluating the start-up and shutdown peak-shaving capacity of steam turbine generator sets based on operational reliability, as proposed in this invention.

[0022] Figure 2 This is a flowchart of the offline training process for the fault risk probability prediction model.

[0023] Figure 3 This is a hierarchical structure diagram of the comprehensive evaluation model based on AHP-FCE. Detailed Implementation

[0024] 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.

[0025] Example 1 In summary, a significant technological gap exists in this field: the lack of a systematic assessment framework that effectively quantifies and integrates specific start-up and shutdown peak-shaving processes, fatigue damage of key components, and the overall failure risk probability revealed by massive operational data. Therefore, those skilled in the art urgently need a new technical solution to overcome the limitations of traditional component-level life assessments and establish a method that can comprehensively evaluate fatigue damage and failure risk probabilities, and output intuitive, decision-making-ready assessment results of the overall turbine's peak-shaving capacity. The establishment of this method has significant technical value in guiding power generation companies to optimize their operating strategies and assisting grid dispatching agencies in making scientific decisions to ensure the safe, economical, and reliable participation of turbine generator units in peak shaving.

[0026] Based on the problems identified in the background art, this invention provides a method for evaluating the start-up and peak-shaving capacity of a steam turbine generator set, referring to... Figure 1 The evaluation method includes the following steps: S1: Processing of operating parameters and construction of load cyclic spectrum.

[0027] S11: Data Acquisition and Preprocessing: Acquire temperature, pressure, and speed parameters from the Distributed Control System (DCS) or Monitoring Information System (SIS). Acquire time-series data including at least main steam temperature, reheat steam temperature, feedwater temperature, key component metal wall temperature, main steam pressure, reheat steam pressure, power output, and turbine speed. Perform data cleaning, synchronization, and interpolation preprocessing on the acquired raw data to generate a complete and accurate multidimensional operating parameter time series, providing a reliable data foundation for subsequent stress analysis.

[0028] S12: Stress-Time History Calculation: Based on the preprocessed multidimensional operating parameter time series, a three-dimensional physical model of key components in the turbine generator set is performed using a finite element analysis (FEA) model, followed by thermo-mechanical coupling calculations. The key components are the critical wall thickness bearing components that are most susceptible to low-cycle fatigue damage during start-up and shutdown due to drastic temperature and pressure changes. The purpose of the modeling is to transform the macroscopic operating parameters of the unit into the microscopic, localized stress and strain distribution of these key components, thereby providing an accurate basis for subsequent fatigue damage calculations.

[0029] Specifically, transient thermal analysis is performed using temperature parameters as thermal boundary conditions to obtain the internal temperature field distribution of key components. T ( x , y , z , t Then, the temperature field distribution is used as a thermal load, and combined with the mechanical load derived from pressure and rotational speed parameters, structural analysis is performed to solve and output the stress time history of key components in high-stress regions. s ( t This step, known as the load spectrum, directly correlates macroscopic operating parameters with the microscopic physical stress state of the components.

[0030] S13: Load Cyclic Spectrum Construction: The load spectrum is constructed using a cyclic counting method (e.g., rainflow counting). s ( t This process decomposes the complex aperiodic stress history into a series of independent stress cycling events. For each identified stress cycling event... i Extract its characteristic parameters, including stress amplitude. s a,i With average stress s m,i Among these factors, the stress amplitude determines the magnitude of stress change in each cycle and is the main factor causing fatigue damage. The mean stress reflects the baseline stress level of the cycle and plays a corrective role in the fatigue life of the material (for example, mean stress in a tensile state accelerates fatigue). Finally, by statistically analyzing the characteristic parameters and the number of occurrences of all stress cycle events, a quantitative load cycle spectrum is formed that reflects the characteristics of the complete cyclic load within the evaluation period; its specific form is usually a two-parameter matrix.

[0031] S2: Calculation of two-dimensional operational reliability indicators.

[0032] This step is used to conduct a parallel quantitative assessment of the unit's operational reliability from two dimensions: "deterministic fatigue damage" and "failure risk probability," in order to overcome the limitation of existing technologies that only have one assessment dimension.

[0033] S21: Reliability Indicator 1: Fatigue Damage Degree D fatigue The calculation.

[0034] This index is used to quantify the deterministic fatigue damage accumulation of critical components due to cyclic loading. The calculation process for this fatigue damage accumulation method is as follows:

[0035] Mean stress correction: For each stress cycle event (including stress amplitude) in the load cycle spectrum constructed in step S1. s a,i With average stress s m,i , i For the first i (For each stress cycle event), a recognized mean stress correction theory, such as the Goodman, Gerber, or Soderberg model, is applied to transform it into an equivalent equivalent stress amplitude with zero mean stress. s ar,i .

[0036] Damage accumulation calculation: Based on the SN curve (stress-life curve) of the key component material, find the amplitude of each equivalent stress. s ar,i The corresponding number of allowed loops N f,i Subsequently, the total fatigue damage should be calculated using a linear cumulative damage rule (preferably the Palmgren-Miner rule) according to the following formula. D fatigue :

[0037] (1); in, n i For the first i The actual number of stress cycles that occurred. The results... D fatigue It is a dimensionless value with a range of [0,1]. The closer it is to 1, the more severe the cumulative physical damage.

[0038] S22: Reliability Indicator Two: Failure Risk Probability P fault The prediction.

[0039] This indicator is used to assess the risk of random and sudden system failures caused by the coupling of multiple factors, and to make up for the shortcomings of physical models.

[0040] like Figure 2 As shown, features characterizing the unit's state are extracted from the historical multidimensional operating parameter time series using a sliding time window approach. For any given moment... tThe window whose input feature vector X t This includes at least the instantaneous values ​​of the original parameters, the mean, variance, and gradient (first-order difference) of the parameters within the window. The extracted feature vectors are labeled using a historical fault database. To enable the model to predict future faults, a crucial early warning time window is introduced during the labeling process. Δt This defines the lead time for forecasting. Specifically, if the historical database records the unit's... T If a certain type of sudden failure occurs at any time, then a length of... L And its ending time t lie in[ T - L - Δt , T - Δt Feature vectors extracted from the sliding time window within the interval X t Labeled as positive samples ( y =1). This "retrospective" labeling method essentially trains the model to recognize and learn before the fault occurs. Δt The "precursor" data pattern of time; feature vectors of other normal operation data segments far removed from the time of failure. X t Then it is labeled as a negative sample. y =0). Therefore, it constitutes a large number of ( X t , y Training dataset for sample pairs D train .

[0041] The fault prediction model is trained using positive and negative samples. The fault prediction model is a deep learning model with time series modeling capabilities, such as LSTM, GRU, TCN, or Transformer. This embodiment uses LSTM as an example.

[0042] To effectively capture the time dependencies of operating parameters and the characteristics of fault precursors, a deep learning prediction model based on a Long Short-Term Memory (LSTM) network is constructed. f A typical model structure may include:

[0043] Input layer: receiving dimension is L × M The feature matrix of , where L This represents the length of the time window (number of steps). M The number of features at each time step.

[0044] LSTM layer: One or more LSTM networks are used to learn long-range dependency features from the input time series and output a hidden state sequence.

[0045] Fully connected layer: Nonlinearly integrates the features output by the LSTM layer.

[0046] Output layer: The Sigmoid activation function is used to output a scalar value between [0,1], which is the predicted failure probability.

[0047] Using the constructed training dataset D train For the model f parameters i Optimization is performed. The goal of the training process is to minimize the difference between the model's predicted output and the true label, which is calculated using the binary cross-entropy loss function. L ( i To measure:

[0048] (2); in, N The total number of training samples. y i It is the first i The true label of each sample The model is for the first i The predicted probability of each sample is calculated. A gradient descent optimization algorithm is used to iteratively update the model parameters through backpropagation. i This continues until the loss function converges or the preset number of training rounds is reached.

[0049] The offline trained model f Deployed within an online monitoring system. During the evaluation process, the system captures the current moment in real time using a sliding time window. t The length of the front is L The continuous multidimensional operating parameter sequence is obtained, and the same feature engineering is performed as described above to obtain the real-time feature vector. X t Real-time feature vectors X t Input into the model that has completed offline training. f In the middle. Because the model has been trained to identify events before a failure occurs. Δt The precursory features of time, therefore its output probability value P fault It can clearly and quantitatively characterize the unit at the current moment. t The future Δt The probability of a sudden failure occurring within a given time period. This process can be represented as:

[0050] (3); in i trained These are the final model parameters after training is complete.

[0051] S3: Comprehensive assessment and decision support based on multi-indicator fusion.

[0052] This step scientifically integrates the multidimensional and specialized reliability indicators obtained in step S2 into a single, intuitive evaluation result, addressing the problem that existing technology evaluation results are difficult to directly guide decision-making. A model combining the Analytic Hierarchy Process (AHP) and the Fuzzy Comprehensive Evaluation Method (FCE) is used for implementation, such as... Figure 3 As shown, it specifically includes:

[0053] S31: Evaluation system construction.

[0054] The definition includes "equipment fatigue damage degree" D fatigue "and overall machine failure probability" P fault "Factor set" U And a collection of comments including several levels such as "excellent", "good", "average" and "poor". V .For example:

[0055] Define factor set U : (4); It is worth noting that the factor set can include multiple indicators. For example, equipment fatigue damage can be subdivided into blade fatigue damage, boiler fatigue damage, etc. The probability of overall machine failure risk can also be subdivided similarly. For simplicity, we will use two indicators as an example here.

[0056] Definition of Comments V : (5); To facilitate the calculation of quantitative scores, a central score can be assigned to each rating level. S V : (6); S32: Weight vector determined.

[0057] A multi-criteria decision-making method (e.g., the Analytic Hierarchy Process, AHP) is employed to determine the factor set through expert knowledge or data analysis. U Weight vector of each factor A = ( oh 1, oh 2). This step is used to objectively quantify the relative importance of fatigue damage and failure probability in the overall assessment.

[0058] S33: Construction of fuzzy relationship matrix.

[0059] (a) Establishing membership function: for each factor in the factor set U k , respectively establish the mapping relationship between its value and the membership degrees of each grade in the comment set V , that is the membership function m kj ( x ). Based on a preset performance threshold, this function converts a specific indicator value into the degree to which it belongs to each grade of "excellent", "good", "medium" and "poor".

[0060] Preferably, trapezoidal or triangular membership functions are used to handle fuzzy boundaries in evaluation criteria.

[0061] For example, for indicator U 1= D fatigue , its membership function for the "excellent" ( V 1), "good" ( V 2), "medium" ( V 3), "poor" ( V 4) four grades can be set as descending half trapezoid, trapezoid, trapezoid and ascending half trapezoid functions, defined by threshold points ( c k1 , c k2 , c k3 , ...) . Taking the membership function of the "excellent" grade m 11 ( x ) as an example:

[0062] (7); wherein x is the calculated value of D fatigue , c 11 , c 12 is a preset grade division threshold.

[0063] (b) Calculating membership degree: substituting the specific values of D fatigue and P fault calculated in step S2 into a respective preset set of membership functions to obtain their membership degree vectors for the comment set V membership degree vector r 1= ( m 11 , m12 , m 13 , m 14 )and r 2 = ( m 21 , m 22 , m 23 , m 24 ).

[0064] (c) Constructing the matrix: Combine membership vectors to form a fuzzy relation matrix. R : (8); In the formula, R It is a fuzzy relation matrix. r i This represents the membership vector of fatigue damage degree or overall machine failure risk probability to different assessment levels. m ij The membership degree of fatigue damage degree or overall machine failure risk probability to different assessment levels, where, i =1 represents the degree of fatigue damage. i =2 represents the probability of overall machine failure risk. j =1 is the optimal value. j =2 is considered good. j =3 is the middle. j =4 is the difference.

[0065] S34: Output of comprehensive evaluation results.

[0066] Weight vector A With fuzzy relation matrix R Perform fuzzy synthesis operation to obtain the comprehensive evaluation result vector. B =( b 1, b 2, b 3, b 4): (9); In the formula, "This is a fuzzy operator."

[0067] The weighted average model was preferred. M (•, +), that is b j = oh 1 m 1j + oh 2 m 2j ,b j The vector represents the overall membership degree corresponding to the evaluation level. B It comprehensively reflects the degree to which the evaluated object belongs to each rating level.

[0068] Based on the comprehensive evaluation result vector B This allows for the output of both qualitative and quantitative evaluation results. The evaluation level (qualitative) is determined using the maximum membership principle, taking a vector value. B maximum value b j Corresponding rating level V j As the final evaluation grade; the comprehensive score (quantitative) is calculated using a weighted average method, defuzzifying the fuzzy evaluation results to obtain a clear score:

[0069] (10); in, s j For rating levels V j The corresponding center score. This score can be used for horizontal comparison of peak-shaving capabilities among different units. Thus, this invention completes the transformation of unit operating conditions into quantifiable, decision-support-available assessment results of start-up and shutdown peak-shaving capabilities.

[0070] Example 2 Based on the same concept, the present invention also provides a steam turbine generator set start-stop peak-shaving capability assessment system, including a data acquisition module, an acquisition module, a calculation module and an assessment module.

[0071] The acquisition module is used to collect the time series of multi-dimensional operating parameters of the steam turbine generator set in real time.

[0072] The acquisition module is used to construct a load cycle spectrum based on the time series of multidimensional operating parameters, perform damage accumulation calculation on the load cycle spectrum, and obtain the fatigue damage degree of the steam turbine generator set; the time series of multidimensional operating parameters is input into the pre-trained fault prediction model to obtain the overall failure risk probability.

[0073] The calculation module is used to substitute fatigue damage degree and overall machine failure risk probability into the membership functions corresponding to different evaluation levels to obtain the corresponding membership vectors; a fuzzy relation matrix is ​​constructed based on the two membership vectors, and the fuzzy relation matrix and the weight vector are fuzzily synthesized to obtain the evaluation result vector; wherein, the mapping relationship between fatigue damage degree and overall machine failure risk probability and the membership degree of different evaluation levels is established to obtain the membership function; by analyzing the importance of fatigue damage degree and overall machine failure risk probability in the evaluation, the weight vector is obtained.

[0074] The evaluation module is used to take the evaluation level corresponding to the maximum value in the evaluation result vector as the overall evaluation level, and to perform a weighted summation of each component value in the evaluation result vector with the central score of its corresponding evaluation level to obtain the overall evaluation score.

[0075] This invention aims to overcome the limitations of existing technologies that assess lifespan solely based on the fatigue damage mechanisms of key components, and instead provides a method for evaluating the start-stop peak-shaving capability of steam turbine generator sets based on operational reliability. This method combines load spectrum analysis reflecting the cyclic stress characteristics of the unit, component fatigue damage calculation based on fatigue cumulative damage theory, and overall unit failure risk probability prediction based on deep learning algorithms to construct a two-dimensional operational reliability assessment index system. Furthermore, a comprehensive evaluation model is used to quantitatively integrate these indicators, enabling a systematic and quantitative assessment of the comprehensive capability of steam turbine generator sets under frequent start-stop modes. The method outputs intuitive and comparable assessment results (such as assessment level or comprehensive score), providing clear and reliable decision-making basis for power generation companies to optimize their operation strategies and for grid dispatching agencies to scientifically allocate peak-shaving tasks.

[0076] (1) It achieves comprehensiveness and systematicness in the evaluation dimensions, and significantly improves the accuracy of the evaluation results.

[0077] Most existing technologies assess lifespan solely from the single dimension of fatigue damage, neglecting the randomness and sudden failure risks caused by the coupling of multiple factors. This invention, through step S2, innovatively constructs a dual-dimensional reliability index system of "deterministic fatigue damage + failure risk probability." It not only quantifies the progressive wear of the unit using fatigue cumulative damage theory but also uncovers hidden fault signs in operational data through deep learning algorithms. This method, which comprehensively considers two completely different types of risks, overcomes the one-sidedness of existing technologies, enabling the assessment results to more comprehensively and realistically reflect the overall reliability level of the steam turbine generator unit under complex peak-shaving conditions.

[0078] (2) It breaks through the limitation of using "points" to represent "areas" and realizes a macro-evaluation of the overall peak-shaving capacity of the unit.

[0079] Existing technologies often focus on high-stress points on specific components such as rotors and welds, and their assessment conclusions are difficult to represent the comprehensive performance of the entire unit as a complex system. The technical solution of this invention, in particular, constructs a probabilistic fault model through comprehensive analysis of the entire plant's DCS / SIS data, and integrates various indicators through the comprehensive evaluation model in step S3. Its evaluation object is the "generator unit" as a whole. Therefore, the output of this invention is no longer the remaining life of a single component, but a system-level, comprehensive evaluation of the entire unit's ability to adapt to and withstand start-up, shutdown, and peak-shaving tasks, which is more in line with the actual needs of power plant operation and grid dispatch.

[0080] (3) It enhances the decision-making guidance and practical value of the evaluation results, and has significant economic and social benefits.

[0081] Existing technologies output specialized data such as damage factors and remaining hours, which are difficult for operations managers and dispatchers to directly use for decision-making. This invention, through the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation model in step S3, transforms complex technical indicators into intuitive evaluation levels (such as excellent, good, average, and poor) and quantified comprehensive scores. This clear and easy-to-understand output format has strong practical value.

[0082] In terms of economic benefits: Power plant managers can allocate tasks differently to units in different health states based on the assessment results, avoiding frequent start-ups and shutdowns of units in poor condition, thereby delaying equipment aging, extending unit life, and reducing operation and maintenance costs. At the same time, it provides a scientific basis for implementing condition-based maintenance, which is expected to reduce unnecessary preventative maintenance costs and costly unplanned downtime.

[0083] In terms of social benefits: Power grid dispatching agencies can optimize dispatching more safely and economically based on the peak-shaving capacity scores of each generating unit. Prioritizing the allocation of peak-shaving tasks to units with stronger capabilities can effectively enhance the power grid's ability to cope with fluctuations in new energy sources, ensuring the overall safe, stable, and reliable operation of the power system.

[0084] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0085] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for evaluating the start-up and peak-shaving capacity of a steam turbine generator set, characterized in that, Includes the following steps: The time series of multi-dimensional operating parameters of the steam turbine generator set are collected in real time. These multi-dimensional operating parameters include temperature, pressure, and speed parameters. The load cycle spectrum is constructed based on the time series of multidimensional operating parameters, and the fatigue damage degree of the steam turbine generator set is obtained by performing damage accumulation calculation on the load cycle spectrum. The time series of multidimensional operating parameters are input into a pre-trained fault prediction model to obtain the overall failure risk probability. Substituting fatigue damage degree and overall machine failure risk probability into membership functions corresponding to different evaluation levels yields the corresponding membership vectors. A fuzzy relation matrix is ​​constructed based on these two membership vectors, and the fuzzy relation matrix is ​​fuzzily synthesized with the weight vector to obtain the evaluation result vector. Specifically, mapping relationships between fatigue damage degree, overall machine failure risk probability, and membership degrees for different evaluation levels are established to obtain the membership functions. The weight vector is obtained by analyzing the importance of fatigue damage degree and overall machine failure risk probability in the evaluation. The evaluation level corresponding to the maximum value in the evaluation result vector is taken as the overall machine evaluation level, and the weighted sum of each component value in the evaluation result vector and its corresponding evaluation level center score is obtained to get the overall machine evaluation score. The fault prediction model employs a deep learning model, and its pre-training includes the following steps: Historical time series of multidimensional operating parameters of steam turbine generator sets are collected, and feature vectors representing the unit status are extracted from the historical time series using a sliding time window method. Set warning time window Δ t If the unit is T If a sudden failure occurs at any time, then the time period [ T - L -Δ t , T -Δ t All feature vectors within the range are labeled as positive samples. L The window length is used; feature vectors of other normally operating data segments far from the time of the failure are labeled as negative samples to construct the training dataset. The fault prediction model is pre-trained using the training dataset.

2. The method for evaluating the start-up and peak-shaving capacity of a steam turbine generator set as described in claim 1, characterized in that, The evaluation level includes Excellent. V 1. Good V 2. Middle V 3 sums and differences V 4. Assign a central score from largest to smallest to Excellent, Good, Average, and Poor. s 1. s 2. s 3 and s 4.

3. The method for evaluating the start-up and peak-shaving capacity of a steam turbine generator set as described in claim 2, characterized in that, The fuzzy relation matrix is ​​shown below: ； In the formula, R It is a fuzzy relation matrix. r i This represents the membership vector of fatigue damage degree or overall machine failure risk probability to different assessment levels. μ ij The membership degree of fatigue damage degree or overall machine failure risk probability to different assessment levels, where, i =1 represents the degree of fatigue damage. i =2 represents the probability of overall machine failure risk. j =1 is considered optimal. j =2 is considered good. j =3 is the middle. j =4 is the difference.

4. The method for evaluating the start-up and peak-shaving capacity of a steam turbine generator set as described in claim 3, characterized in that, The evaluation result vector is shown below: ； in, B = ( b 1, b 2, b 3, b 4)； b j = ω 1 μ 1j + ω 2 μ 2j ； In the formula, B For the evaluation result vector, b j To determine the overall membership degree corresponding to the assessment level, A For the weight vector, For fuzzy operators, ω 1 represents the fatigue damage weight. ω 2 represents the probability weight of overall machine failure risk; The specific evaluation score for the entire machine is as follows: ； In the formula, "Score" represents the overall system evaluation score. s j The center score for different assessment levels.

5. The method for evaluating the start-up and peak-shaving capacity of a steam turbine generator set as described in claim 1, characterized in that, The process of constructing a load cycle spectrum based on the time series of multidimensional operating parameters, and performing damage accumulation calculation on the load cycle spectrum to obtain the fatigue damage degree of the steam turbine generator set, specifically includes the following steps: Three-dimensional physical modeling of key components in a steam turbine generator set is performed using a finite element analysis model. The time series of multi-dimensional operating parameters are input into the key components for thermo-mechanical coupling calculation to obtain the load spectrum. The load spectrum is processed by a cycle counting method, and the load spectrum is decomposed into multiple stress cycle events. By statistically analyzing the characteristic parameters and the number of occurrences of multiple stress cyclic events, the load cyclic spectrum is obtained. Each stress cycle event in the load cycle spectrum is transformed to obtain the equivalent stress amplitude; Based on the fatigue life curves of the key component materials, determine the allowable number of cycles corresponding to each equivalent stress amplitude. The degree of fatigue damage is calculated based on the number of occurrences of each stress cycle event and the allowable number of cycles.

6. The method for evaluating the start-up and peak-shaving capacity of a steam turbine generator set as described in claim 5, characterized in that, The cycle counting method includes rainflow counting, which transforms each stress cycle event in the load cycle spectrum using Goodman, Gerber, or Soderberg models. The fatigue life curves include SN curves and... -N curve.

7. The method for evaluating the start-up and peak-shaving capacity of a steam turbine generator set as described in claim 1, characterized in that, The step of inputting the time series of multidimensional operating parameters into a pre-trained fault prediction model to obtain the overall machine fault risk probability includes the following steps: Capture the current moment in real time using a sliding time window. t The previous length was L The time series is obtained, and the feature vector within the time window is extracted; The feature vector is input into a pre-trained fault prediction model to obtain the overall fault risk probability.

8. An evaluation system based on the method for evaluating the start-stop peak-shaving capacity of a steam turbine generator set as described in claim 1, characterized in that, include: The data acquisition module is used to collect the time series of multi-dimensional operating parameters of the steam turbine generator set in real time. The acquisition module is used to construct a load cycle spectrum based on the time series of multi-dimensional operating parameters, perform damage accumulation calculation on the load cycle spectrum, and obtain the fatigue damage degree of the steam turbine generator set. The time series of multidimensional operating parameters are input into a pre-trained fault prediction model to obtain the overall failure risk probability. The calculation module is used to substitute fatigue damage degree and overall machine failure risk probability into the membership functions corresponding to different evaluation levels to obtain the corresponding membership vectors; construct a fuzzy relation matrix based on the two membership vectors, and fuzzily synthesize the fuzzy relation matrix and the weight vector to obtain the evaluation result vector; wherein, the mapping relationship between fatigue damage degree and overall machine failure risk probability and the membership degree of different evaluation levels is established to obtain the membership functions; and the weight vector is obtained by analyzing the importance of fatigue damage degree and overall machine failure risk probability in the evaluation. The evaluation module is used to take the evaluation level corresponding to the maximum value in the evaluation result vector as the overall evaluation level, and to perform a weighted summation of each component value in the evaluation result vector and the central score of its corresponding evaluation level to obtain the overall evaluation score.

Images