Transformer-based gas turbine secondary air system performance prediction method

Abstract

The invention relates to the technical field of gas turbine performance prediction, and discloses a gas turbine secondary air system performance prediction method based on transformer, and the method comprises the steps: obtaining historical operation data of a healthy gas turbine secondary air system under a steady-state working condition, carrying out the preprocessing, extracting key feature parameters, and carrying out the time serialization, thereby obtaining a key feature sequence; constructing and training a first Transform model by utilizing the key feature sequence, and predicting turbine outer cylinder cavity pressure and compressor exhaust pressure by combining real-time operation data; a residual time sequence is constructed by comparing a measured value with a predicted value, and a future residual is predicted by using a second Transform model, so that the performance degradation degree of the system is quantitatively evaluated, and a health evaluation report is generated; the method can improve the performance prediction accuracy and maintenance efficiency of the gas turbine secondary air system, ensures stable operation of the system, reduces the maintenance cost, and improves the energy utilization efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of gas turbine performance prediction, and more specifically, to a method for predicting the performance of the secondary air system of a gas turbine based on a transformer. Background Art

[0002] As a key device in modern industry, the efficient operation of a gas turbine depends on the stable support of the secondary air system (SAS). This system provides guarantees for turbine cooling, bearing sealing, etc. of the gas turbine by reasonably distributing compressed air, ensuring the long-term stable operation of the device in high-temperature and high-pressure environments. However, with the increase in operating time, the decline in the performance of the secondary air system will directly affect the overall efficiency and component life of the gas turbine. Therefore, real-time monitoring and accurate prediction of the performance of the secondary air system are particularly important.

[0003] In the prior art, the Chinese patent application with the publication number CN113187592A discloses a monitoring method, device, equipment and storage medium for a secondary air system, which determines whether there is a fault in the system by obtaining the flow difference reference value and the auxiliary calculation value of the secondary air pump in different states. However, this method can only determine whether there is a fault in the system and cannot give an early warning of the gradual degradation of the system performance. Specifically, this method determines whether there is a fault in the system by setting preset stability conditions and effectiveness conditions, but this method lacks the ability to predict the decline trend of the system performance and cannot provide early maintenance suggestions for the operation and maintenance personnel.

[0004] The Chinese patent application with the publication number CN117328978A provides a flow estimation method for a secondary air system, which mainly analyzes the instantaneous flow or characteristics of the system by iteratively calculating the mass flow. However, this method is mainly used for instantaneous flow or system characteristic analysis and cannot effectively support the performance evaluation on a long time scale. Specifically, this method analyzes the system characteristics by estimating the mass flow of the secondary air injection valve, but this method lacks the ability to predict the long-term change trend of the system performance and cannot provide comprehensive decision-making support for the long-term operation and maintenance of the gas turbine.

[0005] The existing technical solutions mainly focus on the level of real-time monitoring and fault diagnosis, lacking intelligent prediction of the decline trend of the performance of the secondary air system. Most of these technical solutions rely on static analysis and short-term fault judgment, and do not make full use of the large amount of historical data accumulated during the operation of the gas turbine. In the application of data-driven methods, the advantages of deep learning models are not fully utilized to capture the dynamic interaction relationships between multi-dimensional features. Summary of the Invention

[0006] To overcome the above-mentioned defects of the prior art, the present invention provides a performance prediction method for the secondary air system of a gas turbine based on a transformer. By deeply analyzing the historical and real-time operation data of the secondary air system of the gas turbine, the system performance can be accurately predicted. This method can effectively improve the accuracy of prediction, timely detect signs of system performance decline, provide a comprehensive health assessment report for operation and maintenance personnel, help reasonably arrange maintenance plans, reduce maintenance costs, and ensure the safe and stable operation of the gas turbine.

[0007] The present invention is applicable to various scenarios using gas turbines, such as power plants, power supply systems in industrial production, etc. In these scenarios, the stable operation of the gas turbine is crucial. Through this method, the performance of the secondary air system can be monitored in real time, potential problems can be detected in advance, the efficient and stable operation of the gas turbine can be ensured, and production interruptions and economic losses caused by system failures can be avoided.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A performance prediction method for the secondary air system of a gas turbine based on a transformer, comprising:

[0010] Obtain historical operation data including timestamps under the steady-state condition of a healthy gas turbine secondary air system, and preprocess the historical operation data; extract key characteristic parameters from the preprocessed historical operation data, and serialize the key characteristic parameters according to the timestamps to obtain a key characteristic sequence;

[0011] According to the key characteristic sequence, construct and train a first Transformer model; obtain a real-time operation data sequence including timestamps of the current gas turbine secondary air system, and according to the real-time operation data sequence and the trained first Transformer model, obtain a time series of predicted values of the turbine outer cylinder cavity pressure and a time series of predicted values of the compressor discharge pressure

[0012] Obtain a time series of measured values of the turbine outer cylinder cavity pressure P of the current gas turbine cyl and a time series of measured values of the compressor discharge pressure P comp , according to P cyl , P comp , and obtain the current residual time series of the performance ratio; based on the current residual time series and a pre-constructed second Transformer model, predict the residual time series at the next Y time points According to quantitatively evaluate the degree of performance decline of the current gas turbine secondary air system, and generate a health assessment report.

[0013] Furthermore, the key characteristic parameters include the total inlet temperature T of the compressor in , the static pressure P of the inlet wall of the compressor in , the outlet pressure P of the compressor out , the outlet temperature T of the compressor out and the opening degree V of the turbine cooling valve cool ;

[0014] The obtained key characteristic sequence includes:

[0015] Perform time serialization processing on the total inlet temperature T of the compressor according to the timestamp to obtain the total inlet temperature time series T in ; Perform time serialization processing on the static pressure P of the inlet wall of the compressor according to the timestamp to obtain the static pressure time series P of the inlet wall of the compressor in-seq ; Perform time serialization processing on the outlet pressure P of the compressor according to the timestamp to obtain the outlet pressure time series P of the compressor in ; Perform time serialization processing on the outlet temperature T of the compressor according to the timestamp to obtain the outlet temperature time series T of the compressor in-seq ; Perform time serialization processing on the opening degree V of the turbine cooling valve according to the timestamp to obtain the opening degree time series V of the turbine cooling valve out ; Perform time serialization processing on the opening degree V of the turbine cooling valve according to the timestamp to obtain the opening degree time series V of the turbine cooling valve out-seq ; out ; Perform time serialization processing on the outlet temperature T of the compressor according to the timestamp to obtain the outlet temperature time series T of the compressor out-seq ; cool ; Perform time serialization processing on the opening degree V of the turbine cooling valve according to the timestamp to obtain the opening degree time series V of the turbine cooling valve cool-seq ;

[0016] Align T in-seq , P in-seq , P out-seq , T out-seq and V cool-seq according to the timestamp, and construct the key characteristic sequence TS seq =(TS seq,1 , TS seq,2 ,…, TS seq,n ), where TS seq,k is the k-th element in TS seq , n is the total number of elements in TS seq , and 1 ≤ k ≤ n.

[0017] Furthermore, each element in the TS seq is a six-tuple:

[0018] TS seq,k =(t k , T in,k , P in,k , P out,k , T out,k , V cool,k );

[0019] Among them, t k is the timestamp of the k-th element in TS seq , T in,k is the total inlet temperature of the compressor at the timestamp t k , P in,k is the static pressure of the compressor inlet wall at the timestamp t k , P out,k is the outlet pressure of the compressor at the timestamp t k , T out,k is the outlet temperature of the compressor at the timestamp t k , V cool,k is the opening degree of the turbine cooling valve at the timestamp t k .

[0020] Furthermore, the first Transformer model includes an input embedding layer, a positional encoding layer, a self-attention mechanism, a multi-head attention mechanism, a feed-forward neural network, and an output layer;

[0021] The constructing and training of the first Transformer model includes:

[0022] Extracting the model input feature vector TS seq from each element TS seq,k of the key feature sequence TS in,k =(T in,k , P in,k , P out,k , T out,k , V cool,k );

[0023] Inputting the model input feature vector TS in,k into the input embedding layer for encoding to obtain the embedded feature representation E k at the timestamp t k ;

[0024] Obtaining the positional encoding vector P k at the timestamp t k through the positional encoding layer;

[0025] Adding the positional encoding vector P k to the embedded feature representation E k to obtain the model input feature representation Z k at the timestamp t k ;

[0026] Obtaining the input sequence Z = {Z1, Z2,..., Z k} according to the model input feature representation Z n .

[0027] Furthermore, the building and training of the first Transformer model further includes:

[0028] Through the self-attention mechanism in the first Transformer model, calculate the correlation between the model input feature representations corresponding to different timestamps in the input sequence Z to obtain a one-dimensional feature correlation matrix;

[0029] The multi-head attention mechanism in the first Transformer model calculates the relationships between the various input sub-features in the model input feature representation Z k through n2 attention heads in parallel to obtain a multi-head attention weight matrix;

[0030] Perform a non-linear transformation on the one-dimensional feature correlation matrix and the multi-head attention weight matrix through a feed-forward neural network, and the output layer maps the output of the feed-forward neural network to the turbine outer cylinder cavity pressure and the compressor discharge pressure.

[0031] Furthermore, the position encoding vector P k = [P k,0 , P k,1 , P k,2 , P k,3 , …, P k,2d-1 , where d represents the dimension of the position encoding vector, represents the value of the position encoding vector in the j * -th dimension, and j * is the dimension index in the position encoding vector, 0 ≤ j * ≤ 2d - 1; the values of the even dimensions are calculated by the sine function, and the values of the odd dimensions are calculated by the cosine function.

[0032] Furthermore, the model input feature representation Z k contains five input sub-features, namely the total inlet temperature feature T' k at the timestamp t in,k , the inlet wall static pressure feature P' k at the timestamp t in,k , the outlet pressure feature P' k at the timestamp t out,k , the outlet temperature feature T' k at the timestamp t out,k , and the turbine cooling valve opening feature V' k at the timestamp t cool,k ;

[0033] The n2 attention heads at least include head1 = (T' in,k , P' out,k ), head2 = (V' cool,k , P' in,k) and the head 3 head3 = (V' cool,k , T' out,k ).

[0034] Furthermore, based on the real-time operation data sequence and the trained first Transformer model, obtain the time series of the predicted values of the turbine outer cylinder cavity pressure and the time series of the predicted values of the compressor discharge pressure including:

[0035] Extract the real-time total inlet temperature of the compressor the real-time static pressure of the compressor inlet wall the real-time outlet pressure of the compressor the real-time outlet temperature of the compressor and the real-time opening degree of the turbine cooling valve

[0036] Respectively perform time serialization processing on and to obtain the real-time total inlet temperature series of the compressor the real-time static pressure series of the compressor inlet wall the real-time outlet pressure series of the compressor the real-time outlet temperature series of the compressor and the real-time opening degree series of the turbine cooling valve

[0037] According to and the first Transformer model, obtain the time series of the predicted values of the turbine outer cylinder cavity pressure and the time series of the predicted values of the compressor discharge pressure

[0038] Furthermore, the said according to and the first Transformer model, obtain the time series of the predicted values of the turbine outer cylinder cavity pressure and the time series of the predicted values of the compressor discharge pressure including:

[0039] Align and according to the time stamp to construct the real-time key feature vector set TN seq =(TN seq,1 , TN seq,2 ,…, TN seq,m ), where TN seq,j is the real-time key feature vector at the j-th time point, m is the total number of time points, 1≤j≤m; Among them, is the real-time total inlet temperature of the compressor at the j-th time point, is the static pressure of the compressor inlet wall at the j-th time point, is the real-time compressor outlet pressure at the j-th time point, is the real-time compressor outlet temperature at the j-th time point, is the opening of the turbine cooling valve at the j-th time point;

[0040] Input TN seq,j into the trained first Transformer model to obtain the predicted value of the turbine outer cylinder cavity pressure and the predicted value of the compressor exhaust pressure at the j-th time point at the j-th time point

[0041] According to obtain the time series of the predicted values of the turbine outer cylinder cavity pressure According to obtain the time series of the predicted values of the compressor exhaust pressure

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] By introducing the Transformer deep learning model, the present invention replaces the traditional mechanism modeling in a data-driven manner, makes full use of the operating data of the gas turbine, and overcomes the limitations of the traditional mechanism modeling in dealing with complex systems. With its self-attention mechanism, the Transformer algorithm can effectively capture the dynamic interaction relationships between multi-dimensional features, not only improving the prediction accuracy but also significantly enhancing the model's adaptability to complex dynamic characteristics, enabling the model to better handle the non-linearity and dynamic changes in the gas turbine secondary air system. Secondly, this method quantifies the performance degradation degree of the secondary air system by calculating the residual between the model predicted value and the measured value, and conducts dynamic evaluation in combination with historical trends, realizing the intelligent degradation evaluation of the system performance. The size of the residual directly reflects the obvious degree of performance degradation, providing an intuitive decision-making basis for system maintenance and optimization, achieving intelligent and refined performance management, helping to timely discover potential system problems, reasonably arrange maintenance plans, reduce maintenance costs, and improve the reliability and operating efficiency of the equipment. In addition, this method makes full use of the steady-state data in the historical operating data of the gas turbine, extracts key characteristic parameters as model inputs, and constructs a performance prediction framework based on Transformer. Through deep learning technology, it can extract the implicit long-term trends and interaction characteristics from historical data, realize accurate modeling and prediction of system performance, provide strong technical support for the stable operation and performance optimization of the gas turbine, help to improve energy utilization efficiency, and ensure the safe, stable and efficient operation of the gas turbine. Description of the Drawings

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0045] Figure 1 This is the principle flowchart of the performance prediction method for the secondary air system of a gas turbine based on transformer in the present invention;

[0046] Figure 2 This is the method flowchart for obtaining the key feature sequence in the performance prediction method for the secondary air system of a gas turbine based on transformer in the present invention;

[0047] Figure 3 This is the method flowchart for quantitatively evaluating the performance degradation degree of the current secondary air system of a gas turbine and generating a health assessment report in the performance prediction method for the secondary air system of a gas turbine based on transformer in the present invention. Specific embodiments

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0049] Embodiment 1

[0050] Please refer to Figure 1 As shown, this embodiment provides a performance prediction method for the secondary air system of a gas turbine based on transformer, including:

[0051] Step S1000, obtain the historical operation data including timestamps under the steady-state condition of a healthy gas turbine secondary air system, and preprocess the historical operation data; extract key feature parameters from the preprocessed historical operation data, and serialize the key feature parameters according to the timestamps to obtain a key feature sequence;

[0052] Further, step S1000 includes:

[0053] Step S1100, obtain the historical operation data including timestamps under the steady-state condition of a healthy gas turbine secondary air system, and preprocess the historical operation data;

[0054] Specifically, first, historical operation data under steady-state conditions (i.e., when the operation parameters of the gas turbine are stable and there are no abnormal fluctuations) need to be extracted from the operation records of the gas turbine. These data usually include timestamps, operation parameters (such as temperature, pressure, flow rate, etc.), and equipment status information. The purpose of obtaining these data is to provide basic data support for subsequent feature extraction and model training.

[0055] Preprocessing is an important step in data processing, and its purpose is to convert the original data into a format suitable for model training. The specific operations of preprocessing include:

[0056] Data cleaning: Remove outliers and noise from the data. Outliers may be caused by sensor failures or data transmission errors. If these data are not processed, they may affect the training effect of the model. For example, if the temperature data at a certain time point significantly deviates from the normal range, it can be considered an outlier and needs to be corrected or deleted.

[0057] Data alignment: Ensure that all data points have the same timestamp. Since the operation data of the gas turbine are usually collected by multiple sensors simultaneously, there may be inconsistent timestamps. Through the alignment operation, it can be ensured that the timestamps of each data point are consistent, facilitating subsequent time series analysis.

[0058] Data normalization: Convert the data to the same scale range. Since different operation parameters may have different dimensions and value ranges, directly using the original data for model training may cause the model to be overly sensitive to certain parameters. Through the normalization operation, all parameters can be converted to the same scale range, such as [0, 1] or [-1, 1], thereby improving the training effect of the model.

[0059] Data completion: For missing data points, interpolation or other methods can be used for completion. There may be partial missing data in the operation data of the gas turbine. If these missing data points are not processed, they may affect the training effect of the model. Through the interpolation operation, the missing data points can be completed according to the values of adjacent data points to ensure the integrity of the data.

[0060] Through the above preprocessing operations, the original historical operation data can be converted into a format suitable for model training, providing high-quality data support for subsequent feature extraction and model training.

[0061] Step S1200, extract key feature parameters from the preprocessed historical operation data. The key feature parameters include the total inlet temperature T of the compressor in , the static pressure P on the inlet wall of the compressor in , the outlet pressure P of the compressor out , the outlet temperature T of the compressor outand the opening degree V of the turbine cooling valve cool ;

[0062] Specifically, in step S1200, key characteristic parameters are extracted from the preprocessed historical data. This process is based on the physical mechanism of the gas turbine secondary air system, screening the five parameters that have the most significant impact on the system performance to form the core variable set of the model input.

[0063] The total temperature T at the compressor inlet in Refers to the total temperature of the air at the compressor inlet section, including the static temperature and the dynamic temperature, which is a key parameter for calculating the air density and the compression work. High temperature will reduce the air density, resulting in a decrease in the compressor mass flow rate, and further affecting the cooling capacity of the secondary air system. The static pressure P on the wall of the compressor inlet in The static pressure measured on the wall of the compressor inlet pipeline, reflecting the resistance of the intake system (such as the degree of filter blockage) and the ambient air pressure. A decrease in the static pressure may indicate an increase in the intake resistance, resulting in a decrease in the compressor efficiency. The pressure P at the compressor outlet out Is the pressure value of the air after being compressed by the compressor, directly determining the flow driving force of the cooling air to the outer cylinder cavity of the turbine. The fluctuation of the outlet pressure will affect the flow control accuracy of the cooling valve and is the core index for evaluating the pressure balance of the secondary air system. The temperature T at the compressor outlet out : The temperature rise generated due to work consumption during the compression process, affecting the thermodynamic state of the cooling air. An increase in the outlet temperature will reduce the cooling efficiency and increase the thermal load of the turbine components. The opening degree V of the turbine cooling valve cool Refers to the real-time adjustment state (0%-100%) of the turbine cooling valve, which is the execution parameter for the system to actively control the cooling air flow rate. The change in the opening degree directly affects the pressure in the outer cylinder cavity of the turbine and is the key adjustment variable connecting the compressor output and the turbine cooling demand.

[0064] Through correlation analysis (such as Pearson correlation coefficient > 0.5) and mechanism verification, the strong correlation between the above parameters and the pressure in the outer cylinder cavity of the turbine and the compressor exhaust pressure is determined. For example, the correlation coefficient between the opening degree of the turbine cooling valve and the cavity pressure reaches 0.85, indicating a direct causal relationship between the two. The extracted parameters are all continuous time series data, strictly aligned with the time stamps preprocessed in step S1100, forming the structured data corresponding to "parameters - time", providing a standardized input for the time serialization process in step S1300.

[0065] Exemplarily, Table 1 shows the Pearson correlation coefficients of a certain type of gas turbine. A positive sign indicates that the variables change in the same direction (such as when P out rises, the pressure in the outer cylinder cavity of the turbine rises), and a negative sign indicates that the variables change in the opposite direction (such as when the opening degree V cool increases, the pressure in the outer cylinder cavity of the turbine decreases).

[0066] Table 1: Table of Pearson correlation coefficient results

[0067]

[0068] By performing correlation analysis and mechanism verification, the interference of secondary parameters (such as bearing temperature and lubricating oil pressure) is excluded, and the input dimension of the model is controlled within a reasonable range (5 parameters) to avoid the decline in model training efficiency caused by the "curse of dimensionality". For example, if 20 redundant parameters are included, the computational complexity of the self-attention mechanism will increase by 400%, while the selected key feature set reduces the computational cost while retaining 90% of the effective information. The parameter selection combines thermodynamic principles (such as the energy conversion of the inlet and outlet parameters of the compressor) and control logic (such as the regulating effect of the cooling valve opening), enabling the model to learn feature interactions that conform to the actual operating laws. For example, the negative correlation between the total inlet temperature and the outlet pressure of the compressor (high temperature → low density → decrease in outlet pressure) is determined by the aerodynamic formula, and the model improves the prediction accuracy by learning such mechanism correlations. The extracted parameters are the core elements of the key feature sequence in step S1300, and their quality directly affects the training effect of the first Transformer model. For example, accurate inlet wall static pressure data enables the model to correctly identify the impact of changes in the intake system resistance on the compressor performance, thereby improving the accuracy of exhaust pressure prediction.

[0069] Step S1300: Serialize the key feature parameters according to the time stamp to obtain a key feature sequence;

[0070] Furthermore, as Figure 2 shown, step S1300 includes:

[0071] Step S1310: Perform time serialization processing on the total inlet temperature T in of the compressor according to the time stamp to obtain the total inlet temperature time series T in-seq ;

[0072] Step S1320: Perform time serialization processing on the static pressure P in of the inlet wall of the compressor according to the time stamp to obtain the static pressure time series P in-seq of the inlet wall of the compressor;

[0073] Step S1330: Perform time serialization processing on the outlet pressure P out of the compressor according to the time stamp to obtain the outlet pressure time series P out-seq of the compressor;

[0074] Step S1340: Perform time serialization processing on the outlet temperature T out of the compressor according to the time stamp to obtain the outlet temperature time series T out-seq of the compressor;

[0075] Step S1350, perform time serialization processing on the opening degree V of the turbine cooling valve according to the time stamp, and obtain the time series V of the opening degree of the turbine cooling valve cool ; cool-seq ;

[0076] Step S1360, align T in-seq , P in-seq , P out-seq , T out-seq and V cool-seq according to the time stamp, and construct the key feature sequence TS seq =(TS seq,1 , TS seq,2 , …, TS seq,n ), where TS seq,k is the k-th element in TS seq , n is the total number of elements in TS seq , and 1 ≤ k ≤ n.

[0077] Each element in TS seq is a six-tuple:

[0078] TS seq,k =(t k , T in,k , P in,k , P out,k , T out,k , V cool,k );

[0079] Among them, t k is the time stamp of the k-th element TS seq in TS seq,k , T in,k is the total inlet temperature of the compressor at the time stamp t k , P in,k is the static pressure of the inlet wall of the compressor at the time stamp t k , P out,k is the outlet pressure of the compressor at the time stamp t k , T out,k is the outlet temperature of the compressor at the time stamp t k , and V cool,k is the opening degree of the turbine cooling valve at the time stamp t k .

[0080] Specifically, step S1300 aims to convert the preprocessed historical operation data into a key feature sequence suitable for processing by the Transformer model. By performing time serialization and time stamp alignment, a multi-variable time series data structure containing time dimension information is constructed to solve the problem of structured expression of multi-parameter time series data and provide a standardized input for subsequent model training. In steps S1310 to S1350, for the total inlet temperature T of the compressorin , the static pressure P of the compressor inlet wall in , the compressor outlet pressure P out , the compressor outlet temperature T out and the opening degree V of the turbine cooling valve cool These five key characteristic parameters are respectively processed by time series. The specific operation is to arrange the measured values of the same parameter at different times in a continuous time series according to the time stamp of data acquisition. This process solves the discretization problem of the association between parameter values and time in the original data, enabling the historical change trend of each parameter to be captured by the model. Taking the total temperature T in of the compressor inlet as an example, its time series T in-seq reflects the long-term influence of the ambient temperature on the compressor efficiency. During the operation of the gas turbine, an increase in the total inlet temperature will cause a decrease in the air density, thereby affecting the compression efficiency of the compressor and the cooling effect of the secondary air system. Through time series, the model can learn the lagging effect of the change of T in under different seasons or diurnal temperature differences on the system performance. Similarly, the time series P in of the static pressure P of the compressor inlet wall is used to characterize the change of the intake system resistance, and the time series P in-seq of the outlet pressure P is directly related to the driving force of the cooling air, and the time series T out of the outlet temperature T out-seq reflects the thermal effect of the compression process, and the time series V out of the opening degree V of the turbine cooling valve out-seq reflects the adjustment strategy of the system for the cooling demand. cool of the opening degree V of the turbine cooling valve cool-seq reflects the adjustment strategy of the system for the cooling demand.

[0081] In step S1360, align T in-seq , P in-seq , P out-seq , T out-seq and V cool-seq according to the time stamp to construct a key characteristic sequence TS seq in the form of a six-tuple. Each six-tuple TS seq,k = (t k , T in,k , P in,k , P out,k , T out,k , V cool,k ) corresponds to the multi-parameter state at the same moment t k , where t k is the time stamp, T in,k is the total temperature of the compressor inlet at moment t k , P in,k is the static pressure of the compressor inlet wall at moment t kThe static pressure of the inlet wall at a certain moment, and so on. Timestamp alignment solves the possible clock deviation problem in multi-sensor data acquisition, ensuring strict correspondence of different parameters in the time dimension. For example, if the sudden drop in the compressor outlet pressure P out at time t k is associated with the increase in the total inlet temperature T in at the same moment, the aligned sequence enables the model to capture the immediate coupling relationship between the two. By integrating multi-parameters into a unified time series structure, TS seq provides an input format containing spatio-temporal correlation information for the Transformer model, meeting the model's processing requirements for multi-dimensional time series data.

[0082] Steps S1310 - S1350 transform the historical data of a single parameter into an information flow with time order, solving the problem of expressing the dynamic characteristics of a single parameter; step S1360 establishes the time synchronization relationship between multi-parameters through timestamp alignment, solving the problem of spatio-temporal misalignment of multi-source data. The combination of single-parameter time serialization and timestamp alignment enables the key feature sequence TS seq to retain the time-dependent relationship of each parameter and construct the spatial correlation of multi-parameters at the same time point, providing a structured input for the self-attention mechanism of the subsequent Transformer model.

[0083] The performance of the gas turbine secondary air system is affected by multi-parameter coupling. For example, the adjustment of the turbine cooling valve opening V cool will immediately affect the pressure in the outer cylinder cavity of the turbine, while the change in the total inlet temperature T in of the compressor needs to pass through the compression process and lag to affect the outlet pressure P out . The construction of TS seq enables the model to learn such complex correlations across parameters and time. When TS seq is input into the first Transformer model, the self-attention mechanism can calculate the correlation between T in and P out at different time points, and the multi-head attention mechanism analyzes the local interactions of parameters such as V cool and P in in parallel, ultimately improving the model's fitting ability for the system's dynamic characteristics.

[0084] Through time serialization and timestamp alignment, the key feature sequence TS seq transforms discrete historical data into a continuous sequence at equal time intervals, facilitating the model to learn the time trend and periodic characteristics of parameters, such as capturing the regular decline in compressor efficiency during high-temperature periods in summer. Ensuring the one-to-one correspondence of multi-parameter states under the same timestamp, avoiding correlation analysis errors caused by time misalignment, such as accurately correlating t kThe change in the opening of the moment cooling valve and the simultaneous cavity pressure fluctuation. Form a multi-variable time series format suitable for processing by the Transformer model, enabling the model to utilize the self-attention mechanism to mine cross-parameter and cross-time dependencies, such as discovering the lag effect law of the compressor outlet temperature T out on the turbine cooling effect.

[0085] Step S2000, construct and train the first Transformer model according to the key feature sequence; obtain the real-time operation data sequence with timestamps of the current gas turbine secondary air system, and according to the real-time operation data sequence and the trained first Transformer model, obtain the turbine outer cylinder cavity pressure prediction value time series and the compressor discharge pressure prediction value time series

[0086] Furthermore, step S2000 includes:

[0087] Step S2100, construct and train the first Transformer model according to the key feature sequence, and the first Transformer model includes an input embedding layer, a position encoding layer, a self-attention mechanism, a multi-head attention mechanism, a feed-forward neural network, and an output layer;

[0088] Specifically, step S2100 aims to construct and train the first Transformer model using the key feature sequence. Through a series of complex and orderly operations, the model can learn the internal laws in the operation data of the gas turbine secondary air system, and then achieve accurate prediction of the turbine outer cylinder cavity pressure and the compressor discharge pressure. This process solves the problem that traditional models are difficult to capture dynamic interaction relationships when processing multi-parameter time series data, and improves the model's fitting ability for the non-linear behavior of complex systems through structured feature input and deep neural network architecture.

[0089] Furthermore, step S2100 includes:

[0090] Step S2110, extract the model input feature vector TS seq from each element TS seq,k of the key feature sequence TS in,k =(T in,k ,P in,k ,P out,k ,T out,k ,V cool,k );

[0091] Step S2120, input the model input feature vector TS in,k into the input embedding layer for encoding to obtain the embedded feature representation E k at the timestamp tk ;

[0092] The input of the model input feature vector TS in,k into the input embedding layer for encoding includes:

[0093] The model input feature vector TS is mapped to a high-dimensional space representation through a linear transformation: in,k :

[0094] E k = W e ·TS in,k + b e ;

[0095] where W e is the embedding matrix and b e is the bias term.

[0096] Step S2130: Obtain the position encoding vector P k at time stamp t k ;

[0097] The position encoding vector P k = [P k,0 , P k,1 , P k,2 , P k,3 , …, P k,2d-1 , where d represents the dimension of the position encoding vector, represents the value of the position encoding vector at the j * -th dimension, and j * is the dimension index in the position encoding vector, 0 ≤ j * ≤ 2d - 1; the values of the even dimensions are calculated by the sine function, i.e., The values of the odd dimensions are calculated by the cosine function, i.e., where i * is the dimension index in the position encoding vector.

[0098] Step S2140: Add the position encoding vector P k to the embedded feature representation E k to obtain the model input feature representation Z k at time stamp t k = E k + P k ; The model input feature representation Z k contains five input sub-features, namely the total inlet temperature feature T' k at time stamp t in,k , the static pressure feature P' k of the inlet wall at time stamp t in,k, the outlet pressure characteristic P' at time stamp t k ; out,k , the outlet temperature characteristic T' at time stamp t k ; out,k and the opening characteristic V' of the turbine cooling valve at time stamp t k ; cool,k ;

[0099] Step S2150, obtain the input sequence Z = {Z1, Z2, …, Z k} according to the model input feature representation Z, where 1 ≤ k ≤ n; n

[0100] Step S2160, calculate the correlation between the model input feature representations corresponding to different time stamps in the input sequence Z through the self-attention mechanism in the first Transformer model to obtain a one-dimensional feature correlation matrix;

[0101] Step S2170, the multi-head attention mechanism in the first Transformer model calculates the relationship between each input sub-feature in the model input feature representation Z k through n2 attention heads in parallel to obtain a multi-head attention weight matrix;

[0102] The n2 attention heads at least include head1 = (T' in,k , P' out,k ), head2 = (V' cool,k , P' in,k ) and head3 = (V' cool,k , T' out,k );

[0103] Step S2180, perform a non-linear transformation on the one-dimensional feature correlation matrix and the multi-head attention weight matrix through a feed-forward neural network, and the output layer maps the output of the feed-forward neural network to the outer cylinder cavity pressure and the compressor discharge pressure of the turbine.

[0104] Specifically, in step S2110, extract the model input feature vector TS seq from each element TS seq,k in the key feature sequence TS, that is, extract the model input feature vector TS in,k = (T in,k , P in,k , P out,k , T out,k , V cool,k ), that is, extract the time stamp t k ​Five key parameters: the total inlet temperature of the compressor, the static pressure of the inlet wall, the outlet pressure, the outlet temperature, and the opening of the turbine cooling valve. This operation converts the multivariate time series data into a vector form that can be processed by the model, solving the problem of mismatch between the original data format and the model input requirements. In step S2120, the TS in,k is linearly transformed through the input embedding layer to obtain the embedded feature representation E k = W e ·TS in,k + b e , where W e is the embedding matrix and b e is the bias term. W e and b e are preset parameters that determine the mapping method and effect. This transformation maps the original features to a high-dimensional space, enhancing the expressive power of the features. For example, if the original feature vector is five-dimensional, the embedding matrix W e can be set to 5×d dimensions (d is the embedding dimension). Through linear transformation, the scalar value of each parameter is converted into a d-dimensional vector, enabling the features of different parameters to form distinguishable distributions in the high-dimensional space. This operation solves the problem of model training deviation caused by dimensional differences in the original data. For example, the total inlet temperature of the compressor (unit: K) and the opening of the turbine cooling valve (unit: %) achieve scale unity through embedding coding, facilitating the model to learn the correlation between parameters.

[0105] The purpose of position encoding in step S2130 is to provide the model with the sequential information of the time series, solving the problem that the Transformer model itself does not have the ability to perceive time series. For example, for timestamps t1 and t2, the corresponding position encoding vectors P k distinguish the order through different sine and cosine values, enabling the model to recognize the sequence relationship between "t1" and "t2", thereby capturing the time dependence of parameter changes. For example, the lag effect of the increase in the total inlet temperature of the compressor at time t1 on the outlet pressure at time t2. In step S2140, the position encoding vector P k is added to the embedded feature E k to obtain the model input feature representation Z k = E k + P k . This operation deeply fuses the time sequence information with the parameter features, forming an input representation containing spatio-temporal information. For example, if E k represents the parameter feature vector at time t1 and P k represents the position encoding of t1, then Z kIt contains both the parameter status and time position at this moment, enabling the model to distinguish the dependency relationships at different time points when processing the features at subsequent time point t2. For example, it can determine the influence degree of the change in the opening of the turbine cooling valve at time t1 on the cavity pressure of the turbine outer cylinder at time t2. The input sequence Z = {Z1, Z2, …, Z n} forms a sequence structure suitable for Transformer processing. This sequence preserves the time order and spatial correlation of key feature parameters, providing an input basis for the self-attention mechanism. For example, the input sequence Z can be regarded as an n×d-dimensional matrix, where each row corresponds to the spatio-temporal fusion features at a time point, enabling the model to calculate the correlation of features at different time points globally. The elements in the single-dimensional feature correlation matrix represent the correlation strength between the features of two time points. Step S2170 calculates the relationships between input sub-features in parallel through n2 attention heads. For example, head 1 focuses on the total inlet temperature and outlet pressure (T' in,k , P' out,k ), head 2 focuses on the opening of the turbine cooling valve and the static pressure of the inlet wall surface (V' cool,k , P' in,k ), and head 3 focuses on the opening of the turbine cooling valve and the outlet temperature (V' cool,k , T' out,k ). This design solves the problem of capturing the complex interaction relationships between multiple parameters. For example, the self-attention mechanism can discover the cross-time correlation between the increase in the total inlet temperature at time t1 and the decrease in the outlet pressure at time t3, and the multi-head attention captures the local interactions of different parameter pairs in parallel (such as the immediate impact of the adjustment of the cooling valve opening on the inlet static pressure). Step S2180 inputs the single-dimensional feature correlation matrix and the multi-head attention weight matrix into the feed-forward neural network, extracts high-order features through non-linear transformation, and finally maps them to the predicted values of the cavity pressure of the turbine outer cylinder and the exhaust pressure of the compressor by the output layer. The non-linear transformation ability of the feed-forward network solves the problem of modeling the non-linear behavior of the system. For example, the non-linear relationship between the outlet pressure of the compressor, the total inlet temperature, and the opening of the cooling valve is realized through a multi-layer perceptron structure to achieve complex mapping. The output layer uses a regression model to convert the high-dimensional features into specific pressure prediction values, ensuring the dimensional consistency with the actual physical quantities.

[0106] Step S1300 provides structured multi-parameter input through time serialization and timestamp alignment. Step S2100 realizes the in-depth processing of features through embedding encoding, position encoding, and the attention mechanism. The fusion of position encoding and embedded features enables the model to perceive the time order and capture the parameter correlations across time points in combination with the self-attention mechanism. For example, it can accurately identify the impact of the outlet pressure at 3 time points after the increase in the total inlet temperature of the compressor, solving the limitation of the traditional model relying on fixed time window analysis. The multi-head attention mechanism processes different parameter pairs in parallel (such as V' cool,k and P' in,kThe association) captures local interactions and global dependencies, such as discovering the immediate feedback mechanism of the opening adjustment of the turbine cooling valve on the static pressure of the intake system, and improving the model's fitting ability for complex coupling relationships. The non-linear transformation of the feed-forward neural network adapts to the non-linear characteristics of the gas turbine system, such as the non-linear curve of the compressor efficiency varying with temperature, and achieves precise mapping through multiple activation functions, significantly improving the prediction accuracy compared with the linear model.

[0107] Step S2200: Obtain the real-time operation data sequence including time stamps of the current gas turbine secondary air system, and obtain the predicted value time sequence of the outer cylinder cavity pressure of the turbine according to the real-time operation data sequence and the trained first Transformer model and the predicted value time sequence of the compressor discharge pressure

[0108] Furthermore, step S2200 includes:

[0109] Step S2210: Extract the real-time total temperature at the compressor inlet the real-time static pressure on the wall surface at the compressor inlet the real-time pressure at the compressor outlet the real-time temperature at the compressor outlet and the real-time opening of the turbine cooling valve

[0110] Step S2220: Respectively perform time serialization processing on and to obtain the real-time total temperature sequence at the compressor inlet the real-time static pressure sequence on the wall surface at the compressor inlet the real-time pressure sequence at the compressor outlet the real-time temperature sequence at the compressor outlet and the real-time opening sequence of the turbine cooling valve

[0111] Step S2230: Align and according to the time stamp to construct the real-time key feature vector set TN seq =(TN seq,1 , TN seq,2 ,…, TN seq,m ), where TN seq,j is the real-time key feature vector at the j-th time point, m is the total number of time points, and 1≤j≤m; Among them, is the real-time total temperature at the compressor inlet at the j-th time point, is the real-time static pressure on the wall surface at the compressor inlet at the j-th time point, is the real-time pressure at the compressor outlet at the j-th time point, is the real-time compressor outlet temperature at the j-th time point, is the real-time turbine cooling valve opening at the j-th time point;

[0112] Step S2240: Input TN seq,j into the trained first Transformer model to obtain the predicted value of the outer cylinder cavity pressure of the turbine at the j-th time point and the predicted value of the compressor exhaust pressure at the j-th time point

[0113] Step S2250: According to obtain the time series of the predicted values of the outer cylinder cavity pressure of the turbine According to obtain the time series of the predicted values of the compressor exhaust pressure

[0114] Specifically, step S2200 aims to process the real-time operation data of the current gas turbine secondary air system based on the trained first Transformer model to generate the time series of the predicted values of the outer cylinder cavity pressure of the turbine and the compressor exhaust pressure. This step solves the problems of format matching between real-time operation data and model input, and real-time analysis of multi-parameter coupling relationships by extracting real-time key feature parameters, constructing a set of feature vectors that conform to the model input format, and using the trained model to achieve dynamic prediction. Step S2210 aims to extract five key parameters closely related to the performance of the secondary air system from the current operation data stream of the gas turbine. The real-time operation data sequence is collected through a distributed sensor network and contains a triple of timestamp, parameter type, and measurement value. The real-time total temperature at the compressor inlet reflects the total temperature of the air at the compressor inlet and is a key boundary condition affecting air density and compression efficiency; the real-time static pressure on the wall at the compressor inlet characterizes the static pressure at the compressor inlet wall and is used to evaluate the influence of the intake system resistance and ambient air pressure; the real-time compressor outlet pressure is the pressure value of the air compressed by the compressor and directly determines the driving force and flow distribution of the cooling air; the real-time compressor outlet temperature is the temperature of the air at the compressor outlet, reflects the thermal effect of the compression process, and affects the cooling capacity of the secondary air; the real-time turbine cooling valve opening is the real-time adjustment state of the turbine cooling valve and determines the flow rate of the secondary air distributed to the outer cylinder cavity of the turbine. The above parameters are collected in real time by sensors installed at key positions of the gas turbine, such as temperature sensors, pressure sensors, and valve opening sensors, to ensure the real-time and accuracy of the data.

[0115] Step S2220 performs time serialization processing on the five extracted real-time parameters respectively, that is, according to the time stamps of data acquisition, the measured values of the same parameter at consecutive time points are arranged as a time series. This processing solves the problem of expressing the time order of real-time data, enabling the dynamic change trends of each parameter to be captured by the model. Step S2230 aligns the five serialized real-time parameter sequences according to the time stamps to construct the real-time key feature vector set TN seq , this step is consistent with the construction logic of the key feature sequence in step S1300, ensuring the unified format of real-time data and historical training data, which is convenient for model processing. By aligning the time stamps, the problem of asynchronous multi-sensor data is solved. For example, it avoids misjudgment of the coupling relationship between the two due to the misalignment of the acquisition times of the compressor outlet pressure and the turbine cooling valve opening. Step S2240 inputs TN seq,j into the trained first Transformer model. Each feature vector TN seq,j first undergoes a linear transformation through the input embedding layer to be mapped to a high-dimensional feature space, generating an embedded feature representation; subsequently, a positional encoding vector is superimposed to endow time order information, which is consistent with the feature processing method during model training in step S2100. The model calculates the correlation of feature vectors at different time points through the self-attention mechanism to capture the dynamic interaction between parameters (such as the immediate impact of the change in the turbine cooling valve opening on the compressor outlet pressure); the multi-head attention mechanism analyzes the local associations of different parameter pairs in parallel (such as the cross-time dependence between the inlet total temperature and the outlet temperature), and finally maps to specific pressure prediction values through the feed-forward neural network and the output layer. The prediction results for each time point are aggregated to generate a complete prediction time series.

[0116] Steps S2210 - S2230 and the methods of feature extraction, time serialization, and alignment of historical data in step S1300 form a unified data processing flow, ensuring the format compatibility between real-time data and training data. For example, the five-tuple structure of the real-time key feature vector set TN seq is the same as that of the historical key feature sequence TS seqConsistent, enabling the first Transformer model to process real-time inputs without additional adjustment, avoiding prediction biases caused by data format differences, and enhancing the model's generalization ability and real-time response speed. By aligning timestamps to construct real-time key feature vectors, the problem of clock deviation that may exist in the multi-sensor data acquisition of gas turbines is solved (such as inconsistent acquisition times due to hardware differences among different sensors). For example, if there are slight differences in the timestamps of the compressor outlet pressure sensor and the turbine cooling valve opening sensor, the alignment operation ensures that the parameter data at the same time point strictly corresponds, enabling the model to accurately capture the immediate coupling relationship between parameters (such as the changing trend of cavity pressure during the same period when the cooling valve opening increases), avoiding prediction errors caused by spatio-temporal misalignment, and improving the accuracy of real-time prediction. The application of time serialization processing and positional encoding preserves the time-dependent information of real-time data. For example, the dynamic changes (such as gradually rising with the increase of ambient temperature) of the real-time compressor inlet total temperature sequence are transmitted to the model through positional encoding. Combining the self-attention mechanism for correlation calculation of historical time points, the model can learn the lagging effect of the change in inlet total temperature on the outlet pressure (such as at 2 time points after the inlet total temperature increases, the outlet pressure slightly decreases due to the decrease in air density), thereby more accurately predicting future pressure changes and solving the problem that traditional models are difficult to capture long-distance time series dependencies. The generated predicted value time series and the measured value time series are strictly aligned in the time dimension, providing direct comparison data for residual calculation and performance evaluation in the subsequent step S3000. For example, the predicted value of the turbine outer cylinder cavity pressure and the measured value

[0117] at the jth time point directly reflect the fitting degree of the model to the current working condition, and the trend analysis of the long-term predicted value sequence (such as continuous increase or decrease) can quantify the degradation direction of the system performance, providing data support for operation and maintenance decisions. cyl and the measured value time series P comp of the compressor discharge pressure, and obtain the current residual time series of the performance ratio according to P cyl , P comp , and ; Based on the current residual time series and the pre-constructed second Transformer model, predict the residual time series for the next Y time points The residual time series contains Y predicted residual values; According to the predicted residual time series

[0118] Further, step S3000 includes:

[0119] Step S3100, obtaining the measured value time series P of the turbine outer cylinder cavity compaction of the current gas turbine cyl =(P cyl,1 ,P cyl,2 ,…,P cyl,m ) and the measured value time series P of the compressor exhaust pressure comp =(P comp,1 ,P comp,2 ,…,P comp,m ), and obtaining the current residual time series of the performance ratio according to P cyl , P comp , and ; where P cyl,m is the measured value of the m-th turbine outer cylinder cavity compaction, and P comp,m is the measured value of the m-th compressor exhaust pressure;

[0120] The obtaining the current residual time series of the performance ratio according to P cyl , P comp , and includes:

[0121]

[0122] Where:

[0123] r j : the residual value at the j-th time point, 1≤j≤m;

[0124] P cyl,j : the measured value of the turbine outer cylinder cavity compaction at the j-th time point;

[0125] P comp,j : the measured value of the compressor exhaust pressure at the j-th time point;

[0126] According to r j constitute the current residual time series r=(r1, r2, …, r m ) of the performance ratio.

[0127] Specifically, step S3100 aims to calculate the current residual time series of the performance ratio by obtaining the measured time series of the turbine outer cylinder cavity pressure and the measured time series of the compressor discharge pressure in the secondary air system of the gas turbine, and combining with the previously obtained predicted values, which is used as the key basis for evaluating the current performance state of the system. By calculating the residual time series, the deviation degree between the model predicted value and the actual measured value can be intuitively reflected. The larger the residual, the higher the degree of deviation of the system performance from the healthy state. For example, if the residual value continues to increase over a period of time, it means that the gap between the predicted value and the actual value of the model is constantly expanding, which may imply that there are abnormal conditions in the secondary air system of the gas turbine, such as equipment component wear, operating parameter drift, etc. By real-time monitoring the residual time series, operators can timely discover potential problems in the system, take corresponding maintenance measures in advance, avoid the occurrence of system failures, and ensure the safe and stable operation of the gas turbine. In addition, the residual time series also provides an important data basis for subsequent evaluation of the system performance degradation degree, helps to further analyze the change trend of the system performance, and provides strong support for the formulation of the equipment maintenance plan and the optimization of operation.

[0128] Step S3200, based on the current residual time series and the pre-constructed second Transformer model, predicts the residual time series at the next Y time points The residual time series contains Y predicted residual values;

[0129] Specifically, the construction of the second Transformer model is similar to that of the first Transformer model, and also includes components such as an input embedding layer, a positional encoding layer, a self-attention mechanism, a multi-head attention mechanism, a feed-forward neural network, and an output layer. During the construction process, the parameters of the model, such as the number of network layers, the dimension of the hidden layer, and the number of attention heads, need to be reasonably determined according to the characteristics of the residual time series data and the requirements of the prediction task. For example, according to the length and complexity of the historical residual data, the appropriate dimension d of the positional encoding vector is determined to ensure that the model can accurately capture the position information in the time series.

[0130] When training the second Transformer model, the historical residual time series data is used as the training set. The model improves its prediction ability for future residuals by learning the patterns and trends in the historical residual time series data. During the training process, an optimization algorithm is used to continuously adjust the parameters of the model to minimize the error between the model predicted value and the actual residual value. The optimization objective is usually to minimize the mean square error (MSE) or cross-entropy loss, etc. The gradient is calculated through the backpropagation algorithm, and the weights and biases of the model are updated. Through multiple iterative trainings, the model gradually learns the pattern of the residual changing over time.

[0131] After the model training is completed, the current residual time series is used as the input, and the position encoding layer in the model is utilized to incorporate the time information of each residual data point into the model input. The position encoding is calculated through specific sine and cosine functions, assigning unique encodings to the residual data at each time point, enabling the model to understand the time sequence and the dependency relationships between the residual data at different time points.

[0132] Next, the self-attention mechanism comes into play, calculating the mutual influence between each time point in the input data, thereby learning the trend patterns among the historical residual data. The self-attention mechanism determines the importance of each time point for predicting future residuals by calculating the attention weights of the residual data at different time points. For example, for a sequence containing residual data at multiple time points, the self-attention mechanism can automatically identify which time points' residual data are more crucial for future predictions, thus capturing the trend of residual changes more accurately.

[0133] Based on the results of the self-attention calculation, the feed-forward neural network further processes the data, extracting more advanced features through non-linear transformations. Finally, the output layer of the model generates predicted values for the residuals at the next Y time points, forming a predicted residual time series that contains Y predicted residual values.

[0134] By predicting the future residual time series, the changing trend of the performance of the gas turbine secondary air system can be predicted in advance. For example, if it is predicted that the future residuals gradually increase, it indicates that the system performance may continue to decline. Operators can arrange equipment maintenance in advance, adjust operating parameters, etc., to avoid equipment failures and production interruptions caused by the deterioration of system performance. This helps to improve the reliability and availability of the equipment, reduce maintenance costs, and ensure the efficient operation of the gas turbine. At the same time, the prediction results also provide a scientific basis for formulating the equipment maintenance plan, making the maintenance work more targeted and forward-looking, and improving the efficiency and scientific nature of equipment management.

[0135] Step S3300, according to the predicted residual time series at the next Y time points Quantitatively evaluate the degree of performance degradation of the current gas turbine secondary air system and generate a health assessment report.

[0136] Furthermore, as Figure 3 shown, step S3300 includes:

[0137] Step S3310, perform linear regression to calculate the trend slope k';

[0138] The calculation of the trend slope k' includes:

[0139]

[0140] Where:

[0141] Y: Predicted number of future time points;

[0142] i: Time point index from 1 to Y;

[0143] Predicted residual value at time m + i;

[0144] Linear regression reveals the trend of data changes by establishing a linear relationship model between variables. The slope k' depends on the predicted residuals and the time index i. As increases or as time i progresses, k' will change accordingly. The slope k' represents the rate of change of the residuals over time. If k' is positive, it indicates that the residuals are increasing, which may mean that the system performance is degrading. If k' is negative, it means that the residuals are decreasing and the system performance has a tendency to improve; if k' is close to zero, it indicates that the system performance is relatively stable. By calculating the trend slope, the rate of system performance degradation can be quantified, providing an important indicator for evaluating system performance changes. This helps operators to timely understand the trend of system performance changes and take measures in advance to deal with possible problems, such as adjusting equipment operation parameters, arranging equipment maintenance, etc., to ensure the stable operation of the system.

[0145] Step S3320, calculate the cumulative value S of the predicted residual values for the next Y time points;

[0146]

[0147] Specifically, step S3320 aims to calculate the cumulative value S of the predicted residual values for the next Y time points. This cumulative value is a key indicator for evaluating the total amount of performance degradation of the gas turbine secondary air system, providing an important basis for comprehensively understanding system performance changes. The cumulative value S comprehensively considers the overall situation of the deviation between system performance and health status over a future period of time. The principle is that each predicted residual value reflects the difference between the predicted value of system performance and the theoretical value in the healthy state at the corresponding time point. By adding up these differences, a comprehensive indicator can be obtained to measure the overall degree of system performance degradation within the next Y time points. When the cumulative value S is larger, it means that during this prediction time, the degree of deviation of system performance from the healthy state is more serious, and the total amount of performance degradation is more significant. For example, if the S value gradually increases from a relatively small value, it indicates that the system performance is deteriorating over time, and there may be problems such as increased component wear and gradually deviated operating parameters from the optimal values.

[0148] From the perspective of system maintenance and management, the cumulative value S provides an intuitive and quantitative indicator for operation and maintenance personnel. By continuously monitoring the change of the S value, operation and maintenance personnel can more comprehensively understand the long-term change trend of system performance. If the S value approaches or exceeds a pre-set threshold, this warns operation and maintenance personnel that they need to conduct a more in-depth inspection and maintenance of the system, such as arranging equipment maintenance, adjusting operation parameters, etc., to prevent the further deterioration of system performance, ensure the stable operation of the gas turbine, and reduce the risk of downtime and maintenance costs caused by equipment failures.

[0149] Step S3330, calculate the performance degradation index PDI according to the trend slope k' and the cumulative value S;

[0150]

[0151] Where α is the weight coefficient of the trend slope, and β is the weight coefficient of the ratio of the cumulative value S to the predicted number of future time points Y;

[0152] Specifically, the trend slope k' reflects the change rate of the residual over time and the speed of system performance degradation; the cumulative value S comprehensively considers the total amount of system performance degradation over a period of time in the future. By combining the two, PDI can more comprehensively reflect the degradation of system performance. The settings of the weight coefficients α and β enable the impact degrees of the trend slope and the cumulative value to be adjusted according to the actual situation during the evaluation process; α and β are calibrated through expert experience; for example, in some application scenarios that are more sensitive to the speed of performance degradation, the value of α can be appropriately increased to highlight the impact of the trend slope on the evaluation result; while in cases where more attention is paid to the total amount of performance degradation, the proportion of β can be increased.

[0153] As a comprehensive evaluation indicator, the advantage of PDI is that it can integrate multiple factors reflecting system performance degradation into a quantitative value, providing an intuitive and unified evaluation standard for operation and maintenance personnel. By comparing PDI with a pre-set threshold, operation and maintenance personnel can quickly judge the performance status of the system, determine whether corresponding maintenance measures need to be taken, and the degree of maintenance measures to be taken. This helps to improve the pertinence and effectiveness of equipment maintenance, reasonably allocate maintenance resources, avoid over-maintenance or under-maintenance, and thus ensure the safe, stable and efficient operation of the gas turbine.

[0154] Step S3340, classify the health level of the performance of the current gas turbine secondary air system according to the performance degradation index PDI; and put forward maintenance suggestions according to the health level;

[0155] PDI < θ1: healthy state, where θ1 is the health threshold of the secondary air system;

[0156] θ1 ≤ PDI < θ2: Warning state, where θ2 is the warning threshold of the secondary air system;

[0157] PDI ≥ θ2: Dangerous state.

[0158] Specifically, when it is in the healthy state, the system performance is stable and no additional intervention is required; when it is in the warning state, it is recommended to increase the monitoring frequency and closely monitor the system performance changes; when it is in the dangerous state, it is necessary to immediately stop the machine for maintenance to avoid serious system failures. The healthy threshold θ1 and the warning threshold θ2 of the secondary air system are set by those skilled in the art according to experience. Preferably, θ1 = 0.3 and θ2 = 0.7.

[0159] The health level classification method based on PDI can help the operation and maintenance personnel perform equipment maintenance according to the regular maintenance plan, reasonably allocate maintenance resources, and reduce unnecessary maintenance costs. For example, when the gas turbine is in the healthy state, the interval time of equipment inspection can be appropriately extended, reducing the workload of maintenance personnel while ensuring the normal operation of the equipment. For a system in the warning state, increasing the monitoring frequency can timely capture the subtle changes in system performance so as to take measures before the problem deteriorates. By more frequently monitoring key parameters such as the total inlet temperature and outlet pressure of the compressor, the operation and maintenance personnel can more accurately judge whether the system is developing towards the dangerous state, formulate maintenance plans in advance, such as adjusting operation parameters and arranging preventive maintenance, to avoid the occurrence of system failures and reduce the economic losses caused by downtime. For a system in the dangerous state, immediately stopping the machine for maintenance is the key measure to ensure equipment safety and avoid major accidents. In this case, timely conducting a comprehensive inspection and maintenance of the equipment, replacing damaged components, and adjusting system parameters can restore the system to the normal operation state, ensuring the reliability and safety of the gas turbine.

[0160] Through the clear health level classification and targeted maintenance suggestions, this method can help the operation and maintenance personnel respond to system performance changes in a timely and accurate manner, effectively improve the reliability and operation efficiency of the secondary air system of the gas turbine, reduce the equipment failure rate, extend the equipment service life, and provide a strong guarantee for the stable operation of the gas turbine.

[0161] Step S3350, generate a health assessment report, and the health assessment report includes the residual time series Trend slope k', the cumulative value S of the predicted residual values at the next Y time points, the performance degradation index PDI, and the health level of the current secondary air system of the gas turbine.

[0162] Specifically, the residual time series This records the time-varying differences between predicted system performance at Y future time points and the theoretical values in a healthy state. By analyzing this series, operations and maintenance personnel can visually observe the degree of deviation in system performance at each time point, as well as the changing trend of this deviation. For example, if the values in the residual time series gradually increase, it indicates that system performance is continuously deteriorating; if the values fluctuate, it may indicate system instability, requiring further analysis.

[0163] The trend slope k' represents the rate of change of the residual over time, reflecting the rate of system performance degradation. As mentioned earlier, a positive k' indicates an increasing residual and a downward trend in system performance; its value reflects the speed of performance degradation. By analyzing the trend slope, operations and maintenance personnel can determine the urgency of system performance changes. For example, a large k' value indicates a rapid system performance degradation, necessitating prompt intervention to prevent system failure.

[0164] The cumulative value S combines the predicted residual values at Y future time points, measuring the total amount of system performance degradation. A larger S value indicates that system performance has deviated significantly from a healthy state during this period, reflecting long-term system performance degradation. For example, when S exceeds a certain threshold, even if the system is still able to operate, the risk of long-term equipment failure is high, and comprehensive inspection and maintenance are required.

[0165] The Performance Degradation Index (PDI) is a quantitative indicator derived from the combined trend slope and cumulative value. It comprehensively reflects the degree of system performance degradation. Using the PDI, operations and maintenance personnel can quickly determine the system's health level and, consequently, decide on the appropriate maintenance strategy. When the PDI is within the healthy range, routine maintenance can be maintained; when in a warning state, monitoring frequency should be increased; and when in a critical state, the system should be immediately shut down for maintenance.

[0166] The health level of a gas turbine's secondary air system is determined by comparing the PDI with a set threshold. This provides operators with an intuitive indicator of system performance. The health level clearly informs operators of the system's current condition, whether action is required, and what action to take.

[0167] The health assessment report integrates this information, providing a comprehensive and systematic reference for the operation and maintenance management of gas turbines. Operation and maintenance personnel can formulate targeted maintenance plans based on the report content, reasonably arrange maintenance resources, timely detect potential problems and take corresponding measures, thereby ensuring the safe and stable operation of gas turbines, improving the reliability and service life of equipment, reducing operating costs, and reducing the risk of production interruption caused by equipment failures. For example, when formulating a monthly maintenance plan, the operation and maintenance team can, according to the health assessment report, arrange different levels of maintenance work for gas turbines in different health grades. For equipment in good health, conduct regular inspections; for equipment in the warning state, increase key monitoring items; for equipment in the dangerous state, give priority to arranging a comprehensive overhaul, so as to achieve efficient and scientific equipment management.

[0168] In addition, parts of the above technical solutions provided in the embodiments of the present application that are consistent with the corresponding technical solutions in the prior art in terms of implementation principles are not described in detail to avoid excessive elaboration.

[0169] As described in the specific embodiments above, the purpose, technical solutions and beneficial effects of the present invention have been further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A performance prediction method for the secondary air system of a gas turbine based on a transformer, characterized in that, The method includes: Obtaining historical operation data including timestamps under the steady-state condition of the secondary air system of a healthy gas turbine, and preprocessing the historical operation data; extracting key feature parameters from the preprocessed historical operation data, and time-series serializing the key feature parameters according to the timestamps to obtain key feature sequences; Construct and train a first Transformer model according to the key feature sequence; obtain a real-time operation data sequence including timestamps of the current gas turbine secondary air system, and obtain a predicted value time series of the turbine outer cylinder cavity pressure according to the real-time operation data sequence and the trained first Transformer model and a predicted value time series of the compressor discharge pressure Obtain the time series P of the measured values of the turbine outer cylinder cavity compaction of the current gas turbine cyl and the time series P of the measured values of the compressor exhaust pressure comp , and according to P cyl , P comp , and obtain the current residual time series of the performance ratio; based on the current residual time series and the pre-built second Transformer model, predict the residual time series at the next Y time points According to quantitatively evaluate the performance degradation degree of the secondary air system of the current gas turbine and generate a health assessment report.

2. The performance prediction method of the secondary air system of a gas turbine based on transformer according to claim 1, wherein The key characteristic parameters include the total inlet temperature T of the compressor in , the static pressure P of the inlet wall of the compressor in , the outlet pressure P of the compressor out , the outlet temperature T of the compressor out and the opening V of the turbine cooling valve cool ; The obtaining of the key feature sequences includes: Perform time serialization processing on the total inlet temperature T of the compressor according to the timestamp, and obtain the time series T of the total inlet temperature of the compressor in ; Perform time serialization processing on the static pressure P of the inlet wall of the compressor according to the timestamp, and obtain the time series P of the static pressure of the inlet wall of the compressor in-seq ; Perform time serialization processing on the outlet pressure P of the compressor according to the timestamp, and obtain the time series P of the outlet pressure of the compressor in ; Perform time serialization processing on the outlet temperature T of the compressor according to the timestamp, and obtain the time series T of the outlet temperature of the compressor in-seq ; Perform time serialization processing on the outlet temperature T of the compressor according to the timestamp, and obtain the time series T of the outlet temperature of the compressor out ; Perform time serialization processing on the outlet pressure P of the compressor according to the timestamp, and obtain the time series P of the outlet pressure of the compressor out-seq ; Perform time serialization processing on the outlet temperature T of the compressor according to the timestamp, and obtain the time series T of the outlet temperature of the compressor out ; Perform time serialization processing on the outlet temperature T of the compressor according to the timestamp, and obtain the time series T of the outlet temperature of the compressor out-seq ; Perform time serialization processing on the opening V of the turbine cooling valve according to the timestamp, and obtain the time series V of the opening of the turbine cooling valve cool ; Perform time serialization processing on the opening V of the turbine cooling valve according to the timestamp, and obtain the time series V of the opening of the turbine cooling valve cool-seq ; Align T in-seq , P in-seq , P out-seq , T out-seq and V cool-seq according to the timestamp to construct the key feature sequence TS seq =(TS seq,1 , TS seq,2 ,…, TS seq,n ), where TS seq,k is the k-th element in TS seq , n is the total number of elements in TS seq , and 1 ≤ k ≤ n.

3. The performance prediction method for the secondary air system of a gas turbine based on a transformer according to claim 2, characterized in that, The said TS seq Each element in it is a six-tuple: TS seq,k =(t k ,T in,k ,P in,k ,P out,k ,T out,k ,V cool,k ); Among them, t k is the timestamp of the k-th element in TS seq , T in,k is the total inlet temperature of the compressor at timestamp t k , P in,k is the static pressure of the compressor inlet wall at timestamp t k , P out,k is the outlet pressure of the compressor at timestamp t k , T out,k is the outlet temperature of the compressor at timestamp t k , V cool,k is the opening degree of the turbine cooling valve at timestamp t k .

4. The performance prediction method for the secondary air system of a gas turbine based on transformer according to claim 3, wherein The first Transformer model includes an input embedding layer, a positional encoding layer, a self-attention mechanism, a multi-head attention mechanism, a feed-forward neural network, and an output layer; The constructing and training of the first Transformer model includes: Extract the model input feature vector TS from each element TS of the key feature sequence TS seq of seq,k the key feature sequence TS in,k =(T in,k , P in,k , P out,k , T out,k , V cool,k ); Input the model input feature vector TS in,k into the input embedding layer for encoding to obtain the timestamp t k under the embedded feature representation E k ; Obtain the timestamp t through the positional encoding layer k of the positional encoding vector P k ; Add the position encoding vector P k to the embedded feature representation E k , to obtain the model input feature representation Z k at the timestamp t k ; According to the model input feature representation Z k obtain the input sequence Z = {Z1, Z2, …, Z n}.

5. The performance prediction method of the secondary air system of a gas turbine based on a transformer according to claim 4, characterized in that, The constructing and training of the first Transformer model further includes: Calculating the correlation between the model input feature representations corresponding to different timestamps in the input sequence Z through the self-attention mechanism in the first Transformer model to obtain a one-dimensional feature correlation matrix; The multi-head attention mechanism in the first Transformer model calculates the relationships between the input sub-features in the model input feature representation Z in parallel through n2 attention heads, and obtains the multi-head attention weight matrix; k ​ Non-linearly transforming the one-dimensional feature correlation matrix and the multi-head attention weight matrix through a feed-forward neural network, and the output layer mapping the output of the feed-forward neural network to the turbine outer cylinder cavity pressure and the compressor discharge pressure.

6. The performance prediction method of the secondary air system of a gas turbine based on a transformer according to claim 4, characterized in that, The position encoding vector P k = [P k,0 , P k,1 , P k,2 , P k,3 , …, P k,2d-1 , where d represents the dimension of the position encoding vector, represents the value of the position encoding vector on the j * -th dimension, and j * is the dimension index in the position encoding vector, 0 ≤ j * ≤ 2d - 1; the values of even dimensions are calculated by the sine function, and the values of odd dimensions are calculated by the cosine function.

7. The performance prediction method of the secondary air system of a gas turbine based on a transformer according to claim 5, wherein The model input feature representation Z k contains five input sub - features, namely the total inlet temperature feature T' k at time stamp t in,k , the static wall pressure feature P' k at time stamp t in,k , the outlet pressure feature P' k at time stamp t out,k , the outlet temperature feature T' k at time stamp t out,k and the turbine cooling valve opening feature V' k at time stamp t cool,k .

8. The performance prediction method of the secondary air system of a gas turbine based on a transformer according to claim 1, characterized in that Obtaining a time series of predicted values of the turbine outer cylinder cavity pressure based on the real-time operation data series and the trained first Transformer model and a time series of predicted values of the compressor discharge pressure including: Extract the real-time total temperature at the compressor inlet from the real-time operating data series Real-time static pressure on the compressor inlet wall Real-time compressor outlet pressure Real-time compressor outlet temperature and the real-time opening of the turbine cooling valve Time series processing is performed on and respectively to obtain the real-time total inlet temperature series of the compressor the real-time static pressure series of the compressor inlet wall the real-time outlet pressure series of the compressor the real-time outlet temperature series of the compressor and the real-time opening degree series of the turbine cooling valve According to and the first Transformer model, a time series of predicted values of the turbine outer cylinder cavity pressure is obtained and a time series of predicted values of the compressor discharge pressure 9. The method for predicting the performance of the secondary air system of a gas turbine based on a transformer according to claim 8, wherein Said according to and the first Transformer model, obtain the time series of the predicted values of the turbine outer cylinder cavity pressure and the time series of the predicted values of the compressor exhaust pressure including: Will and According to the timestamp alignment, construct the real-time key feature vector set TN seq =(TN seq,1 ,TN seq,2 ,…,TN seq,m ), where TN seq,j is the real-time key feature vector at the jth time point, m is the total number of time points, 1≤j≤m; in, is the real-time compressor inlet total temperature at the jth time point, is the real-time static pressure of the compressor inlet wall at the jth time point, is the real-time compressor outlet pressure at the jth time point, is the real-time compressor outlet temperature at the jth time point, is the real-time turbine cooling valve opening at the jth time point; Input TN seq,j into the trained first Transformer model to obtain the predicted value of the turbine outer cylinder cavity pressure at the j-th time point and the predicted value of the compressor exhaust pressure at the j-th time point According to obtain the time series of predicted values of the outer cylinder cavity pressure of the turbine According to obtain the time series of predicted values of the compressor discharge pressure 10. The method for predicting the performance of the secondary air system of a gas turbine based on a transformer according to claim 5, characterized in that, The n2 attention heads at least include head 1 head1 = (T' in,k , P' out,k ), head 2 head2 = (V' cool,k , P' in,k ), and head 3 head3 = (V' cool,k , T' out,k ).

Images