Gas turbine speed prediction method

Abstract

The application provides a gas turbine rotating speed prediction method, which can be applied to the technical field of gas turbines. The method comprises the following steps: obtaining fuel quantity data corresponding to a plurality of operating conditions respectively in a predetermined historical operating period; splitting the fuel quantity data into fuel instantaneous data and fuel trend data; calling a trained rotating speed prediction model, wherein the rotating speed prediction model comprises a fuel instantaneous gated recurrent unit, a fuel trend gated recurrent unit, a fusion layer and a fully connected layer; inputting the fuel instantaneous data and the fuel trend data into the fuel instantaneous gated recurrent unit and the fuel trend gated recurrent unit of the rotating speed prediction model respectively, and outputting fuel instantaneous hidden features and fuel trend hidden features corresponding to the fuel instantaneous data and the fuel trend data respectively; inputting the fuel instantaneous hidden features and the fuel trend hidden features into the fusion layer, and outputting fusion features; and inputting the fusion features into the fully connected layer, and outputting the predicted rotating speed of the gas turbine.

Description

Technical Field

[0001] This application relates to the field of gas turbine technology, and more specifically, to a method for predicting gas turbine speed. Background Technology

[0002] A gas turbine is a high-performance power machine that drives a turbine mechanism to continuously perform work by generating a high-temperature gaseous working fluid. As a high-precision energy power engine with the highest cycle temperature, the gas turbine's high efficiency, compactness, and flexibility make it irreplaceable in many fields. For example, in energy systems, it is the core equipment of the most efficient combined cycle power generation system, and can form a distributed cogeneration model, further improving energy utilization efficiency and adapting to new energy systems. In industrial production, gas turbines are widely used to drive the operation of booster units in long-distance natural gas / oil pipelines, ensuring the lifeline of energy transportation. In the shipbuilding field, the gas turbine is the core of a ship's power and a key piece of equipment determining its performance. Digital analysis can better understand the operating status of gas turbines. However, there are still shortcomings in related technologies for digitally analyzing the speed of gas turbines. Summary of the Invention

[0003] In view of this, embodiments of this application provide a gas turbine speed prediction method, apparatus, electronic device, storage medium, and program product.

[0004] One aspect of this application provides a gas turbine speed prediction method, comprising: acquiring fuel quantity data corresponding to various operating conditions of the gas turbine during a predetermined historical operating cycle; splitting the fuel quantity data into instantaneous fuel data and fuel trend data, wherein the instantaneous fuel data represents the instantaneous value of the gas turbine's fuel quantity at a predetermined monitoring time during the historical operating cycle, and the fuel trend data represents the changing trend of the gas turbine's fuel quantity during the historical operating cycle; calling a trained speed prediction model, the speed prediction model including an instantaneous fuel gated loop unit, a fuel trend gated loop unit, a fusion layer, and a fully connected layer; inputting the instantaneous fuel data and fuel trend data into the instantaneous fuel gated loop unit and the fuel trend gated loop unit of the speed prediction model, respectively, and outputting instantaneous fuel hidden features and fuel trend hidden features corresponding to the instantaneous fuel data and the fuel trend data, respectively; inputting the instantaneous fuel hidden features and the fuel trend hidden features into the fusion layer of the speed prediction model, and outputting fused features; inputting the fused features into the fully connected layer of the speed prediction model, and outputting the predicted speed of the gas turbine.

[0005] According to an embodiment of this application, the instantaneous fuel gating loop unit includes a first reset subunit, a first update subunit, a first candidate state subunit, and a first output subunit. The instantaneous fuel hidden features are obtained through the following operations: instantaneous fuel data and historical instantaneous fuel hidden features of the instantaneous fuel gating loop unit are input to the first reset subunit, and an instantaneous fuel reset vector is output, wherein the instantaneous fuel reset vector represents the input ratio of historical instantaneous fuel hidden features in the first candidate state subunit; instantaneous fuel data and the instantaneous fuel reset vector are input to the first candidate state subunit, and a candidate instantaneous fuel hidden state is output; instantaneous fuel data and historical instantaneous fuel hidden features are input to the first update subunit, and an instantaneous fuel update vector is output, wherein the instantaneous fuel update vector is used to represent the fusion ratio of historical instantaneous fuel hidden features and candidate instantaneous fuel hidden states in the first output subunit; and instantaneous fuel update vector, candidate instantaneous fuel hidden states, and historical instantaneous fuel hidden features are input to the first output subunit, and instantaneous fuel hidden features are output.

[0006] According to an embodiment of this application, the fuel trend gating loop unit includes a second reset subunit, a second update subunit, a second candidate state subunit, and a second output subunit. The fuel trend hidden features are obtained through the following operations: fuel trend data and historical fuel trend hidden features of the fuel trend gating loop unit are input to the second reset subunit, and a fuel trend reset vector is output, wherein the fuel trend reset vector represents the input ratio of historical fuel trend hidden features in the second candidate state subunit; fuel trend data and fuel trend reset vector are input to the second candidate state subunit, and a candidate fuel trend hidden state is output; fuel trend data and historical fuel trend hidden features are input to the second update subunit, and a fuel trend update vector is output, wherein the fuel trend update vector is used to represent the fusion ratio of historical fuel trend hidden features and candidate fuel trend hidden states in the second output subunit; the fuel trend update vector, candidate fuel trend hidden states, and historical fuel trend hidden features are input to the second output subunit, and the fuel trend hidden features are output.

[0007] According to an embodiment of this application, in the fuel instantaneous reset vector output by the first reset gate subunit, the input proportion of the historical fuel instantaneous hidden feature in the first candidate state subunit is less than the first input threshold; in the fuel trend reset vector output by the second reset gate subunit, the input proportion of the historical fuel trend hidden feature in the second candidate state subunit is greater than the second input threshold.

[0008] According to an embodiment of this application, the instantaneous fuel data and the historical instantaneous fuel hidden features of the instantaneous fuel gating loop unit are input to the first reset gate sub-unit, and the instantaneous fuel reset vector is output. This includes: processing the instantaneous fuel data and the historical instantaneous fuel hidden features according to the reset weight matrix and the reset bias vector, and outputting the instantaneous fuel reset vector, wherein the number of columns in the reset weight matrix is ​​less than the number of elements in the reset bias vector.

[0009] According to embodiments of this application, the operating conditions include at least one of starting condition, idling condition, rated condition, or variable condition; wherein, in the starting condition, the load of the gas turbine is less than a first proportion of the rated load and the fluctuation within a preset time period is greater than or equal to a second proportion; in the idling condition, the load of the gas turbine is less than a first proportion of the rated load and the fluctuation within a preset time period is less than a third proportion; in the rated condition, the load of the gas turbine is greater than or equal to the rated load and the fluctuation within a preset time period is less than a third proportion; in the variable condition, the load of the gas turbine is greater than or equal to a fourth proportion of the rated load and the fluctuation within a preset time period is greater than or equal to a third proportion.

[0010] According to an embodiment of this application, fuel quantity data is split into instantaneous fuel quantity data and fuel trend data, including: when the fuel quantity data corresponds to either idling or rated operating conditions, the average value of the original fuel quantity in the fuel quantity data is used as instantaneous fuel quantity data, and fuel trend data is obtained based on the fuel quantity data of a first number of time steps; when the fuel quantity data corresponds to either starting or variable operating conditions, the original fuel quantity in the fuel quantity data is used as instantaneous fuel quantity data, and fuel trend data is obtained based on the fuel quantity data of a second number of time steps, wherein the first number is greater than the second number.

[0011] According to an embodiment of this application, the fuel quantity data is normalized data, wherein the fuel quantity data corresponding to the rated operating condition and the idling operating condition is obtained by the quantile normalization method, and the fuel quantity data corresponding to the starting operating condition and the variable operating condition is obtained by the maximum and minimum value normalization method.

[0012] According to an embodiment of this application, the method further includes: concatenating the operating condition as an independent heat vector with the original fuel quantity data to obtain fuel quantity data.

[0013] According to an embodiment of this application, the instantaneous fuel hidden feature and the fuel trend hidden feature are input into the fusion layer of the speed prediction model, and the fused feature is output. The method includes: calculating the attention scores of the instantaneous fuel hidden feature and the fuel trend hidden feature respectively; converting the attention scores of the instantaneous fuel hidden feature and the fuel trend hidden feature into attention weights through a normalized exponential function; and fusing the instantaneous fuel hidden feature and the fuel trend hidden feature according to the attention weights of the instantaneous fuel hidden feature and the fuel trend hidden feature to obtain the fused feature.

[0014] Another aspect of this application provides a gas turbine speed prediction device, comprising: an acquisition module for acquiring fuel quantity data of a gas turbine corresponding to various operating conditions during a predetermined historical operating cycle; a splitting module for splitting the fuel quantity data into instantaneous fuel data and fuel trend data, wherein the instantaneous fuel data represents the instantaneous value of the gas turbine's fuel quantity at a predetermined monitoring time during the historical operating cycle, and the fuel trend data represents the changing trend of the gas turbine's fuel quantity during the historical operating cycle; and a calling module for calling a trained speed prediction model, the speed prediction model including a fuel instantaneous gated loop. The system comprises a unit, a fuel trend gating loop unit, a fusion layer, and a fully connected layer; a first output module, used to input instantaneous fuel data and fuel trend data into the instantaneous fuel gating loop unit and the fuel trend gating loop unit of the speed prediction model, respectively, and output the instantaneous fuel hidden features and the fuel trend hidden features corresponding to the instantaneous fuel data and the fuel trend data, respectively; a second output module, used to input the instantaneous fuel hidden features and the fuel trend hidden features into the fusion layer of the speed prediction model, and output the fused features; and a third output module, used to input the fused features into the fully connected layer of the speed prediction model, and output the predicted speed of the gas turbine.

[0015] Another aspect of this application provides an electronic device, including: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0016] Another aspect of this application provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0017] Another aspect of this application provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the steps of the above-described method.

[0018] According to the embodiments of this application, fuel quantity data corresponding to various operating conditions of a gas turbine in its historical operating cycle is obtained and split into instantaneous fuel data and fuel trend data, which characterize the instantaneous value and trend of fuel quantity. Then, a speed prediction model is called. The speed prediction model includes a fuel instantaneous gated loop unit, a fuel trend gated loop unit, a fusion layer, and a fully connected layer. The fuel instantaneous gated loop unit and the fuel trend gated loop unit process the fuel instantaneous data and fuel trend data in parallel. Fuel quantity data is obtained for operating conditions, thus distinguishing multiple operating conditions. Furthermore, dual-channel heterogeneous processing is adopted for different features, so accurate speed prediction can be achieved using only fuel quantity data. The input features are simple and adaptable to different operating condition categories. Attached Figure Description

[0019] The above and other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 This illustration schematically shows an exemplary system architecture of a gas turbine speed prediction method, apparatus, electronic device, storage medium, and program product according to embodiments of this application;

[0021] Figure 2 A flowchart illustrating a gas turbine speed prediction method according to an embodiment of this application is shown schematically.

[0022] Figure 3 A schematic diagram of the structure of a fuel instantaneous gating circulation unit according to an embodiment of this application is shown.

[0023] Figure 4 A schematic diagram of the structure of a gated loop unit according to an embodiment of this application is shown.

[0024] Figure 5 This illustration schematically shows a gated loop unit structure of a speed prediction model according to an embodiment of the present application;

[0025] Figure 6 This illustration shows a schematic diagram of the changes between the predicted speed and the actual speed of the speed prediction model according to an embodiment of this application;

[0026] Figure 7 This illustration schematically shows a diagram illustrating the error between the predicted speed and the actual speed of a speed prediction model according to an embodiment of this application.

[0027] Figure 8 A block diagram of a gas turbine speed prediction device according to an embodiment of this application is schematically shown; and

[0028] Figure 9A block diagram of an electronic device suitable for implementing the methods described above, according to an embodiment of this application, is illustrated schematically. Detailed Implementation

[0029] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0030] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0031] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0032] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0033] In the embodiments of this application, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard user personal information security and network security.

[0034] In the embodiments of this application, the user's authorization or consent was obtained before obtaining or collecting the user's personal information.

[0035] Digital modeling of gas turbines is a key technological support for promoting upgrades throughout their entire lifecycle, including design, operation, maintenance, and optimization. However, due to the highly nonlinear thermodynamic operating principle of gas turbines, the performance variation characteristics under different operating conditions vary significantly, and the thermodynamic and aerodynamic operating mechanism of gas turbines themselves is not fully understood. Therefore, it is very difficult to establish a high-response, accurate mathematical model of a gas turbine that covers all operating conditions. It is necessary to combine artificial intelligence methods to reduce the reliance on complex physical principles in the modeling process.

[0036] In related technologies, an adaptive performance simulation method based on particle swarm optimization (PSO) algorithm is used to address deviations in component characteristic maps caused by manufacturing tolerances, assembly, and maintenance operations in gas turbines, providing theoretical support for early warning and adaptive diagnosis. However, this method is still based on the thermodynamic component method for modeling and relies on complex and time-consuming preliminary component characteristic simulation calculations or experiments, and cannot directly utilize whole-machine test data. Related technologies also perform performance calculations by loading component models, which is also based on the thermodynamic component method and has limited accuracy for performance deviating from rated environmental conditions.

[0037] Traditional gas turbine modeling employs a thermodynamic component method based on physical models, relying on complex thermo-aerodynamic operating mechanism analysis. This results in extremely long computation times for gas turbine performance calculations, making it difficult to use for rapid development and application throughout the entire lifecycle. Furthermore, using traditional recurrent neural networks to train gas turbine models suffers from problems. Because time-series information is passed through chain multiplication, the gradient grows exponentially with the sequence length when long-term operating data is input. This not only leads to excessively long computation times but also makes the models susceptible to interference from noise during gas turbine operation, failing to adapt to the highly nonlinear operating characteristics of gas turbines. Related neural network modeling methods all use uniform operating conditions and single-channel processing, failing to design differentiated data processing flows based on the different operating characteristics of gas turbines, and neglecting to distinguish between heterogeneous gating logic for instantaneous values ​​and trend channels, thus failing to adapt to the highly nonlinear characteristics across all operating conditions.

[0038] In view of this, embodiments of this application provide a gas turbine speed prediction method. By acquiring fuel quantity data corresponding to various operating conditions during the historical operating cycle of the gas turbine, and splitting it into instantaneous fuel data and fuel trend data characterizing the instantaneous value and trend of fuel quantity, a speed prediction model is then invoked. This model includes an instantaneous fuel gating loop unit, a fuel trend gating loop unit, a fusion layer, and a fully connected layer. The instantaneous fuel gating loop unit and the fuel trend gating loop unit process the instantaneous fuel data and fuel trend data in parallel. Fuel quantity data is acquired based on the operating conditions, achieving differentiation among multiple operating conditions. Furthermore, dual-channel heterogeneous processing is adopted for different features, enabling accurate speed prediction using only fuel quantity data. The input features are simple and adaptable to different operating condition categories.

[0039] Figure 1 This illustration schematically depicts an exemplary system architecture for a gas turbine speed prediction method, apparatus, electronic device, storage medium, and program product according to embodiments of this application. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this application, in order to help those skilled in the art understand the technical content of this application, but do not mean that the embodiments of this application cannot be used in other devices, systems, environments or scenarios.

[0040] like Figure 1 As shown, the system architecture according to this embodiment may include a gas turbine 110 and a predictive processor 120.

[0041] The prediction processor 120 can acquire fuel quantity data of the gas turbine 110 and execute the gas turbine speed prediction method of this application embodiment to output the predicted speed of the gas turbine.

[0042] The predicted speed output by the predictive processor 120 can be compared with the actual speed of the gas turbine 110 to determine whether the gas turbine 110 is operating normally.

[0043] Figure 2 A flowchart illustrating a gas turbine speed prediction method according to an embodiment of this application is shown schematically.

[0044] like Figure 2 As shown, the method includes operations S210~S260.

[0045] In operation S210, fuel quantity data corresponding to various operating conditions of the gas turbine in a predetermined historical operating cycle are obtained.

[0046] The predetermined historical operating cycle can be the operating period before and including the current moment. The historical operating cycle can include multiple monitoring points. The operating conditions of the gas turbine at these multiple monitoring points can be determined based on the gas turbine's operating power, speed, exhaust temperature, etc. Fuel quantity data refers to the fuel consumption of the gas turbine at that monitoring point.

[0047] In operation S220, fuel quantity data is split into instantaneous fuel data and fuel trend data.

[0048] Instantaneous fuel data represents the instantaneous value of fuel quantity in the gas turbine at a predetermined monitoring time during the historical operating cycle, while fuel trend data represents the changing trend of fuel quantity in the gas turbine during the historical operating cycle.

[0049] Fuel quantity data can be split into instantaneous fuel data and fuel trend data according to the time sequence of historical operating cycles.

[0050] The scheduled monitoring time can be the last monitoring time in the historical operating cycle, that is, the scheduled monitoring time can be the current time. The instantaneous fuel data can be the fuel consumption of the gas turbine at the scheduled monitoring time, that is, the fuel quantity of the gas turbine at the current time.

[0051] Fuel trend data can be the changing trend of fuel quantity composition at each monitoring moment in the historical operating cycle.

[0052] When operating S230, the trained speed prediction model is invoked.

[0053] The speed prediction model includes a fuel instantaneous gated cycle unit, a fuel trend gated cycle unit, a fusion layer, and a fully connected layer.

[0054] A speed prediction model can be trained using historical operating data of gas turbines.

[0055] In operation S240, instantaneous fuel data and fuel trend data are input into the instantaneous fuel gating loop unit and the fuel trend gating loop unit of the speed prediction model, respectively, and the instantaneous fuel hidden features and the fuel trend hidden features corresponding to the instantaneous fuel data and the fuel trend data are output.

[0056] The instantaneous fuel gating loop unit and the fuel trend gating loop unit can be two parallel units with similar structures. The instantaneous fuel gating loop unit can process instantaneous fuel data and output instantaneous fuel hidden features. The fuel trend gating loop unit can process fuel trend data and output fuel trend hidden features.

[0057] By using dual-channel processing, each unit can extract different features.

[0058] In operation S250, the instantaneous fuel hidden features and the fuel trend hidden features are input into the fusion layer of the speed prediction model, and the fused features are output.

[0059] The fusion layer can fuse fuel instantaneous hidden features and fuel trend hidden features in a certain proportion and output fused features.

[0060] In operation S260, the fused features are input into the fully connected layer of the speed prediction model, and the predicted speed of the gas turbine is output.

[0061] The fully connected layer can map the fused features to obtain and output the predicted speed of the gas turbine.

[0062] According to the embodiments of this application, fuel quantity data corresponding to various operating conditions of a gas turbine in its historical operating cycle is obtained and split into instantaneous fuel data and fuel trend data, which characterize the instantaneous value and trend of fuel quantity. Then, a speed prediction model is called. The speed prediction model includes a fuel instantaneous gated loop unit, a fuel trend gated loop unit, a fusion layer, and a fully connected layer. The fuel instantaneous gated loop unit and the fuel trend gated loop unit process the fuel instantaneous data and fuel trend data in parallel. Fuel quantity data is obtained for operating conditions, thus distinguishing multiple operating conditions. Furthermore, dual-channel heterogeneous processing is adopted for different features, so accurate speed prediction can be achieved using only fuel quantity data. The input features are simple and adaptable to different operating condition categories.

[0063] According to embodiments of this application, the operating conditions include at least one of starting conditions, idling conditions, rated conditions, or variable conditions.

[0064] Specifically, under starting conditions, the gas turbine load is less than a first proportion of the rated load, and its fluctuation within a preset time period is greater than or equal to a second proportion. Under idling conditions, the gas turbine load is less than a first proportion of the rated load, and its fluctuation within a preset time period is less than a third proportion. Under rated conditions, the gas turbine load is greater than or equal to the rated load, and its fluctuation within a preset time period is less than a third proportion. Under variable operating conditions, the gas turbine load is greater than or equal to a fourth proportion of the rated load, and its fluctuation within a preset time period is greater than or equal to a third proportion.

[0065] The K-means clustering algorithm can be used to classify the time series data of gas turbines into starting conditions, idling conditions, rated conditions, or variable conditions according to their operating conditions.

[0066] The preset duration can be 3 seconds, the first percentage can be 1%, the second percentage can be 10%, the third percentage can be 5%, and the fourth percentage can be 5%.

[0067] For example, during startup, the gas turbine load is less than 1% of the rated load and the fluctuation within 3 seconds is greater than or equal to 10%. During idling, the gas turbine load is less than 1% of the rated load and the fluctuation within 3 seconds is less than 5%. During rated operation, the gas turbine load is greater than or equal to the rated load and the fluctuation within a preset time is less than 5%. The gas turbine load is greater than or equal to 5% of the rated load and the fluctuation within 3 seconds is greater than or equal to 5%.

[0068] By differentiating fuel quantity data according to operating conditions, targeted processing can be performed based on different operating conditions, thereby increasing the accuracy of gas turbine speed prediction.

[0069] According to an embodiment of this application, the fuel quantity data is normalized data, wherein the fuel quantity data corresponding to the rated operating condition and the idling operating condition is obtained by the quantile normalization method, and the fuel quantity data corresponding to the starting operating condition and the variable operating condition is obtained by the maximum and minimum value normalization method.

[0070] It can acquire the raw operating data of the gas turbine, and combined with the operating characteristics of the gas turbine, it can remove the fault signal data from the raw operating data, filter out the raw fuel quantity data, and normalize it according to the operating conditions to obtain the fuel quantity data.

[0071] Depending on the operating conditions, fuel quantity data for each operating condition can be normally processed independently to eliminate dimensional differences while retaining the operating characteristics of each condition, thus adapting the fuel quantity data to the input and output requirements of the speed prediction model. Differential normalization methods are employed based on the type of operating condition.

[0072] For stable operating conditions such as rated operating conditions and idling conditions, quantile normalization can be used to eliminate points with small-amplitude steady-state fluctuations and retain steady-state baseline characteristics. The lower quantile is set to 1%, and the upper quantile to 99%.

[0073] For dynamic operating conditions such as starting and changing operating conditions, maximum and minimum value normalization can be used to fully preserve the dynamic change range. The normalization process is shown in equation (1).

[0074] (1);

[0075] Where Y' represents the fuel quantity data, Y represents the original fuel quantity data, and Y max Y min These represent the maximum and minimum fuel quantities under this operating condition, respectively.

[0076] According to an embodiment of this application, the method may further include: concatenating the operating condition as an independent heat vector with the original fuel quantity data to obtain fuel quantity data.

[0077] A working condition feature embedding layer can be added to the speed prediction model, encoding the four types of operating conditions as unique heat vectors [0001, 0010, 0100, 1000]. The fuel quantity data can be the concatenation of operating condition and fuel quantity data, which can be represented as the input feature vector [X]. t Label i ](X t Represents the fuel quantity at monitoring time t in the historical operating cycle, Label i(This represents the label of the i-th operating condition). By inputting the fuel quantity data with the combined operating condition labels into the speed prediction model, the speed prediction model can quickly identify the operating characteristics of the gas turbine under different operating conditions.

[0078] According to embodiments of this application, fuel quantity data is split into instantaneous fuel data and fuel trend data, including:

[0079] When the fuel quantity data corresponds to either the idling condition or the rated condition, the average value of the original fuel quantity in the fuel quantity data is used as the instantaneous fuel data, and the fuel trend data is obtained based on the fuel quantity data of the first number of time steps.

[0080] When the fuel quantity data corresponds to either the starting condition or the variable condition, fuel trend data is obtained based on the original fuel quantity in the fuel quantity data as instantaneous fuel data and the fuel quantity data based on the time step of the second quantity, wherein the first quantity is greater than the second quantity.

[0081] The feature vector [X] output from the embedded layer of working condition features t Label i [Split into instantaneous fuel data X] t 1 And fuel trend data X t 2 The instantaneous fuel data can be the data from the last monitoring moment in the historical operating cycle, such as the current moment. The fuel trend data can be the fuel quantity data corresponding to a certain number of monitoring moments, such as 5 monitoring moments, where each monitoring moment can be one operating cycle of the gas turbine.

[0082] When the fuel quantity data corresponds to either the idling condition or the rated condition, fuel trend data is obtained by using the average of the original fuel quantity data as the instantaneous fuel data and the fuel quantity data based on a first number of time steps. Each time step can be a monitoring moment, and the first number can be, for example, 5. The fuel trend data can be the differential trend value of the fuel quantity data at 5 monitoring moments, for example, X. t -X t−1 This can weaken short-term fluctuations and strengthen steady-state trends.

[0083] When the fuel quantity data corresponds to either the starting condition or the variable condition, the fuel trend data is obtained by using the original fuel quantity in the fuel quantity data as the instantaneous fuel data and the fuel quantity data according to the second number of time steps. For example, the second number can be 3. Then the fuel trend data can be the differential trend value of the fuel quantity data at 3 monitoring times, which can enhance the capture of dynamic change characteristics.

[0084] Next, the instantaneous fuel data and fuel trend data are input into the instantaneous fuel gating loop unit and the fuel trend gating loop unit, respectively.

[0085] Figure 3 A schematic diagram of the structure of a fuel instantaneous gating circulation unit according to an embodiment of this application is shown.

[0086] like Figure 3 As shown, the fuel instantaneous gating cycle unit 300 includes a first reset gate subunit 310, a first update gate subunit 320, a first candidate state subunit 330, and a first output subunit 340.

[0087] The instantaneous fuel hidden features are obtained through the following operations: Instantaneous fuel data and historical instantaneous fuel hidden features from the instantaneous fuel gating loop unit are input to the first reset gate subunit, and an instantaneous fuel reset vector is output, where the instantaneous fuel reset vector represents the input ratio of historical instantaneous fuel hidden features in the first candidate state subunit; instantaneous fuel data and the instantaneous fuel reset vector are input to the first candidate state subunit, and a candidate instantaneous fuel hidden state is output; instantaneous fuel data and historical instantaneous fuel hidden features are input to the first update gate subunit, and an instantaneous fuel update vector is output, where the instantaneous fuel update vector represents the fusion ratio of historical instantaneous fuel hidden features and candidate instantaneous fuel hidden states in the first output subunit; the instantaneous fuel update vector, candidate instantaneous fuel hidden states, and historical instantaneous fuel hidden features are input to the first output subunit, and the instantaneous fuel hidden features are output.

[0088] The fuel trend gating loop unit includes a second reset subunit, a second update subunit, a second candidate state subunit, and a second output subunit. The structure of the fuel trend gating loop unit is similar to that of the fuel instantaneous gating loop unit, which can be referenced. Figure 3 .

[0089] The fuel trend hidden feature is obtained through the following operations: Fuel trend data and historical fuel trend hidden features from the fuel trend gating recurrent unit are input into the second reset sub-unit, and a fuel trend reset vector is output, where the fuel trend reset vector represents the input ratio of historical fuel trend hidden features in the second candidate state sub-unit; fuel trend data and the fuel trend reset vector are input into the second candidate state sub-unit, and a candidate fuel trend hidden state is output; fuel trend data and historical fuel trend hidden features are input into the second update sub-unit, and a fuel trend update vector is output, where the fuel trend update vector represents the fusion ratio of historical fuel trend hidden features and candidate fuel trend hidden states in the second output sub-unit; the fuel trend update vector, candidate fuel trend hidden states, and historical fuel trend hidden features are input into the second output sub-unit, and the fuel trend hidden feature is output.

[0090] The following details the processing procedures of the instantaneous fuel gating cycle unit and the fuel trend gating cycle unit.

[0091] Both the instantaneous fuel gated recurrent unit and the fuel trend gated recurrent unit can be implemented based on the gated recurrent unit (GRU). The GRU can solve the gradient vanishing and gradient explosion problems faced by traditional recurrent neural networks when processing long sequence data.

[0092] Figure 4 A schematic diagram of the structure of a gated loop unit according to an embodiment of this application is shown.

[0093] like Figure 4 As shown, the input X of the gated loop unit t The output R is processed through the reset gate. t After processing by the update gate, the output Z is... t The reset and update gates also receive the previous hidden state from the input. t-1 Finally, the candidate hidden state is processed through the candidate state output. The final output is the hidden feature H. t .

[0094] The calculation of the reset gate sub-unit of the two gated loop units is shown in Equation (2).

[0095] (2);

[0096] Where k=1, 2, correspond to the instantaneous fuel gating cycle unit and the fuel trend gating cycle unit, respectively, R t,k X is the reset vector output by the corresponding gated loop unit. t,k H is the fuel quantity data input to the corresponding gated cycle unit. t−1,k W represents the hidden historical state of the corresponding gated loop unit at the previous time step. xr,k W hr,k These are the reset weight matrices for the fuel quantity data of the corresponding reset gate sub-unit and the reset weight matrices for the historical hidden state at the previous time step, respectively. r,k σ is the reset bias vector corresponding to the reset gate sub-unit, and σ is the Sigmoid function.

[0097] According to an embodiment of this application, the instantaneous fuel data and the historical instantaneous fuel hidden features of the instantaneous fuel gating loop unit are input to the first reset gate sub-unit, and the instantaneous fuel reset vector is output. This includes: processing the instantaneous fuel data and the historical instantaneous fuel hidden features according to the reset weight matrix and the reset bias vector, and outputting the instantaneous fuel reset vector, wherein the number of columns in the reset weight matrix is ​​less than the number of elements in the reset bias vector.

[0098] The reset weight matrix can have a dimension of 1×16, resulting in a small receptive field and focusing on single-point precision. The reset bias vector can have a dimension of 1×32, resulting in a large receptive field and focusing on data patterns. The reset weight matrix and reset bias vector of the fuel trend gating loop unit can also be set in this way, which will not be elaborated further here.

[0099] According to an embodiment of this application, in the fuel instantaneous reset vector output by the first reset gate subunit, the input proportion of the historical fuel instantaneous hidden feature in the first candidate state subunit is less than the first input threshold; in the fuel trend reset vector output by the second reset gate subunit, the input proportion of the historical fuel trend hidden feature in the second candidate state subunit is greater than the second input threshold.

[0100] The two reset sub-units are not homogeneous and computed in parallel. Instead, they employ differentiated gating weight update rules designed for "instantaneous accuracy" and "temporal trend." In the first reset sub-unit of the fuel instantaneous gating loop unit, historical information is more easily discarded, and the focus is placed on the current input. That is, the proportion of historical fuel instantaneous hidden features in the input of the first candidate state sub-unit is less than the first input threshold.

[0101] Historical information is more easily preserved and change patterns are captured in the second reset sub-unit of the fuel trend gating recurrent unit. Finally, the two types of heterogeneous features are weighted and fused through an attention layer. That is, the proportion of historical fuel trend hidden features in the input of the second candidate state sub-unit is greater than the second input threshold. The first input threshold is less than the second input threshold.

[0102] Two gated loop units can perform the calculation of the update gate sub-unit in parallel, as shown in equation (3).

[0103] (3);

[0104] Among them, Z t,k This represents the update vector output by the update subunit of the k-th gated loop unit, used to control the fusion ratio of the historical fuel instantaneous hidden features and the candidate fuel instantaneous hidden states of the gas turbine.

[0105] Two gated loop units can perform candidate state calculations in parallel for candidate state subunits, combined with the reset vector R output by the reset gate subunit. t,k As shown in equation (4).

[0106] (4);

[0107] in, The candidate hidden state output by the candidate state sub-unit. W xz,k W is the input weight matrix (1×h) for the candidate state sub-units. hh,k b is the implicit weight matrix (with dimensions h×h) for the candidate state sub-units. h,k To update the gate bias (dimension 1×h). tanh represents the hyperbolic tangent function, mapping the output to the interval [-1, 1]. H t-1,k This represents the candidate hidden state from the previous time step (i.e., the output value from the previous time step). The operator ⊙ multiplies the array elements sequentially, enabling selective resetting of the gas turbine's historical fuel quantity data.

[0108] Combine the updated gate sub-unit output Z t,k The output sub-unit is processed as shown in equation (5).

[0109] (5);

[0110] Among them, H t,k To hide features.

[0111] According to an embodiment of this application, the instantaneous fuel hidden feature and the fuel trend hidden feature are input into the fusion layer of the speed prediction model, and the fused feature is output. The method includes: calculating the attention scores of the instantaneous fuel hidden feature and the fuel trend hidden feature respectively; converting the attention scores of the instantaneous fuel hidden feature and the fuel trend hidden feature into attention weights through a normalized exponential function; and fusing the instantaneous fuel hidden feature and the fuel trend hidden feature according to the attention weights of the instantaneous fuel hidden feature and the fuel trend hidden feature to obtain the fused feature.

[0112] The instantaneous fuel concealment feature and the fuel trend concealment feature are respectively H t,1 H t,2 .

[0113] The attention scores for the instantaneous fuel hidden feature and the fuel trend hidden feature are calculated separately and converted into attention weights α1 and α2 through a normalized exponential function, where α1 + α2 = 1. The weights are calculated as shown in equation (6).

[0114] (6);

[0115] Among them, W αHere, is the attention weight matrix (16×1), and exp is the exponential function. The calculation of these weights allows the fusion layer to automatically focus on features more effective for speed prediction during fusion.

[0116] The instantaneous fuel hidden features and the fuel trend hidden features are weighted and fused based on two attention weights to obtain the fused features. .

[0117] The fused features are mapped to a one-dimensional output through a fully connected layer, which is the predicted rotational speed of the gas turbine at the next moment.

[0118] The training and application processes of the speed prediction model can be similar. Before training, the weight matrix and bias are random initial values, and the input fuel quantity cannot obtain a speed value that matches the experimental data. Therefore, it is necessary to match the gas turbine fuel quantity-speed time series data and then carry out targeted training.

[0119] First, a speed prediction model can be built: based on the past fuel input W of the gas turbine. f The future speed of the gas turbine is predicted. The speed prediction model can be composed of five parts: an operating condition feature embedding layer, a fuel instantaneous gated cycle unit, a fuel trend gated cycle unit, a fusion layer, and a fully connected layer. The structure of each layer is adapted to the characteristics of the gas turbine's fuel quantity data and significantly reduces the number of parameters.

[0120] Among them, the operating condition feature embedding layer encodes the gas turbine operating condition into a unique heat vector, which is then concatenated with the fuel quantity data.

[0121] Before inputting the instantaneous fuel gating cycle unit and the fuel trend gating cycle unit, the fuel quantity data can be split to obtain instantaneous fuel data and fuel trend data.

[0122] The instantaneous fuel gating loop unit and the fuel trend gating loop unit can extract hidden features from instantaneous fuel data and fuel trend data in parallel.

[0123] The fusion layer fuses the instantaneous fuel hidden features and the fuel trend hidden features to obtain fused features, which are then mapped to the predicted rotational speed via a fully connected layer.

[0124] The input and output of the speed prediction model are as follows (7).

[0125] (7);

[0126] in, Let u be the predicted rotational speed of the gas turbine at time k+1. k The fuel quantity data for the gas turbine at time k and during a predetermined operating cycle is... Label iθ represents the working condition feature label, and θ represents the parameter set.

[0127] Figure 5 The diagram illustrates a gated loop unit structure of a speed prediction model according to an embodiment of this application.

[0128] like Figure 5 As shown, to balance the calculation pressure and accuracy of gas turbine speed prediction, a time series length of L=5 is selected. With sufficient computing resources, any value can be selected based on the characteristics of the gas turbine data.

[0129] The training process for the speed prediction model can be as follows.

[0130] The normalized fuel quantity data and corresponding speed time series data of the gas turbine were divided into training, validation, and test sets in a 7:2:1 ratio. A suitable optimization algorithm was selected as the optimization method during the training process, and training parameters suitable for the fuel quantity data and speed time series data were determined through testing. For example, the learning rate was initially set to 0.001, the batch size was initially set to 32 / 64, and the number of iterations was initially set to 1000, after which training was carried out.

[0131] The fuel quantity data and engine speed time series data from the training set are input into the engine speed prediction model. Through forward propagation, the predicted engine speed is obtained, and the mean square error loss value between the predicted engine speed and the actual engine speed is calculated. The predicted engine speed output by the engine speed prediction model can be restored to the actual engine speed value of the gas turbine through the inverse operation of formula (1).

[0132] The gradients of each weight matrix and bias are calculated using the backpropagation algorithm based on the loss value. The gradients are then updated and the parameter values ​​are adjusted using an optimization algorithm to suppress the gradient vanishing problem of traditional recurrent neural networks.

[0133] After each iteration, the validation set is input into the speed prediction model. If the loss of the validation set does not decrease after a certain number of iterations, such as 50, an early stopping mechanism is triggered to avoid overfitting of the speed prediction model.

[0134] After training is completed, the generalization ability of the speed prediction model is verified by using the test set. Finally, the parameters in formulas (1) to (6) are obtained through training, and the trained speed prediction model is obtained.

[0135] Figure 6 The diagram illustrates the changes in predicted and actual rotational speeds of the rotational speed prediction model according to an embodiment of this application. Figure 7 The diagram illustrates the error between the predicted speed and the actual speed of the speed prediction model according to an embodiment of this application.

[0136] like Figure 6 and Figure 7The predicted rotational speed achieves a good fit of over 90% with the actual rotational speed, and the prediction error is controlled within 3%. Furthermore, the gating mechanism of the gated recurrent neural network solves the gradient vanishing problem of traditional recurrent neural networks. Combined with the normalization of fuel quantity data from the gas turbine and customized training parameters, it effectively filters noise information during operation, accurately capturing the rotational speed variation pattern under the strong nonlinear characteristics of the gas turbine, exhibiting high prediction accuracy and strong anti-interference capability.

[0137] The gas turbine speed prediction method of this application combines fuel quantity data and speed time series data under all operating conditions of the gas turbine with a gated recurrent neural network. It does not rely on the complex thermo-gas dynamics principles of the gas turbine, but directly uses whole-machine test data for training. This avoids the time-consuming preliminary simulation calculations of the thermo-component method and can be quickly applied to the entire life cycle of gas turbine design, operation and maintenance, and optimization, thus improving the efficiency of model building.

[0138] Furthermore, the gas turbine speed prediction method of this application has strong model adaptability and generalization ability: training parameters such as time series length and learning rate can be flexibly adjusted according to the gas turbine operating conditions and computing resources, adapting to all operating conditions such as starting conditions, idling conditions, rated conditions or variable conditions, and can achieve accurate speed prediction using only fuel quantity data. The input features are simple and the method is highly practical for engineering applications.

[0139] Furthermore, the gas turbine speed prediction method of this application embodiment has good model stability: during the training process, a validation set and an early stop mechanism are set to effectively avoid model overfitting. The weight matrix and bias obtained by training are adapted to the characteristics of the gas turbine time series operation data, and the model runs stably and reliably.

[0140] Furthermore, the gas turbine speed prediction method in this application embodiment is customized according to the operating conditions and performs heterogeneous dual-channel processing: different data processing flows are designed for starting conditions, idling conditions, rated conditions or variable conditions, and heterogeneous gating calculation rules are used for instantaneous values ​​and time-series trends, which are different from the general gating loop network and homogeneous dual-channel method, forming a unique technical barrier.

[0141] Figure 8 A block diagram of a gas turbine speed prediction device according to an embodiment of this application is shown schematically.

[0142] like Figure 8 As shown, the gas turbine speed prediction device 800 includes an acquisition module 810, a splitting module 820, a calling module 830, a first output module 840, a second output module 850, and a third output module 860.

[0143] The acquisition module 810 is used to acquire fuel quantity data of the gas turbine corresponding to various operating conditions in a predetermined historical operating cycle.

[0144] The splitting module 820 is used to split the fuel quantity data into instantaneous fuel data and fuel trend data. The instantaneous fuel data represents the instantaneous value of the fuel quantity of the gas turbine at a predetermined monitoring time in the historical operating cycle, and the fuel trend data represents the changing trend of the fuel quantity of the gas turbine in the historical operating cycle.

[0145] Module 830 is invoked to invoke the trained speed prediction model, which includes a fuel instantaneous gated recurrent unit, a fuel trend gated recurrent unit, a fusion layer, and a fully connected layer.

[0146] The first output module 840 is used to input instantaneous fuel data and fuel trend data into the instantaneous fuel gating loop unit and the fuel trend gating loop unit of the speed prediction model, respectively, and output the instantaneous fuel hidden features and the fuel trend hidden features corresponding to the instantaneous fuel data and the fuel trend data, respectively.

[0147] The second output module 850 is used to input the instantaneous fuel hidden features and the fuel trend hidden features into the fusion layer of the speed prediction model and output the fused features.

[0148] The third output module 860 is used to input the fused features into the fully connected layer of the speed prediction model and output the predicted speed of the gas turbine.

[0149] Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or implemented by hardware or firmware in any other reasonable manner by integrating or packaging circuits, or implemented in any one of software, hardware, and firmware, or in a suitable combination of any of these. Alternatively, one or more of the modules, submodules, units, and subunits according to the embodiments of this application can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0150] For example, any and more of the following modules can be combined into one module / unit / subunit: the acquisition module 810, the splitting module 820, the calling module 830, the first output module 840, the second output module 850, and the third output module 860. Alternatively, any one of these modules / units / subunits can be split into multiple modules / units / subunits. Or, at least some of the functionality of one or more of these modules / units / subunits can be combined with at least some of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of this application, at least one of the acquisition module 810, splitting module 820, calling module 830, first output module 840, second output module 850, and third output module 860 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits (ASICs), or any other reasonable means of integrating or packaging circuits, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 810, splitting module 820, calling module 830, first output module 840, second output module 850, and third output module 860 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0151] Figure 9 A block diagram of an electronic device suitable for implementing the methods described above, according to an embodiment of this application, is illustrated schematically. Figure 9 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0152] like Figure 9 As shown, an electronic device 900 according to an embodiment of this application includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0153] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 902 and / or RAM 903. It should be noted that the programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0154] According to embodiments of this application, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the input / output (I / O) interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0155] According to embodiments of this application, the method flow according to embodiments of this application can be implemented as a computer software program. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 909, and / or installed from removable medium 911. When the computer program is executed by processor 901, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0156] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0157] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0158] For example, according to embodiments of this application, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0159] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this application. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the methods provided in the embodiments of this application.

[0160] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0161] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0162] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0163] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations are not explicitly described in this application. In particular, without departing from the spirit and teachings of this application, the features described in the various embodiments of this application can be combined and / or combined in various ways. All such combinations and / or combinations fall within the scope of this application.

[0164] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A method for predicting the speed of a gas turbine, characterized in that, The method includes: Acquire fuel quantity data for the gas turbine under various operating conditions during a predetermined historical operating cycle; The fuel quantity data is split into instantaneous fuel quantity data and fuel trend data. The instantaneous fuel quantity data represents the instantaneous value of the fuel quantity of the gas turbine at a predetermined monitoring time in the historical operating cycle, and the fuel trend data represents the changing trend of the fuel quantity of the gas turbine in the historical operating cycle. The trained engine speed prediction model is invoked. The engine speed prediction model includes a fuel instantaneous gated recurrent unit, a fuel trend gated recurrent unit, a fusion layer, and a fully connected layer. The instantaneous fuel data and the fuel trend data are respectively input into the instantaneous fuel gating loop unit and the fuel trend gating loop unit of the speed prediction model, and the instantaneous fuel hidden feature and the fuel trend hidden feature corresponding to the instantaneous fuel data and the fuel trend data are output respectively. The instantaneous fuel hidden features and the fuel trend hidden features are input into the fusion layer of the speed prediction model, and the fused features are output. The fused features are input into the fully connected layer of the speed prediction model to output the predicted speed of the gas turbine.

2. The method according to claim 1, characterized in that, The instantaneous fuel gating loop unit includes a first reset subunit, a first update subunit, a first candidate state subunit, and a first output subunit. The instantaneous fuel hidden feature is obtained through the following operation: The instantaneous fuel data and the historical instantaneous fuel hidden features of the instantaneous fuel gating loop unit are input to the first reset gate subunit, and the instantaneous fuel reset vector is output, wherein the instantaneous fuel reset vector represents the input ratio of the historical instantaneous fuel hidden features in the first candidate state subunit; The instantaneous fuel data and the instantaneous fuel reset vector are input into the first candidate state sub-unit, and the candidate fuel instantaneous hidden state is output. The instantaneous fuel data and the historical instantaneous fuel hiding features are input into the first update gate sub-unit, and the instantaneous fuel update vector is output. The instantaneous fuel update vector is used to characterize the fusion ratio of the historical instantaneous fuel hiding features and the candidate instantaneous fuel hiding state in the first output sub-unit. The instantaneous fuel update vector, the instantaneous hidden state of the candidate fuel, and the instantaneous hidden features of the historical fuel are input into the first output sub-unit, and the instantaneous hidden features of the fuel are output.

3. The method according to claim 2, characterized in that, The fuel trend gated loop unit includes a second reset sub-unit, a second update sub-unit, a second candidate state sub-unit, and a second output sub-unit. The fuel trend hidden feature is obtained through the following operation: The fuel trend data and the historical fuel trend hidden features of the fuel trend gated loop unit are input to the second reset gate subunit, and the fuel trend reset vector is output, wherein the fuel trend reset vector represents the input ratio of the historical fuel trend hidden features in the second candidate state subunit; The fuel trend data and the fuel trend reset vector are input into the second candidate state sub-unit, and the candidate fuel trend hidden state is output. The fuel trend data and the historical fuel trend hidden features are input into the second update gate sub-unit, and the fuel trend update vector is output. The fuel trend update vector is used to characterize the fusion ratio of the historical fuel trend hidden features and the candidate fuel trend hidden states in the second output sub-unit. The fuel trend update vector, the candidate fuel trend hiding state, and the historical fuel trend hiding features are input into the second output sub-unit, and the fuel trend hiding features are output.

4. The method according to claim 3, characterized in that, In the fuel instantaneous reset vector output by the first reset gate subunit, the input ratio of the historical fuel instantaneous hidden feature in the first candidate state subunit is less than the first input threshold. In the fuel trend reset vector output by the second reset gate subunit, the input proportion of the historical fuel trend hidden feature in the second candidate state subunit is greater than the second input threshold.

5. The method according to claim 2, characterized in that, The step of inputting the instantaneous fuel data and the historical instantaneous fuel hidden features of the instantaneous fuel gating loop unit into the first reset gate subunit and outputting the instantaneous fuel reset vector includes: The instantaneous fuel data and the historical instantaneous fuel hidden features are processed according to the reset weight matrix and the reset bias vector to output the instantaneous fuel reset vector, wherein the number of columns in the reset weight matrix is ​​less than the number of elements in the reset bias vector.

6. The method according to claim 1, characterized in that, The operating conditions include at least one of the following: starting condition, idling condition, rated condition, or variable condition; In the aforementioned starting condition, the load of the gas turbine is less than a first proportion of the rated load and the fluctuation within a preset time period is greater than or equal to a second proportion. Under the idling condition, the load on the gas turbine is less than the first proportion of the rated load and the fluctuation within the preset time period is less than the third proportion; Under the rated operating conditions, the load of the gas turbine is greater than or equal to the rated load and the fluctuation within the preset time period is less than the third ratio; Under the variable operating conditions, the load of the gas turbine is greater than or equal to the fourth proportion of the rated load, and the fluctuation within the preset time period is greater than or equal to the third proportion.

7. The method according to claim 6, characterized in that, The step of splitting the fuel quantity data into instantaneous fuel data and fuel trend data includes: When the fuel quantity data corresponds to either the idling condition or the rated condition, the fuel trend data is obtained by using the average value of the original fuel quantity in the fuel quantity data as the instantaneous fuel data and the fuel quantity data according to the first number of time steps. When the fuel quantity data corresponds to either the starting condition or the variable condition, the fuel trend data is obtained based on the original fuel quantity in the fuel quantity data as the instantaneous fuel data and the fuel quantity data according to the second number of time steps, wherein the first number is greater than the second number.

8. The method according to claim 6, characterized in that, The fuel quantity data is normalized data. Specifically, the fuel quantity data corresponding to the rated operating condition and the idling operating condition is obtained by the quantile normalization method, and the fuel quantity data corresponding to the starting operating condition and the variable operating condition is obtained by the maximum and minimum value normalization method.

9. The method according to any one of claims 1 to 8, characterized in that, The method further includes: The operating condition is used as an independent heat vector and concatenated with the original fuel quantity data to obtain the fuel quantity data.

10. The method according to any one of claims 1 to 8, characterized in that, The step of inputting the instantaneous fuel hidden features and the fuel trend hidden features into the fusion layer of the speed prediction model and outputting fused features includes: Calculate the attention scores for the instantaneous fuel hiding feature and the trend fuel hiding feature, respectively; The attention scores of the instantaneous fuel hidden feature and the fuel trend hidden feature are converted into attention weights using a normalized exponential function; The fuel instantaneous hidden feature and the fuel trend hidden feature are fused based on the attention weights of the fuel instantaneous hidden feature and the fuel trend hidden feature to obtain the fused feature.

Images