Optimization method and system of gas turbine unit APS starting control system

Abstract

The embodiment of the application provides a kind of gas unit APS starting control system optimization method and system, the system fault feature set under different system simulation test strategies is respectively encoded and output as the fault knowledge representation variable of same feature knowledge network, the strategy operation node of fault knowledge representation variable is set for each system simulation test strategy, according to strategy operation node, the fault knowledge representation variable of same feature knowledge network is fused, and fusion fault knowledge representation variable is generated after fault classification is carried out to fusion fault knowledge representation variable, so that the gas unit APS starting control system is optimized.Compared with single system simulation test strategy, by converting the test feedback data under different system simulation test strategies to the same feature knowledge network for vector representation, the association between test feedback data under different strategies can be combined, and the fault classification accuracy can be improved based on limited test feedback data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gas turbine units, in particular to an optimization method and system of an APS starting control system of a gas turbine unit. BACKGROUND

[0002] A gas turbine unit is composed of a compressor, a combustion chamber and a turbine. The working process of a gas-steam combined cycle unit is roughly as follows: air is sucked into the compressor, compressed and mixed with natural gas, and then burned in the combustion chamber to convert into heat energy. The high-temperature and high-pressure gas generated is introduced into the turbine of the gas turbine to expand step by step to do work, and the mechanical energy formed drives the generator to become electrical energy. In this process, auxiliary machine starting, gas turbine starting, steam turbine starting and unit load increasing need to be started by the APS starting control system in order to improve the operation reliability. Based on this, the operation stability of the APS starting control system of the gas turbine unit is related to the stability of the gas turbine unit. Therefore, it is necessary to timely eliminate possible faults of the APS starting control system of the gas turbine unit. In the related art, the APS starting control system of the gas turbine unit is usually simulated and tested based on a single system simulation test strategy to determine its potential faults, but the fault classification accuracy of this scheme is very limited. SUMMARY

[0003] Therefore, the purpose of the present application is to provide an optimization method and system of an APS starting control system of a gas turbine unit.

[0004] According to a first aspect of the present application, an optimization method of an APS starting control system of a gas turbine unit is provided, applied to a cloud server, and the method comprises:

[0005] obtaining a system fault feature set of the APS starting control system of the gas turbine unit under different system simulation test strategies;

[0006] based on a feature encoding unit of a trained fault classification network, encoding and outputting the system fault feature set under different system simulation test strategies into fault knowledge representation variables of the same feature knowledge network respectively;

[0007] setting a strategy running node of the fault knowledge representation variables for each system simulation test strategy, fusing the fault knowledge representation variables of the same feature knowledge network according to the strategy running node, and generating a fused fault knowledge representation variable;

[0008] based on a fully connected output unit of the fault classification network, classifying the fused fault knowledge representation variable, generating a hit coefficient of the APS starting control system of the gas turbine unit under different fault labels, and optimizing the APS starting control system of the gas turbine unit based on the hit coefficient of the APS starting control system of the gas turbine unit under different fault labels.

[0009] In a possible implementation of the first aspect, the step of obtaining the system fault feature set of the gas turbine unit APS start-up control system under different system simulation test strategies comprises:

[0010] The test feedback data of the gas turbine unit APS start-up control system collected in the system simulation test software is summarized to obtain the test feedback data under different system simulation test strategies.

[0011] The test feedback data under different system simulation test strategies is split, and the test feedback data under the different system simulation test strategies is processed into a system fault feature set under the corresponding system simulation test strategy.

[0012] In a possible implementation of the first aspect, the system fault feature set under different system simulation test strategies includes a system crash feature set, a system dead loop feature set, and a system delay operation feature set. The step of encoding the system fault feature set under different system simulation test strategies into a fault knowledge representation variable of the same feature knowledge network based on the feature encoding unit of the trained fault classification network comprises:

[0013] For the crash description log in the system crash feature set, the feature encoding unit based on the crash feature encoding network converts the crash description log into a first encoded knowledge representation vector through a feature encoding array, and superimposes the fault trigger node of the crash description log to generate a crash fault knowledge representation variable.

[0014] For the system dead loop description log in the system dead loop feature set, the feature encoding unit based on the crash feature encoding network converts the system dead loop description log into a second encoded knowledge representation vector through a feature mapping space, and superimposes the fault trigger node of the system dead loop description log to generate a system dead loop fault knowledge representation variable.

[0015] For the system delay operation description log in the system delay operation feature set, the feature encoding unit based on the crash feature encoding network converts the system delay operation description log into a third encoded knowledge representation vector through a feature mapping space, and superimposes the fault trigger node of the system dead loop description log to generate a system delay operation fault knowledge representation variable.

[0016] The crash fault knowledge representation variable, the system dead loop fault knowledge representation variable, and the system delay operation fault knowledge representation variable are mapped to the fault knowledge representation variable of the same feature knowledge network according to the spatial domain algorithm.

[0017] In a possible implementation of the first aspect, the step of mapping the collapse fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay operation fault knowledge representation variable to the fault knowledge representation variables of the same feature knowledge network according to the spatial domain algorithm comprises:

[0018] calculating a mapping vector of the fault knowledge domain based on the interaction between the collapse fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay operation fault knowledge representation variable;

[0019] mapping the collapse fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay operation fault knowledge representation variable to the fault knowledge representation variables of the same feature knowledge network through a mapping bitmap generated by the mapping vector of the fault knowledge domain.

[0020] In a possible implementation of the first aspect, before the step of encoding and outputting the system fault feature sets under different system simulation test strategies into the fault knowledge representation variables of the same feature knowledge network and setting the strategy operation nodes of the fault knowledge representation variables for each system simulation test strategy based on the feature encoding unit of the trained fault classification network, the method further comprises:

[0021] loading the system fault feature sets of the target gas turbine unit APS start control system under different system simulation test strategies into a neural network for network iterative learning and updating to generate a fault classification network; the neural network comprises a feature encoding unit for vectorizing representation of the system fault feature sets and a fully connected output unit for fault label learning of the fused fault knowledge representation variables, the fully connected output unit comprises a self-attention layer and a classification layer, and the step of loading the system fault feature sets of the target gas turbine unit APS start control system under different system simulation test strategies into the neural network for network iterative learning and updating to generate the fault classification network comprises:

[0022] processing the system fault feature sets of the target gas turbine unit APS start control system under different system simulation test strategies into the fused fault knowledge representation variables based on the feature encoding unit; and analyzing the fused fault knowledge representation variables based on the strategy operation nodes based on the self-attention layer of the fully connected output unit to obtain self-attention coefficients between the fused fault knowledge representation variables and update the fused fault knowledge representation variables according to the self-attention coefficients;

[0023] The updated fusion fault knowledge representation variable is subjected to fault label learning based on a classification layer of the full connection output unit, a hit coefficient value of the target gas turbine unit APS start control system on different fault labels is predicted, and a fault classification network is constructed.

[0024] In a possible implementation of the first aspect, the method further includes:

[0025] In the iterative learning update process of the neural network, after at least part of the knowledge representation vectors in the fusion fault knowledge representation variable are hidden based on the mask processing, it is predicted whether the at least part of the knowledge representation vectors match the fusion fault knowledge representation variable, and a plurality of learning tasks are generated.

[0026] In a possible implementation of the first aspect, the method further includes:

[0027] A learning cost calculation unit is set, and the network weight parameters in the fault classification network are updated by using a gradient descent method based on a learning cost value obtained by the learning cost calculation unit.

[0028] According to a second aspect of the present application, a cloud server is provided, which includes a machine readable storage medium and a processor, the machine readable storage medium stores machine executable instructions, and the processor implements the optimization method of the gas turbine unit APS start control system when executing the machine executable instructions.

[0029] According to a third aspect of the present application, a computer readable storage medium is provided, which stores computer executable instructions, and the computer executable instructions implement the optimization method of the gas turbine unit APS start control system when executed.

[0030] According to any one of the above aspects, in the present application, the system fault feature set of the gas turbine unit APS starting control system under different system simulation test strategies is acquired, the system fault feature set under different system simulation test strategies is respectively encoded and output as the fault knowledge representation variable of the same feature knowledge network based on the feature encoding unit of the trained fault classification network, the strategy running node of the fault knowledge representation variable is set for each system simulation test strategy, the fault knowledge representation variable of the same feature knowledge network is fused according to the strategy running node, the fused fault knowledge representation variable is generated, the fused fault knowledge representation variable is classified based on the fully connected output unit of the fault classification network, the hit coefficient of the gas turbine unit APS starting control system under different fault labels is generated, and the gas turbine unit APS starting control system is optimized based on the hit coefficient of the gas turbine unit APS starting control system under different fault labels. Compared with the single system simulation test strategy, the test feedback data under different system simulation test strategies is converted to the same feature knowledge network for vector representation, the association between the test feedback data under different strategies can be combined, the fault classification accuracy can be improved based on limited test feedback data. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0032] Figure 1 The flowchart of the optimization method of the gas turbine unit APS starting control system provided by the embodiments of the present application is shown.

[0033] Figure 2 The component structure schematic diagram of the cloud server for implementing the optimization method of the gas turbine unit APS starting control system provided by the embodiments of the present application is shown. EMBODIMENT

[0034] For the purposes of the present application embodiments, the technical solutions and advantages, the following will be based on the drawings in the present application embodiments, the technical solutions in the present application embodiments are clearly and completely described, it should be understood that the drawings in the present application only play the purpose of description and description, and do not limit the protection scope of the present application. In addition, it should be understood that the schematic drawing is not drawn according to the actual proportion. The flowchart used in the present application shows the operation realized by some embodiments based on the embodiments of the present application. It should be understood that the operations of the flowchart can not be implemented in sequence, and the steps without logical context relationship can be reversed in sequence or implemented simultaneously. In addition, under the guidance of the content of the present application, one or more other operations can be added to the flowchart, or one or more operations can be deleted from the flowchart.

[0035] In addition, the described embodiments are only part of the embodiments of the present application, not all the embodiments. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0036] Figure 1 The flowchart of the optimization method of the gas turbine unit APS starting control system provided by the embodiments of the present application is shown, it should be understood that in other embodiments, the order of part of the steps of the optimization method of the gas turbine unit APS starting control system of the present embodiment can be shared according to actual needs, or part of the steps can be omitted or maintained. The optimization method of the gas turbine unit APS starting control system in detail includes:

[0037] Step S101, obtaining a set of system fault characteristics of the gas turbine unit APS starting control system under different system simulation test strategies.

[0038] In the present embodiment, different system simulation test strategies can be set based on different test directions (such as running efficiency test direction, running stability test direction, running temperature control test direction, etc.), thereby corresponding system simulation test is carried out on the gas turbine unit APS starting control system, and then a set of system fault characteristics of the gas turbine unit APS starting control system under different system simulation test strategies is obtained.

[0039] For example, the test feedback data of the gas turbine unit APS starting control system collected in the system simulation test software can be tested, and the test feedback data of the gas turbine unit APS starting control system is summarized to obtain a system fault feature set under different system simulation test strategies. The output direction of the system fault feature set under different system simulation test strategies is different. The test feedback data of the gas turbine unit APS starting control system is obtained through the system simulation test software, and the test feedback data is processed into a system fault feature set under different system simulation test strategies. Further, based on the fusion fault knowledge representation variable of the system fault feature set under different system simulation test strategies, fault classification is performed. The test feedback data of each system simulation test strategy can be uniformly processed, the dependency relationship between each system simulation test strategy is considered, and the fault classification accuracy is improved.

[0040] In step S102, the feature encoding unit of the trained fault classification network encodes and outputs the system fault feature set under different system simulation test strategies into the fault knowledge representation variable of the same feature knowledge network.

[0041] The fault classification network is obtained by training the system fault feature set under different system simulation test strategies using the target gas turbine unit APS starting control system. The neural network includes a feature encoding unit for vectorizing the system fault feature set and a fully connected output unit for learning the fault label of the fault knowledge representation variable. The feature encoding unit can process the system fault feature set and convert it into a fault knowledge representation variable. In this process, each feature under different system simulation test strategies is not learned separately, but is updated uniformly in the network iteration learning and updating process of the entire neural network.

[0042] In the process of processing the system fault feature set under each system simulation test strategy, the system fault feature set processing process of different system simulation test strategies is not the same. The system fault feature set under different system simulation test strategies includes a system crash feature set, a system dead loop feature set, and a system delay running feature set. For the crash description log in the system crash feature set, the feature coding unit based on the crash feature coding network converts the crash description log into a first encoded knowledge representation vector through a feature coding array, and superimposes the fault trigger node of the crash description log to generate a crash fault knowledge representation variable. For the system dead loop description log in the system dead loop feature set, the feature coding unit based on the crash feature coding network converts the system dead loop description log into a second encoded knowledge representation vector through a feature mapping space, and superimposes the fault trigger node of the system dead loop description log to generate a system dead loop fault knowledge representation variable. For the system delay running description log in the system delay running feature set, the feature coding unit based on the crash feature coding network converts the system delay running description log into a third encoded knowledge representation vector through a feature mapping space, and superimposes the fault trigger node of the system dead loop description log to generate a system delay running fault knowledge representation variable. The crash fault knowledge representation variable, the system dead loop fault knowledge representation variable, and the system delay running fault knowledge representation variable are mapped to the same feature knowledge network fault knowledge representation variable according to a spatial domain algorithm.

[0043] In step S103, the strategy running node of the fault knowledge representation variable is set for each system simulation test strategy. The fault knowledge representation variable of the same feature knowledge network is fused according to the strategy running node to generate a fused fault knowledge representation variable.

[0044] Considering that different system simulation test strategies have different feature output attributes, and considering the association between the fault knowledge representation variables of different system simulation test strategies, the association information of the fault interaction between different system simulation test strategies can be obtained while setting the strategy running node of the fault knowledge representation variable for each system simulation test strategy, and the corresponding strategy running node is set for the association information of the fault interaction, so that the association between the test feedback data of different system simulation test strategies is combined in the fusion process of the fault knowledge representation variable, and the fused fault knowledge representation variable containing rich fault knowledge information is obtained.

[0045] Step S104, based on the full connection output unit of the fault classification network, classifying the fusion fault knowledge representation variable, generating the hit coefficient of the gas turbine unit APS starting control system under different fault labels, and optimizing the gas turbine unit APS starting control system based on the hit coefficient of the gas turbine unit APS starting control system under different fault labels.

[0046] For example, the fault label with a hit coefficient greater than a set coefficient can be determined as the target fault label of the gas turbine unit APS starting control system, and then a target cloud optimization program corresponding to the target fault label is called from the cloud optimization program library, and the target cloud optimization program is used to optimize the gas turbine unit APS starting control system.

[0047] Based on the above steps, a set of system fault features of the gas turbine unit APS starting control system under different system simulation test strategies is obtained, the set of system fault features under different system simulation test strategies is respectively encoded and output as fault knowledge representation variables of the same feature knowledge network based on the feature encoding unit of the trained fault classification network, the strategy running node of the fault knowledge representation variable is set for each system simulation test strategy, the fault knowledge representation variables of the same feature knowledge network are fused according to the strategy running node, the fusion fault knowledge representation variable is generated, the fusion fault knowledge representation variable is classified based on the full connection output unit of the fault classification network, and the hit coefficient of the gas turbine unit APS starting control system under different fault labels is generated. Compared with a single system simulation test strategy, by converting the test feedback data under different system simulation test strategies to a vector representation in the same feature knowledge network, the association between the test feedback data under different strategies can be combined, the fault classification accuracy can be improved based on limited test feedback data.

[0048] Next, further embodiments are provided, and the specific steps include:

[0049] Step S201, by summarizing the test feedback data of the gas turbine unit APS starting control system collected in the system simulation test software, test feedback data under different system simulation test strategies is obtained.

[0050] Step S202, the test feedback data under different system simulation test strategies is split, and the test feedback data under the different system simulation test strategies is processed into a set of system fault features under the corresponding system simulation test strategy.

[0051] Step S203, based on the set of system fault features of the target gas turbine unit APS starting control system under different system simulation test strategies loaded into the neural network for network iterative learning and updating, a fault classification network is generated.

[0052] The neural network includes a feature encoding unit for vectorizing representation of the system fault feature set and a fully connected output unit for learning fault labels of the fusion fault knowledge representation variable. The fully connected output unit includes a self-attention layer and a classification layer. The system fault feature set of the target gas turbine unit APS starting control system under different system simulation test strategies can be processed into the fusion fault knowledge representation variable based on the feature encoding unit. The self-attention layer of the fully connected output unit analyzes the fusion fault knowledge representation variable based on the strategy running node, obtains the self-attention coefficient between the fusion fault knowledge representation variables, and updates the fusion fault knowledge representation variable according to the self-attention coefficient. The classification layer of the fully connected output unit learns the fault labels of the updated fusion fault knowledge representation variable, predicts the hit coefficient value of the target gas turbine unit APS starting control system under different fault labels, and constructs a fault classification network.

[0053] In the neural network network iterative learning update process, after at least part of the knowledge representation vectors in the fusion fault knowledge representation variable are hidden based on the mask processing method, it is predicted whether the at least part of the knowledge representation vectors match the fusion fault knowledge representation variable, and a plurality of learning tasks are generated. Then, a learning cost calculation unit is set, and the network weight parameters in the fault classification network are updated using the gradient descent method based on the learning cost value calculated by the learning cost calculation unit.

[0054] Step S204, based on the feature encoding unit of the trained fault classification network, the system fault feature set under different system simulation test strategies is respectively encoded and output as the fault knowledge representation variable of the same feature knowledge network.

[0055] The system failure feature set under different system simulation test strategies here includes a system crash feature set, a system dead loop feature set and a system delay running feature set. Specifically, for the crash description log in the system crash feature set, the feature coding unit based on the crash feature coding network converts the crash description log into a first encoded knowledge representation vector through a feature coding array, and superimposes the fault trigger node of the crash description log to generate a crash fault knowledge representation variable; for the system dead loop description log in the system dead loop feature set, the feature coding unit based on the crash feature coding network converts the system dead loop description log into a second encoded knowledge representation vector through a feature mapping space, and superimposes the fault trigger node of the system dead loop description log to generate a system dead loop fault knowledge representation variable; for the system delay running description log in the system delay running feature set, the feature coding unit based on the crash feature coding network converts the system delay running description log into a third encoded knowledge representation vector through a feature mapping space, and superimposes the fault trigger node of the system dead loop description log to generate a system delay running fault knowledge representation variable; the crash fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay running fault knowledge representation variable are mapped to the fault knowledge representation variable of the same feature knowledge network according to the spatial domain algorithm.

[0056] By fusing the features under different system simulation test strategies, a joint representation space is obtained, and then the fault knowledge representation variables converted under different system simulation test strategies are mapped to the same feature knowledge network using the joint representation space. Specifically, based on the interaction between the crash fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay running fault knowledge representation variable, the mapping vector of the fault knowledge domain is calculated, and then the crash fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay running fault knowledge representation variable are mapped to the fault knowledge representation variable of the same feature knowledge network through the mapping bitmap generated by the mapping vector of the fault knowledge domain.

[0057] Step S205, for each system simulation test strategy, set the strategy running node of the fault knowledge representation variable, and fuse the fault knowledge representation variable of the same feature knowledge network according to the strategy running node to generate a fused fault knowledge representation variable.

[0058] Step S206, based on the full connection output unit of the fault classification network, classifying the fused fault knowledge representation variable to generate the hit coefficient of the gas turbine unit APS starting control system under different fault labels.

[0059] Based on the above steps, by encoding the system fault feature sets under different system simulation test strategies into fault knowledge representation variables of the same feature knowledge network respectively, setting the strategy running nodes of the fault knowledge representation variables for each system simulation test strategy, fusing the fault knowledge representation variables of the same feature knowledge network according to the strategy running nodes, generating the fused fault knowledge representation variables after the fusion, and classifying the fused fault knowledge representation variables, the gas turbine unit APS start control system is optimized. Compared with a single system simulation test strategy, by converting the test feedback data under different system simulation test strategies to the same feature knowledge network for vector representation, the association between the test feedback data under different strategies can be combined, the limited test feedback data can be fully based on, and the fault classification accuracy is improved.

[0060] Figure 2 A cloud server 100 that can be used to implement various embodiments described in the present application is schematically shown.

[0061] For one embodiment, Figure 2 The cloud server 100 is shown having one or more processors 102, a control module (chipset) 104 coupled to one or more of the processor(s) 102, a memory 106 coupled to the control module 104, a non-volatile memory (NVM) / storage device 108 coupled to the control module 104, one or more input / output devices 110 coupled to the control module 104, and a network interface 112 coupled to the control module 104.

[0062] The processor(s) 102 can include one or more single core or multi core processors, which can include any combination of general-purpose processors or dedicated processors (e.g., graphics processors, application processors, baseband processors, etc.). In some example design scenarios, the cloud server 100 can be capable of acting as a server device such as a gateway described in embodiments of the present application.

[0063] In some example design scenarios, the cloud server 100 can include one or more computer readable media (e.g., the memory 106 or the NVM / storage device 108) having instructions 114 and one or more processors 102 integrated with the one or more computer readable media configured to execute the instructions 114 to implement modules to perform the actions described in the present disclosure.

[0064] For one embodiment, the control module 104 can include any suitable interface controllers to provide any suitable interface to one or more of the processor(s) 102 and / or any suitable device or component in communication with the control module 104.

[0065] The control module 104 can include a memory controller module to provide an interface to the memory 106. The memory controller module can be a hardware module, a software module, and / or a firmware module.

[0066] The memory 106 can be used to load and store data and / or instructions 114, for example, for the cloud server 100. For one embodiment, the memory 106 can include any suitable volatile memory, for example, suitable DRAM. In some example design considerations, the memory 106 can include Double Data Rate Type Four Synchronous Dynamic Random Access Memory (DDR4 SDRAM).

[0067] For one embodiment, the control module 104 can include one or more input / output controllers to provide an interface to the NVM / storage device 108 and the input / output device(s) 110.

[0068] The NVM / storage device 108 can be used to store data and / or instructions 114, for example. The NVM / storage device 108 can include any suitable non-volatile memory (e.g., flash memory) and / or can include any suitable non-volatile storage device(s) (e.g., one or more hard disk drives (HDDs), one or more compact disk (CD) drives, and / or one or more digital versatile disk (DVD) drives).

[0069] The NVM / storage device 108 can include storage resources that are physically part of the device on which the cloud server 100 is installed, or it can be accessed by the device remotely and not necessarily part of the device. For example, the NVM / storage device 108 can be accessed via the input / output device(s) 110 over a network.

[0070] The input / output device(s) 110 can provide an interface for the cloud server 100 to communicate with any other suitable device(s), and the input / output device(s) 110 can include communication components, pinyin components, sensor components, etc. The network interface 112 can provide an interface for the cloud server 100 to communicate over one or more networks, and the cloud server 100 can wirelessly communicate with one or more components of a wireless network according to any of one or more wireless network standards and / or protocols, for example, to access a wireless network according to a communication standard, such as Wi-Fi®, WWAN, 2G, 3G, 4G, 5G, etc., or combinations thereof.

[0071] For one embodiment, one or more of the processor(s) 102 can be loaded with logic of one or more controllers of the control module 104 (e.g., a memory controller module). For one embodiment, one or more of the processor(s) 102 can be loaded with logic of one or more controllers of the control module 104 to form a system on a package (SwP). For one embodiment, one or more of the processor(s) 102 can be integrated on the same die as logic of one or more controllers of the control module 104. For one embodiment, one or more of the processor(s) 102 can be integrated on the same die as logic of one or more controllers of the control module 104 to form a system on a chip (SoC).

[0072] In various embodiments, cloud server 100 can be, but is not limited to, a cloud server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet, a netbook, etc.), and the like. In various embodiments, cloud server 100 can have more or less components, and / or different architectures. For example, in some example design considerations, cloud server 100 includes one or more cameras, a keyboard, a liquid crystal display (LCD) screen (including touch screen displays), non- volatile memory port, multiple antennas, a graphics chip, an application specific integrated circuit (ASIC), and a speaker.

[0073] The above describes the embodiments of the present application in detail, and the principles and implementation modes of the present application are described by applying specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range can be changed, and the above description of the present application should not be understood as a limitation.

Claims

1. An optimization method for a gas turbine unit APS start control system, characterized in that, The method applied to a cloud server comprises: obtaining a system fault feature set of a gas turbine unit APS starting control system under different system simulation test strategies; encoding and outputting the system fault feature set under different system simulation test strategies into fault knowledge representation variables of a same feature knowledge network based on a feature encoding unit of a trained fault classification network; setting a strategy running node of the fault knowledge representation variables for each system simulation test strategy, fusing the fault knowledge representation variables of the same feature knowledge network according to the strategy running node, and generating fused fault knowledge representation variables; classifying the fused fault knowledge representation variables based on a full connection output unit of the fault classification network, generating a hit coefficient of the gas turbine unit APS starting control system under different fault labels, and optimizing the gas turbine unit APS starting control system based on the hit coefficient of the gas turbine unit APS starting control system under different fault labels; the step of obtaining the system fault feature set of the gas turbine unit APS starting control system under different system simulation test strategies comprises: obtaining test feedback data under different system simulation test strategies by summarizing the test feedback data of the gas turbine unit APS starting control system collected in the system simulation test software; splitting the test feedback data under different system simulation test strategies, and processing the test feedback data under the different system simulation test strategies into system fault feature sets under corresponding system simulation test strategies; the system fault feature set under the different system simulation test strategies comprises a system crash feature set, a system dead loop feature set, and a system delay running feature set, and the step of encoding and outputting the system fault feature set under different system simulation test strategies into fault knowledge representation variables of a same feature knowledge network based on a feature encoding unit of a trained fault classification network comprises: for a crash description log in the system crash feature set, converting the crash description log into a first encoded knowledge representation vector through a feature encoding array based on a feature encoding unit of a crash feature encoding network, and generating a crash fault knowledge representation variable by superimposing a fault trigger node of the crash description log; for a system dead loop description log in the system dead loop feature set, converting the system dead loop description log into a second encoded knowledge representation vector through a feature mapping space based on the feature encoding unit of the crash feature encoding network, and generating a system dead loop fault knowledge representation variable by superimposing a fault trigger node of the system dead loop description log; for a system delay running description log in the system delay running feature set, converting the system delay running description log into a third encoded knowledge representation vector through a feature mapping space based on the feature encoding unit of the crash feature encoding network, and generating a system delay running fault knowledge representation variable by superimposing a fault trigger node of the system dead loop description log. The collapse fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay operation fault knowledge representation variable are mapped to the fault knowledge representation variables of the same feature knowledge network according to a spatial domain algorithm.

2. The method of claim 1, wherein, The step of mapping the collapse fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay operation fault knowledge representation variable to the fault knowledge representation variables of the same feature knowledge network according to a spatial domain algorithm comprises: Based on the interaction between the collapse fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay operation fault knowledge representation variable, a mapping vector of the fault knowledge domain is calculated; The collapse fault knowledge representation variable, the system dead loop fault knowledge representation variable and the system delay operation fault knowledge representation variable are mapped to the fault knowledge representation variables of the same feature knowledge network by a mapping bitmap generated by the mapping vector of the fault knowledge domain.

3. The method of claim 1, wherein, Before the feature encoding unit of the fault classification network based on the training is completed, the system fault feature set under different system simulation test strategies is respectively encoded and output as the fault knowledge representation variables of the same feature knowledge network, and before the strategy operation node of the fault knowledge representation variable is set for each system simulation test strategy, the method further comprises: Based on the system fault feature set of the target gas turbine unit APS start control system under different system simulation test strategies being loaded into the neural network for network iterative learning update, a fault classification network is generated; the neural network comprises a feature encoding unit for vectorization representation of the system fault feature set and a fully connected output unit for fault label learning of the fused fault knowledge representation variable, the fully connected output unit comprises a self-attention layer and a classification layer, and the step of loading the system fault feature set of the target gas turbine unit APS start control system under different system simulation test strategies into the neural network for network iterative learning update to generate the fault classification network comprises: Based on the feature encoding unit, the system fault feature set of the target gas turbine unit APS start control system under different system simulation test strategies is processed into the fused fault knowledge representation variable; based on the self-attention layer of the fully connected output unit, the fused fault knowledge representation variable is analyzed to obtain the self-attention coefficient between the fused fault knowledge representation variables, and the fused fault knowledge representation variable is updated according to the self-attention coefficient; Based on the classification layer of the fully connected output unit, the updated fused fault knowledge representation variable is subjected to fault label learning, the hit coefficient value of the target gas turbine unit APS start control system under different fault labels is predicted, and the fault classification network is constructed.

4. The method of claim 3, wherein, The method further comprises: In the iterative learning update process of the neural network, after at least part of the knowledge representation vectors in the fused fault knowledge representation variable are hidden based on a mask processing method, it is predicted whether at least part of the knowledge representation vectors match the fused fault knowledge representation variable, and a plurality of learning tasks are generated.

5. The method of optimization of a gas turbine package APS start control system according to claim 3 or 4, characterized in that, The method further comprises: A learning cost calculation unit is set, and network weight parameters in the fault classification network are updated by using a gradient descent method based on a learning cost value calculated by the learning cost calculation unit.

6. A computer-readable storage medium, characterized in that, A computer readable storage medium storing machine executable instructions, which when executed by a processor implement the optimization method of the gas turbine unit APS start control system according to any one of claims 1-5.

7. A cloud server, characterized by A computer readable storage medium storing machine executable instructions, which when executed by a processor implement the optimization method of the gas turbine unit APS start control system according to any one of claims 1-5.

Images