Artificial intelligence-based water turbine generator voiceprint monitoring method and system

Abstract

The application provides an artificial intelligence-based water-turbine-generator voiceprint monitoring method and system, and relates to the technical field of artificial intelligence.In the application, an initial voiceprint analysis network is extracted;at least two processing units of a voiceprint processing complex subnetwork are adjusted to form a voiceprint processing simplified subnetwork, the number of processing units of the voiceprint processing simplified subnetwork is less than that of the voiceprint processing complex subnetwork; the voiceprint processing simplified subnetwork is used to replace the voiceprint processing complex subnetwork in the initial voiceprint analysis network to form a target voiceprint analysis network; the target voiceprint analysis network is used to perform voiceprint information optimization operation on target voiceprint information to output optimized voiceprint information; and based on the optimized voiceprint information, operation abnormality analysis operation is performed on a target water-turbine-generator to output target operation abnormality analysis data.Based on the above, the efficiency of water-turbine-generator voiceprint monitoring can be improved to a certain extent.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based method and system for monitoring the acoustic signature of a hydro-generator. Background Technology

[0002] Artificial Intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or computer-controlled computing to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce new intelligent machines that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities. AI technology is a multidisciplinary field, encompassing both hardware and software technologies. Among its many applications, AI technology can be used to analyze voiceprint information.

[0003] In existing technologies, there is a problem of low efficiency in using artificial intelligence technology to analyze the acoustic fingerprint monitoring results (acoustic fingerprint information of hydro-generators), that is, to perform anomaly analysis on the acoustic fingerprint information of hydro-generators. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an artificial intelligence-based method and system for monitoring the acoustic signature of hydro-generators, so as to improve the efficiency of acoustic signature monitoring of hydro-generators to a certain extent.

[0005] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0006] An artificial intelligence-based method for monitoring the acoustic signature of a hydro-generator, the method comprising:

[0007] An initial voiceprint analysis network is extracted, which includes a complex voiceprint processing sub-network. The complex voiceprint processing sub-network is used to mine key data content of exemplary voiceprint information. The complex voiceprint processing sub-network includes at least two processing units.

[0008] At least two processing units of the complex subnetwork for voiceprint processing are adjusted to form a simplified subnetwork for voiceprint processing. The simplified subnetwork for voiceprint processing has a smaller number of processing units than the complex subnetwork for voiceprint processing. The simplified subnetwork for voiceprint processing is used to extract key data content of the target voiceprint information, which is obtained by voiceprint monitoring of the target hydro-generator.

[0009] The simplified subnetwork for voiceprint processing is used to replace the complex subnetwork for voiceprint processing in the initial voiceprint analysis network to form the target voiceprint analysis network.

[0010] Using the target voiceprint analysis network, the target voiceprint information is optimized to output the corresponding optimized voiceprint information.

[0011] Based on the optimized acoustic fingerprint information, an operational anomaly analysis operation is performed on the target hydro-generator to output target operational anomaly analysis data corresponding to the target hydro-generator. The target operational anomaly analysis data is used to reflect the operating status of the target hydro-generator.

[0012] In some preferred embodiments, in the above-described artificial intelligence-based hydro-generator acoustic signature monitoring method, the number of processing units included in the complex acoustic signature processing sub-network is equal to a first number. The step of adjusting at least two processing units of the complex acoustic signature processing sub-network to form a simplified acoustic signature processing sub-network includes:

[0013] Based on the configured processing unit selection method, a second number of sampling processing units are selected from the first number of processing units in the complex sub-network of voiceprint processing, wherein the second number is less than or equal to the first number.

[0014] The second number of sampling processing units are adjusted to form a simplified subnetwork for voiceprint processing.

[0015] In some preferred embodiments, in the above-described AI-based hydro-generator acoustic signature monitoring method, the second number of extraction processing units belong to the processing units of the first processing method. The processing units of the first processing method are used to perform key information mining operations on the loaded information to be processed in the complex acoustic signature processing sub-network. The processing units of the first processing method include at least one feature mining layer; and the step of adjusting the second number of extraction processing units to form a simplified acoustic signature processing sub-network includes:

[0016] The feature mining layer in the second number of extraction processing units is adjusted in a first manner to form a second number of processing units including the first feature mining structure; and the second number of processing units including the first feature mining structure are integrated to form a corresponding voiceprint processing simplified sub-network.

[0017] The step of adjusting the feature mining layer in the second number of extraction processing units in a first manner to form a second number of processing units including a first feature mining structure includes:

[0018] When the number of feature mining layers in the xth extraction processing unit is equal to one, the feature mining layer in the xth extraction processing unit is marked as a first feature mining structure, and the xth extraction processing unit belongs to one of the second number of extraction processing units; and when the number of feature mining layers in the xth extraction processing unit is greater than one, the feature mining layers in the xth extraction processing unit are integrated into mining layers, and the feature mining layer formed by the integration operation is marked as a first feature mining structure.

[0019] In some preferred embodiments, in the above-described AI-based hydro-generator acoustic signature monitoring method, the second number of extraction processing units includes processing units of a first processing method and processing units of a second processing method; the processing unit of the first processing method is used to perform feature mining operations on the loaded information to be processed in the complex acoustic signature processing sub-network, and the processing unit of the first processing method includes at least one feature mining layer; the processing unit of the second processing method is used to perform input-output linking operations on the loaded information to be processed in the complex acoustic signature processing sub-network, the number of processing units of the first processing method is equal to a third number, and the third number is less than the second number; and the step of adjusting the second number of extraction processing units to form a simplified acoustic signature processing sub-network includes:

[0020] The third number of processing units using the first processing method are adjusted using a first method to form a third number of processing units including a first feature mining structure; and the fourth number of processing units using the second processing method are adjusted using a second method to form a fourth number of processing units including a second feature mining structure, wherein the fourth number is equal to the difference between the second number and the third number; and the third number of processing units including the first feature mining structure and the fourth number of processing units including the second feature mining structure are integrated to form a corresponding simplified sub-network for voiceprint processing.

[0021] The processing unit of the first processing method further includes a parameter mapping processing structure, which is used to perform parameter mapping operations on the key data content mined by the feature mining layer in the processing unit of the first processing method.

[0022] In some preferred embodiments, in the above-described AI-based hydro-generator acoustic signature monitoring method, the first number of processing units includes a processing unit with a first processing mode, and the processing unit with the first processing mode includes at least one feature mining layer; and the step of selecting a second number of extraction processing units from the first number of processing units of the complex acoustic signature processing sub-network based on the configured processing unit selection method includes:

[0023] From the first number of processing units in the complex sub-network of voiceprint processing, a second number of extraction processing units are arbitrarily selected; or

[0024] Based on a predetermined target size, a second number of extraction processing units are selected from the processing units of the first processing method, wherein the size of the feature mining layer in the second number of extraction processing units is equal to the target size; or

[0025] Based on a predetermined target number of samples, a second number of sampling processing units are selected from the processing units of the first processing method. The number of feature mining layers in the second number of sampling processing units is equal to the target number of samples.

[0026] In some preferred embodiments, in the above-described AI-based hydro-generator acoustic signature monitoring method, the target acoustic signature analysis network includes a fifth number of simplified acoustic signature processing sub-networks; the target acoustic signature analysis network further includes a key information connection processing unit; the step of using the target acoustic signature analysis network to perform acoustic signature information optimization operations on the target acoustic signature information to output corresponding optimized acoustic signature information includes at least:

[0027] Using the key information connection processing unit, the pending key data content of the target voiceprint information and the key data content mined by the y-th voiceprint processing simplified sub-network are integrated to form the integrated key data content of the target voiceprint information. The y-th voiceprint processing simplified sub-network belongs to one of the fifth number of voiceprint processing simplified sub-networks.

[0028] Using the next simplified sub-network for voiceprint processing, the integrated key data content of the target voiceprint information is subjected to key information mining operation to output the key information mining output data of the target voiceprint information. The next simplified sub-network for voiceprint processing refers to a simplified sub-network for voiceprint processing that is cascaded after the y-th simplified sub-network for voiceprint processing.

[0029] Based on the key information of the target voiceprint information, the output data is mined and a feature restoration operation is performed to output the optimized voiceprint information corresponding to the target voiceprint information.

[0030] In some preferred embodiments, the above-described artificial intelligence-based hydro-generator acoustic signature monitoring method further includes:

[0031] Using an initial voiceprint analysis network that includes the voiceprint processing complex sub-network, the exemplary voiceprint information is optimized to output the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information.

[0032] Based on the difference information between the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information and the exemplary tag voiceprint information corresponding to the exemplary voiceprint information, the network parameters included in the voiceprint processing complex sub-network are optimized to form an optimized initial voiceprint analysis network.

[0033] In some preferred embodiments, in the above-described AI-based hydro-generator acoustic signature monitoring method, the number of processing units included in the acoustic signature processing complex sub-network is a first number, which is greater than one. The step of using the initial acoustic signature analysis network including the acoustic signature processing complex sub-network to perform acoustic signature information optimization operation on the exemplary acoustic signature information to output the exemplary optimized acoustic signature information corresponding to the exemplary acoustic signature information includes:

[0034] Using the first number of processing units, key information mining operations are performed on the exemplary voiceprint information to output the first number of key data partial contents corresponding to the exemplary voiceprint information. The key data partial contents and the key data contents are represented in the form of vectors.

[0035] The first number of key data partial contents are integrated to form the integrated key data content corresponding to the exemplary voiceprint information;

[0036] Based on the integrated key data content corresponding to the exemplary voiceprint information, a feature restoration operation is performed to output the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information.

[0037] In some preferred embodiments, in the above-described AI-based hydro-generator acoustic signature monitoring method, the step of integrating the first number of key data partial contents to form the integrated key data content corresponding to the exemplary acoustic signature information includes:

[0038] A concatenated combination operation is performed on the first number of key data local contents to form a concatenated combination feature distribution corresponding to the first number of key data local contents;

[0039] The first number of key data local contents are superimposed to form a superimposed feature distribution corresponding to the first number of key data local contents. The cascaded combination feature distribution and the superimposed feature distribution are represented in the form of vectors.

[0040] The feature distribution size is adjusted for the first number of key data local contents and the cascaded combined feature distribution to form multiple first feature distributions to be processed, and the feature distribution size is consistent among the multiple first feature distributions to be processed.

[0041] The feature distribution size is adjusted for the first number of key data local contents and the superimposed feature distribution to form multiple second feature distributions to be processed, and the feature distribution size is consistent among the multiple second feature distributions to be processed.

[0042] Using a first deep feature mining unit formed by network optimization, the focused feature parameter distribution of each of the multiple first feature distributions to be processed in relation to the cascaded combined feature distribution is analyzed to form a corresponding first focused feature analysis result, which is represented in vector form.

[0043] Using a second deep feature mining unit formed by network optimization, the focused feature parameter distribution of each of the multiple second feature distributions to be processed is analyzed in relation to the superimposed feature distribution, so as to form a corresponding second focused feature analysis result, which is represented in the form of a vector.

[0044] The analysis results of the first focused feature analysis and the second focused feature analysis are aggregated to form the integrated key data content corresponding to the exemplary voiceprint information. The aggregation operation of the analysis results includes the cascading combination or superposition of vectors.

[0045] This invention also provides an artificial intelligence-based hydro-generator acoustic signature monitoring system, including a processor and a memory. The memory stores a computer program, and the processor executes the computer program to implement the above-described hydro-generator acoustic signature monitoring method.

[0046] The artificial intelligence-based acoustic signature monitoring method and system for hydro-generators provided in this invention first extracts an initial acoustic signature analysis network; then, at least two processing units of the complex acoustic signature processing sub-network are adjusted to form a simplified acoustic signature processing sub-network; this simplified sub-network replaces the complex acoustic signature processing sub-network in the initial acoustic signature analysis network to form a target acoustic signature analysis network; the target acoustic signature analysis network is used to optimize the target acoustic signature information to output optimized acoustic signature information; and based on the optimized acoustic signature information, an operational anomaly analysis is performed on the target hydro-generator, outputting target operational anomaly analysis data. Based on the foregoing, since the number of processing units in the simplified acoustic signature processing sub-network is less than that in the complex acoustic signature processing sub-network, the processing efficiency of the target acoustic signature analysis network is improved, thus improving the efficiency of hydro-generator acoustic signature monitoring to a certain extent. Furthermore, since the acoustic signature optimization is performed using the target acoustic signature information, the operational anomaly analysis is based on the optimized acoustic signature information, rather than the collected target acoustic signature information; therefore, the basis is more reliable, and the results are also more reliable.

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0048] Figure 1 The structural block diagram of the artificial intelligence-based hydro generator acoustic signature monitoring system provided in the embodiments of the present invention is shown.

[0049] Figure 2 This is a flowchart illustrating the steps of the artificial intelligence-based acoustic signature monitoring method for hydro-generators provided in this embodiment of the invention.

[0050] Figure 3 This is a schematic diagram of the modules included in the artificial intelligence-based hydro generator acoustic signature monitoring device provided in this embodiment of the invention. Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0052] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0053] like Figure 1 As shown in the figure, this embodiment of the invention provides an artificial intelligence-based hydro-generator acoustic signature monitoring system. The artificial intelligence-based hydro-generator acoustic signature monitoring system may include a memory and a processor.

[0054] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, they can be electrically connected via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that exists in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby implementing the artificial intelligence-based hydro-generator acoustic signature monitoring method provided in this embodiment of the invention.

[0055] It should be understood that, in some feasible implementations, the memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc.

[0056] It should be understood that, in some feasible implementations, the processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0057] It should be understood that, in some feasible implementations, the AI-based hydro-generator acoustic signature monitoring system can be a server with data processing capabilities.

[0058] Combination Figure 2 This invention also provides an artificial intelligence-based method for monitoring the acoustic signature of a hydro-generator, which can be applied to the aforementioned artificial intelligence-based hydro-generator acoustic signature monitoring system. The method steps defined in the process of the artificial intelligence-based hydro-generator acoustic signature monitoring method can be implemented by the artificial intelligence-based hydro-generator acoustic signature monitoring system.

[0059] The following will be about Figure 2 The specific process shown will be explained in detail.

[0060] Step S110: Extract the initial voiceprint analysis network.

[0061] In this embodiment of the invention, the AI-based hydro-generator acoustic signature monitoring system can extract an initial acoustic signature analysis network. The initial acoustic signature analysis network includes a complex acoustic signature processing sub-network, which is used to mine key data content from exemplary acoustic signature information (i.e., perform key information or key data mining operations to obtain key data content). The complex acoustic signature processing sub-network includes at least two processing units. Furthermore, the initial acoustic signature analysis network can be a historically optimized acoustic signature analysis network, or it can be an initially built acoustic signature analysis network without network optimization, and it can be a neural network.

[0062] Step S120: Adjust at least two processing units of the complex sub-network for voiceprint processing to form a simplified sub-network for voiceprint processing.

[0063] In this embodiment of the invention, the AI-based hydro-generator acoustic signature monitoring system can adjust at least two processing units of the complex acoustic signature processing sub-network to form a simplified acoustic signature processing sub-network. The simplified acoustic signature processing sub-network includes fewer processing units than the complex acoustic signature processing sub-network, thus simplifying the processing units. The simplified acoustic signature processing sub-network is used to extract key data content of the target acoustic signature information, which is obtained by acoustic signature monitoring of the target hydro-generator (e.g., through a corresponding acoustic signature detection instrument).

[0064] Step S130: The simplified voiceprint processing subnetwork is used to replace the complex voiceprint processing subnetwork in the initial voiceprint analysis network to form the target voiceprint analysis network.

[0065] In this embodiment of the invention, the AI-based hydro-generator acoustic signature monitoring system can utilize the simplified acoustic signature processing sub-network to replace the complex acoustic signature processing sub-network in the initial acoustic signature analysis network to form a target acoustic signature analysis network. In other words, the complex acoustic signature processing sub-network in the initial acoustic signature analysis network can be deleted, and the simplified acoustic signature processing sub-network can be added to obtain a streamlined target acoustic signature analysis network.

[0066] Step S140: Using the target voiceprint analysis network, perform voiceprint information optimization on the target voiceprint information to output corresponding optimized voiceprint information.

[0067] In this embodiment of the invention, the AI-based hydro-generator acoustic signature monitoring system can utilize the target acoustic signature analysis network to optimize the target acoustic signature information, thereby outputting corresponding optimized acoustic signature information. The acoustic signature optimization function of the target acoustic signature analysis network can be obtained by learning from corresponding exemplary acoustic signature information and corresponding tags. Furthermore, the acoustic signature optimization function can refer to filtering out interference information in the target acoustic signature information, or supplementing missing information in the target acoustic signature information. This missing information may be caused by instrument malfunction during acquisition or formed during data transmission, or it may involve adding information to the target acoustic signature information.

[0068] Step S150: Based on the optimized voiceprint information, perform an operational anomaly analysis on the target hydro-generator to output the target operational anomaly analysis data corresponding to the target hydro-generator.

[0069] In this embodiment of the invention, the AI-based hydro-generator acoustic signature monitoring system can perform operational anomaly analysis on the target hydro-generator based on the optimized acoustic signature information, thereby outputting target operational anomaly analysis data corresponding to the target hydro-generator. The target operational anomaly analysis data reflects the operating status of the target hydro-generator, such as whether anomalies exist, the type of anomalies, and the degree of anomalies. Furthermore, the anomaly analysis operation can be implemented using an anomaly analysis neural network formed through network optimization. This anomaly analysis neural network learns from the corresponding acoustic signature information and corresponding tags, enabling it to perform operational anomaly analysis; the specific learning process is unrestricted.

[0070] Based on the foregoing, since the simplified subnetwork for acoustic fingerprint processing has fewer processing units than the complex subnetwork, the processing efficiency of the target acoustic fingerprint analysis network is improved. Therefore, this can, to some extent, improve the efficiency of acoustic fingerprint monitoring of hydro-generators, thereby addressing the inefficiency problem in existing technologies. Furthermore, because the acoustic fingerprint information optimization operation utilizes the target acoustic fingerprint information, the basis for anomaly analysis is the optimized acoustic fingerprint information, rather than the collected target acoustic fingerprint information. Therefore, the basis is more reliable, leading to more reliable results.

[0071] It should be understood that, in some feasible implementations, the number of processing units included in the complex voiceprint processing subnetwork is equal to the first number. Based on this, step S120 in the above description, namely the step of adjusting at least two processing units of the complex voiceprint processing subnetwork to form a simplified voiceprint processing subnetwork, may further include the specific implementation details described below:

[0072] Based on the configured processing unit selection method, a second number of sampling processing units are selected from the first number of processing units in the complex sub-network of voiceprint processing. The second number is less than or equal to the first number. For example, the first number can be less than the second number. That is, a second number of processing units are selected from the first number of processing units, and each of the second number of processing units is used as a sampling processing unit.

[0073] The second number of extraction processing units are adjusted, such as by integrating the extraction processing units, to form a simplified sub-network for voiceprint processing.

[0074] It should be understood that, in some feasible implementations, the first number of processing units may include processing units of a first processing method. The processing units of the first processing method include at least one feature mining layer. Feature mining can be the mining of the aforementioned key information or key data, for example, it may include a filter matrix or a convolution function. Based on this, the step of selecting a second number of extraction processing units from the first number of processing units in the complex sub-network of voiceprint processing based on the configured processing unit selection method may further include the specific implementation details described below:

[0075] From the first number of processing units in the complex sub-network of voiceprint processing, a second number of extraction processing units are arbitrarily selected; or

[0076] Based on a predetermined target size, a second number of extraction processing units are selected from the processing units of the first processing method. The size of the feature mining layer in the second number of extraction processing units is equal to the target size. For example, the size of the feature mining layer can refer to the size of the filter matrix. That is, the processing units of the first processing method can be extracted, meaning that the processing methods corresponding to the extraction processing units all belong to the first processing method, i.e., used for feature mining. For example, processing units including a 6*6 feature mining layer can be selected from the first number of processing units; or...

[0077] Based on a predetermined target number of selections, a second number of selection processing units are selected from the processing units of the first processing method. The number of feature mining layers in the second number of selection processing units is equal to the target number of selections. That is, in the processing units of the first processing method, each processing unit that includes a number of feature mining layers equal to the target number of selections is selected as a selection processing unit. For example, a processing unit that includes 3 feature mining layers can be selected from the first number of processing units.

[0078] It should be understood that, in some feasible implementations, the second number of extraction processing units belong to the processing units of the first processing method. The processing units of the first processing method are used to perform key information mining operations, i.e., feature mining operations, on the loaded information to be processed in the complex sub-network of voiceprint processing. The processing units of the first processing method include at least one feature mining layer. Based on this, the step of adjusting the second number of extraction processing units to form a simplified sub-network of voiceprint processing may further include the specific implementation content described below:

[0079] The feature mining layer in the second number of extraction processing units is adjusted in a first manner to form a second number of processing units including the first feature mining structure.

[0080] The second number of processing units including the first feature mining structure are integrated to form a corresponding voiceprint processing simplified sub-network. That is, the voiceprint processing simplified sub-network includes the second number of processing units including the first feature mining structure, or it may include a portion of the second number of processing units including the first feature mining structure.

[0081] It should be understood that, in some feasible implementations, the step of adjusting the feature mining layer in the second number of extraction processing units in a first manner to form a second number of processing units including the first feature mining structure may further include the specific implementation details described below:

[0082] When the number of feature mining layers in the xth extraction processing unit is equal to one, the feature mining layer in the xth extraction processing unit is marked as the first feature mining structure. The xth extraction processing unit belongs to one of the second number of extraction processing units, that is, feature mining layer processing is performed on each extraction processing unit.

[0083] If the number of feature mining layers in the xth extraction processing unit is greater than one, the feature mining layers in the xth extraction processing unit are integrated. For example, the size of each feature mining layer in the xth extraction processing unit can be determined first, and then the feature mining layer formed by the integration operation can be determined based on the size of each feature mining layer. For example, in one embodiment, a feature mining layer with a size of 1*1 and a feature mining layer with a size of 4*4 can be integrated into a feature mining layer with a size of 4*4; or, for another example, a feature mining layer with a size of 2*2 and a feature mining layer with a size of 7*7 can be integrated into a feature mining layer with a size of 8*8. The feature mining layer formed by the integration operation is marked as the first feature mining structure.

[0084] It should be understood that, in some feasible implementations, the second number of extraction processing units may include processing units of the first processing method and processing units of the second processing method. The processing unit of the first processing method is used to perform feature mining operations on the loaded information to be processed in the complex sub-network of voiceprint processing, and the processing unit of the first processing method includes at least one feature mining layer. The processing unit of the second processing method is used to perform input-output linking operations on the loaded information to be processed in the complex sub-network of voiceprint processing (as the network depth increases, the more nonlinear functions are passed through, and the data is mapped to a more discrete space, which may lead to gradient anomalies. Therefore, this problem can be improved by performing input-output linking operations, i.e., the key data content loaded into the processing unit of the first processing method and the key data content mined by the processing unit of the first processing method can be the same, i.e., performing residual connection operations, etc.). The number of processing units of the first processing method is equal to the third number, and the third number is less than the second number. Based on this, the step of adjusting the second number of extraction processing units to form a simplified sub-network of voiceprint processing may further include the specific implementation details described below:

[0085] The third number of processing units of the first processing method are adjusted in the first way to form a third number of processing units including the first feature mining structure. For example, for any processing unit, if the number of feature mining layers in the processing unit is equal to one, the feature mining layer in the processing unit is marked as the first feature mining structure. If the number of feature mining layers in the processing unit is greater than one, the feature mining layer in the processing unit is integrated, and the feature mining layer formed by the integration operation is marked as the first feature mining structure. The specific content can be described as described above, and will not be repeated here.

[0086] The fourth number of processing units using the second processing method are adjusted in a second manner to form a fourth number of processing units including a second feature mining structure, wherein the fourth number is equal to the difference between the second number and the third number; and the third number of processing units including a first feature mining structure and the fourth number of processing units including a second feature mining structure are integrated to form a corresponding voiceprint processing simplified sub-network. That is, the voiceprint processing simplified sub-network may include the third number of processing units including a first feature mining structure and the fourth number of processing units including a second feature mining structure, or it may include a portion of the third number of processing units including a first feature mining structure and / or a portion of the fourth number of processing units including a second feature mining structure.

[0087] It should be understood that, in some feasible implementations, the processing unit of the first processing method further includes a parameter mapping processing structure, which is used to perform parameter mapping operations on the key data content mined by the feature mining layer in the processing unit of the first processing method. For example, the parameters included in the key data content can be mapped to the interval (0, 1), etc.

[0088] It should be understood that, in some feasible implementations, the number of simplified voiceprint processing sub-networks included in the target voiceprint analysis network is equal to the fifth number. The target voiceprint analysis network may also include a key information connection processing unit. Based on this, step S140 in the above description, namely, the step of using the target voiceprint analysis network to perform voiceprint information optimization operations on the target voiceprint information to output corresponding optimized voiceprint information, may further include the following specific implementation details:

[0089] Using the key information connection processing unit, the undetermined key data content of the target voiceprint information and the key data content mined by the y-th voiceprint processing simplified sub-network are integrated to form the integrated key data content of the target voiceprint information. The y-th voiceprint processing simplified sub-network belongs to one of the fifth number of voiceprint processing simplified sub-networks (for example, it can be any voiceprint processing simplified sub-network). In addition, the undetermined key data content can be obtained by performing feature mining operations on the target voiceprint information through other feature mining layers, and by performing excitation processing on the results of the feature mining operations. The excitation processing can be implemented by an excitation function configured after the other feature mining layers. For example, for the first voiceprint processing simplified sub-network, the key data content of the target voiceprint information is mined in the voiceprint processing simplified sub-network. The key information connection processing unit can be used to integrate the key data content and the undetermined key data content of the target voiceprint information. For example, the key data content and the undetermined key data content can be superimposed, cascaded, or combined.

[0090] Using the next simplified sub-network for voiceprint processing, the integrated key data content of the target voiceprint information is subjected to key information mining operation to output the key information mining output data of the target voiceprint information. The next simplified sub-network for voiceprint processing refers to a simplified sub-network for voiceprint processing that is cascaded after the y-th simplified sub-network for voiceprint processing. For example, for the second simplified sub-network for voiceprint processing, the integrated key data content output by the key information connection processing unit corresponding to the first simplified sub-network for voiceprint processing can be subjected to key information mining operation to output the key information mining output data of the target voiceprint information. This key information mining output data serves as the key data content mined by the second simplified sub-network for voiceprint processing.

[0091] Based on the key information mining output data of the target voiceprint information, a feature restoration operation is performed to output the optimized voiceprint information corresponding to the target voiceprint information. For example, the key information mining output data of the target voiceprint information output by the last voiceprint processing simplification sub-network can be subjected to a feature restoration operation, such as through a decoding network, to output the optimized voiceprint information corresponding to the target voiceprint information. Alternatively, the key information mining output data of the target voiceprint information output by each voiceprint processing simplification sub-network can be superimposed or cascaded and then the feature restoration operation can be performed to output the optimized voiceprint information corresponding to the target voiceprint information.

[0092] It should be understood that, in some feasible implementations, the artificial intelligence-based hydro-generator acoustic signature monitoring method may further include the specific implementation details described below:

[0093] Using an initial voiceprint analysis network that includes the voiceprint processing complex sub-network, the exemplary voiceprint information is optimized to output the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information.

[0094] Based on the differences between the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information and the exemplary tag voiceprint information corresponding to the exemplary voiceprint information, the network parameters included in the complex sub-network of voiceprint processing are optimized to form an optimized initial voiceprint analysis network. For example, the exemplary voiceprint information is voiceprint information with missing information, the exemplary optimized voiceprint information can be voiceprint information after optimization and supplementation of missing information, and the exemplary tag voiceprint information can be voiceprint information without missing information. For example, partial information deletion processing can be performed on the exemplary tag voiceprint information to form the exemplary voiceprint information. Based on this, the corresponding network optimization cost index can be calculated according to the differences between the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information and the exemplary tag voiceprint information corresponding to the exemplary voiceprint information. Then, the network parameters included in the complex sub-network of voiceprint processing (the network parameters may include the size of the filter matrix, etc.) can be optimized along the direction of reducing the network optimization cost index to form an optimized initial voiceprint analysis network.

[0095] It should be understood that, in some feasible implementations, the number of processing units included in the complex sub-network for voiceprint processing is a first number, which may be greater than one. Based on this, the step of using the initial voiceprint analysis network including the complex sub-network for voiceprint processing to perform voiceprint information optimization on the exemplary voiceprint information to output the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information may further include the specific implementation content described below:

[0096] Using the first number of processing units, key information mining operations are performed on the exemplary voiceprint information to output the first number of key data partial contents corresponding to the exemplary voiceprint information. The key data partial contents and the key data contents are represented in the form of vectors.

[0097] The first number of key data partial contents are integrated to form the integrated key data content corresponding to the exemplary voiceprint information. That is, the first number of key data partial contents are fused or aggregated to obtain information-rich integrated key data content.

[0098] Based on the integrated key data content corresponding to the exemplary voiceprint information, a feature restoration operation is performed to output the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information.

[0099] It should be understood that, in some feasible implementations, the step of integrating the first number of key data partial contents to form the integrated key data content corresponding to the exemplary voiceprint information may further include the specific implementation details described below:

[0100] A cascaded combination operation is performed on the first number of key data local contents to form a cascaded combination feature distribution corresponding to the first number of key data local contents. Thus, the cascaded combination feature distribution can be {key data local content 1, key data local content 2, key data local content 3, key data local content 4, key data local content 5...}.

[0101] The first number of key data local contents are superimposed (for example, weighted calculation can be performed directly, or the mean can be superimposed, etc.) to form a superimposed feature distribution corresponding to the first number of key data local contents. The cascaded combination feature distribution and the superimposed feature distribution are represented in the form of vectors.

[0102] The feature distribution size is adjusted for the first number of key data local contents and the cascaded combined feature distribution to form multiple first feature distributions to be processed. The feature distribution size is consistent among the multiple first feature distributions to be processed. The specific method of adjusting the feature distribution size is not limited, such as stretching or compressing.

[0103] The feature distribution size is adjusted for the first number of key data local contents and the superimposed feature distribution to form multiple second feature distributions to be processed. The feature distribution size is consistent among the multiple second feature distributions to be processed. The specific method of adjusting the feature distribution size is not limited, such as stretching or compressing.

[0104] Using a first deep feature mining unit formed by network optimization, the focused feature parameter distribution of each of the multiple first feature distributions to be processed in relation to the cascaded combined feature distribution is analyzed to form a corresponding first focused feature analysis result, which is represented in vector form.

[0105] Using a second deep feature mining unit formed by network optimization, the focused feature parameter distribution of each of the multiple second feature distributions to be processed is analyzed in relation to the superimposed feature distribution, so as to form a corresponding second focused feature analysis result, which is represented in the form of a vector.

[0106] The analysis results of the first focused feature analysis and the second focused feature analysis are aggregated to form the integrated key data content corresponding to the exemplary voiceprint information. The aggregation operation of the analysis results includes the cascading combination or superposition of vectors.

[0107] It should be understood that, in some feasible implementations, the step of using a first deep feature mining unit formed through network optimization to analyze the focused feature parameter distribution of each of the plurality of first feature distributions to be processed relative to the cascaded combined feature distribution, in order to form a corresponding first focused feature analysis result, may further include the specific implementation details described below:

[0108] For each of the plurality of first feature distributions to be processed, the first mapping matrix, the second mapping matrix and the third mapping matrix included in the first deep feature mining unit formed by network optimization are used to perform mapping operations (i.e. multiplication) on the first feature distribution to be processed, so as to output the first mapped feature distribution, the second mapped feature distribution and the third mapped feature distribution corresponding to the first feature distribution to be processed.

[0109] For each of the plurality of first feature distributions to be processed, the first mapping feature distribution corresponding to the first feature distribution to be processed and the second mapping feature distribution corresponding to the first feature distribution to be processed corresponding to the cascaded combination feature distribution are multiplied together to output the focused feature parameter distribution corresponding to the first feature distribution to be processed. Then, the focused feature parameter distribution and the third mapping feature distribution corresponding to the first feature distribution to be processed corresponding to the cascaded combination feature distribution are multiplied together to output the multiplication result corresponding to the first feature distribution to be processed.

[0110] The multiplication results corresponding to each of the plurality of first feature distributions to be processed are weighted and summed to output the corresponding first focused feature analysis result. For example, in the process of weighted summation, the method for determining the weighting parameter corresponding to each first feature distribution to be processed may include: for each first feature distribution to be processed, calculating the average similarity between the first feature distribution to be processed and other first feature distributions to be processed; and determining a weighting coefficient based on the average similarity, wherein the weighting coefficient may have a positive correlation with the average similarity. The similarity mean can refer to the average of all similarities, and the similarity can refer to the cosine similarity between vectors; or, the multiple first feature distributions to be processed can be clustered to determine a cluster center or multiple clusters. In this way, for each first feature distribution to be processed, the minimum distance between the first feature distribution to be processed and each cluster center can be calculated. Then, the weighting coefficient corresponding to the first feature distribution to be processed can be determined based on the minimum distance. The weighting coefficient can be positively correlated with the minimum distance. In other embodiments, the weighting coefficient can also be determined based on other methods.

[0111] It should be understood that, in some feasible implementations, the step of using a second deep feature mining unit formed through network optimization to analyze the focused feature weight parameters of each of the plurality of second feature distributions to be processed relative to the superimposed feature distribution, so as to form a corresponding second focused feature analysis result, wherein the second focused feature analysis result is represented in vector form, may further include the specific implementation content described below:

[0112] For each of the plurality of second feature distributions to be processed, the fourth mapping matrix, the fifth mapping matrix and the sixth mapping matrix included in the first deep feature mining unit formed by network optimization are used to perform mapping operations (i.e. multiplication) on the second feature distribution to be processed, so as to output the fourth mapping feature distribution, the fifth mapping feature distribution and the sixth mapping feature distribution corresponding to the second feature distribution to be processed.

[0113] For each of the plurality of second feature distributions to be processed, the fourth mapping feature distribution corresponding to the second feature distribution to be processed and the fifth mapping feature distribution corresponding to the second feature distribution to be processed corresponding to the superimposed feature distribution are multiplied together to output the focused feature parameter distribution corresponding to the second feature distribution to be processed. Then, the focused feature parameter distribution and the sixth mapping feature distribution corresponding to the second feature distribution to be processed corresponding to the superimposed feature distribution are multiplied together to output the multiplication result corresponding to the second feature distribution to be processed.

[0114] The multiplication results corresponding to each of the plurality of second feature distributions to be processed are weighted and summed to output the corresponding second focused feature analysis result. For example, in the process of weighted summation, the method for determining the weighting parameter corresponding to each second feature distribution to be processed may include: for each second feature distribution to be processed, calculating the mean similarity between that second feature distribution and other second feature distributions to be processed; and determining a weighting coefficient based on the mean similarity, wherein the weighting coefficient may have a positive correlation with the mean similarity. The similarity mean can refer to the average of all similarities, and the similarity can refer to the cosine similarity between vectors; or, the multiple second feature distributions to be processed can be clustered to determine a cluster center or multiple clusters. In this way, for each second feature distribution to be processed, the minimum distance between the second feature distribution to be processed and each cluster center can be calculated. Then, the weighting coefficient corresponding to the second feature distribution to be processed can be determined based on the minimum distance. The weighting coefficient can be positively correlated with the minimum distance. In other embodiments, the weighting coefficient can also be determined based on other methods.

[0115] It should be understood that, in some feasible implementations, the step of aggregating the analysis results of the first focused feature analysis and the second focused feature analysis to form the integrated key data content corresponding to the exemplary voiceprint information may further include the specific implementation details described below:

[0116] Perform mutual focusing feature analysis operations on the first focusing feature analysis result and the second focusing feature analysis result to form corresponding first mutual focusing result and second mutual focusing result. For example, perform focusing feature analysis operations on the second focusing feature analysis result based on the first focusing feature analysis result to obtain the first mutual focusing result, and perform focusing feature analysis operations on the first focusing feature analysis result based on the second focusing feature analysis result to obtain the second mutual focusing result.

[0117] The first and second mutual focusing results are averaged to output the integrated key data content corresponding to the exemplary voiceprint information.

[0118] Combination Figure 3 This invention also provides an artificial intelligence-based hydro-generator acoustic signature monitoring device, which can be applied to the aforementioned artificial intelligence-based hydro-generator acoustic signature monitoring system. The artificial intelligence-based hydro-generator acoustic signature monitoring device may include:

[0119] An initial network extraction module is used to extract an initial voiceprint analysis network, the initial voiceprint analysis network including a voiceprint processing complex sub-network, the voiceprint processing complex sub-network being used to mine key data content of exemplary voiceprint information, the voiceprint processing complex sub-network including at least two processing units;

[0120] An initial network simplification module is used to adjust at least two processing units of the complex sub-network for voiceprint processing to form a simplified sub-network for voiceprint processing. The simplified sub-network for voiceprint processing has a smaller number of processing units than the complex sub-network for voiceprint processing. The simplified sub-network for voiceprint processing is used to extract key data content of the target voiceprint information, which is obtained by voiceprint monitoring of the target hydro-generator.

[0121] The target network generation module is used to replace the complex voiceprint processing subnetwork in the initial voiceprint analysis network with the simplified voiceprint processing subnetwork to form the target voiceprint analysis network.

[0122] The voiceprint information optimization module is used to optimize the target voiceprint information using the target voiceprint analysis network, so as to output the corresponding optimized voiceprint information.

[0123] An operation anomaly analysis module is used to perform operation anomaly analysis on the target hydro-generator based on the optimized acoustic fingerprint information, so as to output target operation anomaly analysis data corresponding to the target hydro-generator, the target operation anomaly analysis data being used to reflect the operating status of the target hydro-generator.

[0124] In summary, the artificial intelligence-based acoustic signature monitoring method and system for hydro-generators provided by this invention can first extract an initial acoustic signature analysis network; then, adjust at least two processing units of the complex acoustic signature processing sub-network to form a simplified acoustic signature processing sub-network; use the simplified acoustic signature processing sub-network to replace the complex acoustic signature processing sub-network in the initial acoustic signature analysis network to form a target acoustic signature analysis network; use the target acoustic signature analysis network to perform acoustic signature information optimization operations on the target acoustic signature information to output optimized acoustic signature information; and based on the optimized acoustic signature information, perform operational anomaly analysis operations on the target hydro-generator to output target operational anomaly analysis data. Based on the foregoing, since the number of processing units in the simplified acoustic signature processing sub-network is less than that in the complex acoustic signature processing sub-network, the processing efficiency of the target acoustic signature analysis network is improved, thus improving the efficiency of hydro-generator acoustic signature monitoring to a certain extent. Furthermore, since the acoustic signature information optimization operation is performed using the target acoustic signature information, the operational anomaly analysis operation is based on the optimized acoustic signature information, rather than the collected target acoustic signature information; therefore, the basis is more reliable, and the results are also more reliable.

[0125] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring the acoustic signature of a hydro-generator based on artificial intelligence, characterized in that, The acoustic signature monitoring method for the hydro-generator includes: An initial voiceprint analysis network is extracted, which includes a complex voiceprint processing sub-network. The complex voiceprint processing sub-network is used to mine key data content of exemplary voiceprint information. The complex voiceprint processing sub-network includes at least two processing units. At least two processing units of the complex subnetwork for voiceprint processing are adjusted to form a simplified subnetwork for voiceprint processing. The simplified subnetwork for voiceprint processing has a smaller number of processing units than the complex subnetwork for voiceprint processing. The simplified subnetwork for voiceprint processing is used to extract key data content of the target voiceprint information, which is obtained by voiceprint monitoring of the target hydro-generator. The simplified subnetwork for voiceprint processing is used to replace the complex subnetwork for voiceprint processing in the initial voiceprint analysis network to form the target voiceprint analysis network. Using the target voiceprint analysis network, the target voiceprint information is optimized to output the corresponding optimized voiceprint information. Based on the optimized voiceprint information, an operational anomaly analysis operation is performed on the target hydro-generator to output target operational anomaly analysis data corresponding to the target hydro-generator. The target operational anomaly analysis data is used to reflect the operating status of the target hydro-generator. The complex subnetwork for voiceprint processing includes a number of processing units equal to a first number. The step of adjusting at least two processing units of the complex subnetwork for voiceprint processing to form a simplified subnetwork for voiceprint processing includes: Based on the configured processing unit selection method, a second number of sampling processing units are selected from the first number of processing units in the complex sub-network of voiceprint processing, wherein the second number is less than or equal to the first number. The second number of extraction processing units are adjusted to form a simplified sub-network for voiceprint processing; The second number of extraction processing units belong to the processing units of the first processing method. The processing units of the first processing method are used to perform key information mining operations on the loaded information to be processed in the complex sub-network of voiceprint processing. The processing units of the first processing method include at least one feature mining layer; and the step of adjusting the second number of extraction processing units to form a simplified sub-network of voiceprint processing includes: The feature mining layer in the second number of extraction processing units is adjusted in a first manner to form a second number of processing units including the first feature mining structure; and the second number of processing units including the first feature mining structure are integrated to form a corresponding voiceprint processing simplified sub-network. The step of adjusting the feature mining layer in the second number of extraction processing units in a first manner to form a second number of processing units including a first feature mining structure includes: When the number of feature mining layers in the xth extraction processing unit is equal to one, the feature mining layer in the xth extraction processing unit is marked as a first feature mining structure, and the xth extraction processing unit belongs to one of the second number of extraction processing units; and when the number of feature mining layers in the xth extraction processing unit is greater than one, the feature mining layers in the xth extraction processing unit are integrated into mining layers, and the feature mining layer formed by the integration operation is marked as a first feature mining structure.

2. The artificial intelligence-based acoustic signature monitoring method for hydro-generators as described in claim 1, characterized in that, The second number of extraction processing units includes processing units of the first processing method and processing units of the second processing method; the processing unit of the first processing method is used to perform feature mining operation on the loaded information to be processed in the complex sub-network of voiceprint processing, and the processing unit of the first processing method includes at least one feature mining layer; the processing unit of the second processing method is used to perform input-output linking operation on the loaded information to be processed in the complex sub-network of voiceprint processing, and the number of processing units of the first processing method is equal to the third number, and the third number is less than the second number. as well as The step of adjusting the second number of sampling processing units to form a simplified sub-network for voiceprint processing includes: The third number of processing units using the first processing method are adjusted using a first method to form a third number of processing units including a first feature mining structure; and the fourth number of processing units using the second processing method are adjusted using a second method to form a fourth number of processing units including a second feature mining structure, wherein the fourth number is equal to the difference between the second number and the third number; and the third number of processing units including the first feature mining structure and the fourth number of processing units including the second feature mining structure are integrated to form a corresponding simplified sub-network for voiceprint processing. The processing unit of the first processing method further includes a parameter mapping processing structure, which is used to perform parameter mapping operations on the key data content mined by the feature mining layer in the processing unit of the first processing method.

3. The artificial intelligence-based acoustic signature monitoring method for hydro-generators as described in claim 1, characterized in that, The first number of processing units includes a processing unit of a first processing method, and the processing unit of the first processing method includes at least one feature mining layer; And, the step of selecting a second number of processing units from a first number of processing units in the complex sub-network of voiceprint processing, based on the configuration-based processing unit selection method, includes: From the first number of processing units in the complex sub-network of voiceprint processing, a second number of sampling processing units are arbitrarily selected; or Based on a predetermined target size, a second number of extraction processing units are selected from the processing units of the first processing method, wherein the size of the feature mining layer in the second number of extraction processing units is equal to the target size; or Based on a predetermined target number of samples, a second number of sampling processing units are selected from the processing units of the first processing method. The number of feature mining layers in the second number of sampling processing units is equal to the target number of samples.

4. The artificial intelligence-based acoustic signature monitoring method for hydro-generators as described in claim 1, characterized in that, The target voiceprint analysis network includes a number of simplified voiceprint processing sub-networks equal to the fifth number; the target voiceprint analysis network also includes a key information connection processing unit; the step of using the target voiceprint analysis network to perform voiceprint information optimization operations on the target voiceprint information to output corresponding optimized voiceprint information includes at least: Using the key information connection processing unit, the pending key data content of the target voiceprint information and the key data content mined by the y-th voiceprint processing simplified sub-network are integrated to form the integrated key data content of the target voiceprint information. The y-th voiceprint processing simplified sub-network belongs to one of the fifth number of voiceprint processing simplified sub-networks. Using the next simplified sub-network for voiceprint processing, the integrated key data content of the target voiceprint information is subjected to key information mining operation to output the key information mining output data of the target voiceprint information. The next simplified sub-network for voiceprint processing refers to a simplified sub-network for voiceprint processing that is cascaded after the y-th simplified sub-network for voiceprint processing. Based on the key information of the target voiceprint information, the output data is mined and a feature restoration operation is performed to output the optimized voiceprint information corresponding to the target voiceprint information.

5. The artificial intelligence-based acoustic signature monitoring method for hydro-generators as described in claim 1, characterized in that, The acoustic signature monitoring method for hydro-generators also includes: Using an initial voiceprint analysis network that includes the voiceprint processing complex sub-network, the exemplary voiceprint information is optimized to output the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information. Based on the difference information between the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information and the exemplary tag voiceprint information corresponding to the exemplary voiceprint information, the network parameters included in the voiceprint processing complex sub-network are optimized to form an optimized initial voiceprint analysis network.

6. The artificial intelligence-based acoustic signature monitoring method for hydro-generators as described in claim 5, characterized in that, The complex subnetwork for voiceprint processing includes a first number of processing units, where the first number is greater than one. The step of using the initial voiceprint analysis network including the complex subnetwork for voiceprint processing to perform voiceprint information optimization on the exemplary voiceprint information, and outputting exemplary optimized voiceprint information corresponding to the exemplary voiceprint information, includes: Using the first number of processing units, key information mining operations are performed on the exemplary voiceprint information to output the first number of key data partial contents corresponding to the exemplary voiceprint information. The key data partial contents and the key data contents are represented in the form of vectors. The first number of key data partial contents are integrated to form the integrated key data content corresponding to the exemplary voiceprint information; Based on the integrated key data content corresponding to the exemplary voiceprint information, a feature restoration operation is performed to output the exemplary optimized voiceprint information corresponding to the exemplary voiceprint information.

7. The artificial intelligence-based acoustic signature monitoring method for hydro-generators as described in claim 6, characterized in that, The step of integrating the first number of key data partial contents to form the integrated key data content corresponding to the exemplary voiceprint information includes: A concatenated combination operation is performed on the first number of key data local contents to form a concatenated combination feature distribution corresponding to the first number of key data local contents; The first number of key data local contents are superimposed to form a superimposed feature distribution corresponding to the first number of key data local contents. The cascaded combination feature distribution and the superimposed feature distribution are represented in the form of vectors. The feature distribution size is adjusted for the first number of key data local contents and the cascaded combined feature distribution to form multiple first feature distributions to be processed, and the feature distribution size is consistent among the multiple first feature distributions to be processed. The feature distribution size is adjusted for the first number of key data local contents and the superimposed feature distribution to form multiple second feature distributions to be processed, and the feature distribution size is consistent among the multiple second feature distributions to be processed. Using a first deep feature mining unit formed by network optimization, the focused feature parameter distribution of each of the multiple first feature distributions to be processed in relation to the cascaded combined feature distribution is analyzed to form a corresponding first focused feature analysis result, which is represented in vector form. Using a second deep feature mining unit formed by network optimization, the focus feature parameter distribution of each of the multiple second feature distributions to be processed relative to the superimposed feature distribution is analyzed to form a corresponding second focus feature analysis result, which is represented in vector form. The analysis results of the first focused feature analysis and the second focused feature analysis are aggregated to form the integrated key data content corresponding to the exemplary voiceprint information. The aggregation operation of the analysis results includes the cascading combination or superposition of vectors.

8. An artificial intelligence-based acoustic signature monitoring system for hydro-generators, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, and the processor being used to execute the computer program to implement the method of any one of claims 1-7.

Images