Anaerobic tank stirring equipment operation monitoring method and system based on data analysis

Abstract

The application provides an anaerobic tank stirring equipment operation monitoring method and system based on data analysis, and relates to the technical field of data analysis.In the application, firstly, motor monitoring data and transmission monitoring data are acquired;secondly, first encoding is performed to form motor semantic representations corresponding to the motor monitoring data;then, second encoding is performed to form transmission semantic representations corresponding to the transmission monitoring data, wherein the first encoding comprises semantic encoding performed on data segments formed based on the motor monitoring data in multiple time length divisions, and / or the second encoding comprises repetitive semantic encoding performed based on the correlation between multiple data segments in the transmission monitoring data;finally, based on the motor semantic representations and the transmission semantic representations, equipment operation monitoring results are decoded.The above method can improve the relatively low reliability of anaerobic tank stirring equipment operation monitoring in the prior art.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and more specifically, to a method and system for monitoring the operation of anaerobic tank stirring equipment based on data analysis. Background Technology

[0002] Anaerobic digesters are a crucial part of wastewater treatment, primarily using anaerobic bacteria to transform organic matter in wastewater into simpler substances, thus achieving the goal of removing organic matter. For example, in the hydrolysis stage, complex organic matter (such as proteins, fats, and polysaccharides) is first broken down into simpler compounds, such as amino acids, fatty acids, and monosaccharides, through anaerobic hydrolysis. In the acidification stage, the hydrolyzed organic matter is further converted into volatile fatty acids (VFAs) and other acidic products. These acidic products provide the necessary substrate for the subsequent methanation process. Methanation is one of the most important processes in the anaerobic digester, mainly completed by methanogenic bacteria. These bacteria convert the acidification products into methane gas and carbon dioxide. The methanation reaction not only helps reduce the concentration of organic matter in the water but also produces methane gas, which can be recycled as a renewable energy source.

[0003] The primary purpose of using a mixer in an anaerobic tank is to promote uniform mixing of organic matter, prevent its sedimentation, and improve reaction efficiency. Specifically, mixing accelerates the hydrolysis and acidification stages, ensuring that anaerobic bacteria can effectively contact and decompose organic matter, thus enhancing microbial degradation efficiency. Furthermore, mixing helps maintain temperature and pH uniformity within the reaction tank, preventing unsuitable local environments from affecting microbial activity. Therefore, monitoring the operation of the mixing equipment in the anaerobic tank is fundamental to ensuring the normal and effective operation of wastewater treatment. However, current technologies typically rely on manual monitoring or automatic monitoring based on static rules or threshold comparisons. Both methods fail to fully utilize monitoring data, resulting in low reliability of operational monitoring. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a data analysis-based method and system for monitoring the operation of anaerobic tank mixing equipment, so as to improve the problem of relatively low reliability of the operation monitoring of anaerobic tank mixing equipment in the prior art.

[0005] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: A data analysis-based method for monitoring the operation of anaerobic tank mixing equipment includes: Acquire motor monitoring data at the motor monitoring point and transmission monitoring data at the transmission structure monitoring point of the target anaerobic tank mixing equipment. The motor monitoring data is vibration data or sound data of the motor, and the transmission monitoring data is position data of the transmission structure. The transmission structure includes stirring blades and a transmission rod connecting the stirring blades and the motor. Perform the first encoding to form the motor semantic representation corresponding to the motor monitoring data; A second encoding is performed to form a transmission semantic representation corresponding to the transmission monitoring data. The first encoding includes semantic encoding based on data segments formed by multi-time-length segmentation of the motor monitoring data, and / or the second encoding includes repetitive semantic encoding based on the correlation between multiple data segments in the transmission monitoring data. Based on the motor semantic representation and the transmission semantic representation, the equipment operation monitoring result is decoded, wherein the equipment operation monitoring result is used to characterize whether there is any abnormality in the operation of the target anaerobic tank stirring equipment.

[0006] In some preferred embodiments, in the above-described data analysis-based anaerobic tank stirring equipment operation monitoring method, the step of performing a second encoding to form a transmission semantic representation corresponding to the transmission monitoring data includes: The positional semantics of each transmission monitoring data segment in the transmission monitoring data are mined to form a transmission positional semantic representation of each transmission monitoring data segment; The process involves mining the semantic relationships between the transmission position semantic representations of each transmission monitoring data segment to form corresponding position-related semantic representations, mining repetitive semantics of different time lengths from each position-related semantic representation to form repetitive local semantic representations corresponding to each time length, and determining position repetitive semantic representations based on the repetitive local semantic representations corresponding to each time length. Mining anomalous semantics in the transmission position semantic representation of each transmission monitoring data segment to form a position anomalous semantic representation, wherein, for the first step of mining and determining position repetitive semantic representation and the second step of mining and forming position anomalous semantic representation, at least the first step exists; Based on the location repeatability semantic representation, or based on the location repeatability semantic representation and the location anomaly semantic representation, the transmission semantic representation corresponding to the transmission monitoring data is determined.

[0007] In some preferred embodiments, in the above-described data analysis-based method for monitoring the operation of an anaerobic tank mixing device, the steps of mining the semantic relationships between the semantic representations of the transmission positions of each transmission monitoring data segment to form corresponding position-related semantic representations, mining repetitive semantics of different time lengths from each position-related semantic representation to form repetitive local semantic representations corresponding to each time length, and determining the position repetitive semantic representation based on the repetitive local semantic representations corresponding to each time length, include: In the transmission position semantic representation of each transmission monitoring data segment, determine the transmission position semantic representation of the first transmission monitoring data segment and the transmission position semantic representation of the second transmission monitoring data segment, and determine the relevant relationship parameters corresponding to the transmission position semantic representation of the first transmission monitoring data segment and the transmission position semantic representation of the second transmission monitoring data segment. The transmission position semantic representation of the first transmission monitoring data segment is transformed into rows and columns to form a position transformation semantic representation. Based on the position transformation semantic representation, the relevant relation parameters and the transmission position semantic representation of the second transmission monitoring data segment, the position-related semantic representation between the transmission position semantic representation of the first transmission monitoring data segment and the transmission position semantic representation of the second transmission monitoring data segment is determined. Repetitive semantics of different time lengths are extracted from the location-related semantic representations to form repetitive local semantic representations corresponding to each time length; Semantic aggregation processing is performed based on the repetitive local semantic representations corresponding to each time length to form the corresponding positional repetitive semantic representations.

[0008] In some preferred embodiments, in the above-described data analysis-based method for monitoring the operation of anaerobic tank mixing equipment, the step of mining repetitive semantics of different time lengths from the location-related semantic representations to form repetitive local semantic representations corresponding to each time length includes: For each location-related semantic representation, a corresponding target window is determined based on the time length pre-configured for that location-related semantic representation, and based on the target window, the location-related semantic representation is semantically compressed to form a location-aware semantic representation of that time length. For each location-aware semantic representation, based on the correlation between the parameters in the location-aware semantic representation, a location mining semantic representation is mined from the location-aware semantic representation. The semantic representation of each location is mapped using a multilayer perceptron to form a repetitive local semantic representation corresponding to each time length.

[0009] In some preferred embodiments, in the above-described data analysis-based anaerobic tank stirring equipment operation monitoring method, the step of mining anomalous semantics in the transmission position semantic representation of each transmission monitoring data segment to form a positional anomaly semantic representation includes: For each transmission position semantic representation of the transmission monitoring data segment, the correlation semantic representation between the transmission position semantic representation and the mean semantic representation of all transmission position semantic representations centered on the transmission position semantic representation is determined. Based on the difference between the transmission position semantic representation and the associated semantic representation, the position difference semantic representation of the transmission position semantic representation is determined. Based on the semantic representations of the location differences, common semantics are mined to form a semantic representation of differences and common semantics. Based on the splicing result of the semantic representations of the location differences and the semantic representations of differences and common semantics, deep compression is performed to form a semantic representation of location anomalies.

[0010] In some preferred embodiments, in the above-described data analysis-based anaerobic tank stirring equipment operation monitoring method, the step of performing the first encoding to form the motor semantic representation corresponding to the motor monitoring data includes: The motor monitoring data is divided into A time lengths to form a data segment sequence corresponding to each time length. The data segment sequence includes multiple motor monitoring data segments belonging to the corresponding time length. Each data segment sequence is convolved to form a motor semantic representation sequence corresponding to each data segment sequence, wherein the motor semantic representation sequence includes motor segment semantic representations corresponding to each motor monitoring data segment of the corresponding time length. For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, a multi-time length semantic set is determined based on the motor semantic representation sequence corresponding to the x-th time length. The multi-time length semantic set includes the semantic set corresponding to the y-th motor monitoring data segment from the x-th time length to the A-th time length. The semantic set includes the motor segment semantic representation corresponding to the motor monitoring data segments before the y-th motor monitoring data segment. x and y are both integers greater than or equal to 1. Attention fusion is performed on the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length to form a multi-time-length semantic representation. Based on the multi-time-length semantic representation corresponding to each motor monitoring data segment in each of the aforementioned time lengths, the motor semantic representation corresponding to the motor monitoring data is determined.

[0011] In some preferred embodiments, in the above-described data analysis-based anaerobic tank stirring equipment operation monitoring method, the step of determining a multi-time-length semantic set based on the motor semantic representation sequence corresponding to the x-th time length in the y-th motor monitoring data segment of the data segment sequence corresponding to the x-th time length includes: Based on the number of motor monitoring data segments included in the data segment sequence corresponding to the xth time length, the target screening parameters corresponding to the xth time length are determined; For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, when y-1 is less than or equal to the target filtering parameter corresponding to the x-th time length, the semantic representation of the motor segment corresponding to the y-1 motor monitoring data segment is taken as a semantic set, wherein the y-1 motor monitoring data segment is all motor monitoring data segments in the data segment sequence before the y-th motor monitoring data segment; For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, when y-1 is greater than the target screening parameter corresponding to the x-th time length, the semantic representation of the motor segment corresponding to the nearest target screening parameter before the y-th motor monitoring data segment is taken as the semantic set; For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, the semantic set of the y-th motor monitoring data segment from the x-th time length to the A-th time length is determined as a multi-time length semantic set.

[0012] In some preferred embodiments, in the above-described data analysis-based anaerobic tank stirring equipment operation monitoring method, the step of performing attention fusion on the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length to form a multi-time-length semantic representation includes: The splicing result of the semantic representations of each motor segment in the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length is deeply compressed to form a motor compressed semantic representation. The deep compression includes convolution and attention processing. The result of concatenating the motor compressed semantic representation and the motor segment semantic representation of the y-th motor monitoring data segment is pooled and compressed to form the motor pooled semantic representation; Based on the motor pooling semantic representation mapping, a weighted adjustment parameter is formed, and based on the weighted adjustment parameter, the motor segment semantic representation of the y-th motor monitoring data segment is weighted and adjusted to form a multi-time-length semantic representation.

[0013] In some preferred embodiments, in the above-described data analysis-based anaerobic tank stirring equipment operation monitoring method, the step of decoding the equipment operation monitoring results based on the motor semantic representation and the transmission semantic representation includes: The average value of the motor semantic representation and the transmission semantic representation is calculated to achieve preliminary fusion and form an initial fused semantic representation. Random noise is injected into the initial fused semantic representation to form a noise fused semantic representation. Based on the motor semantic representation, the noise fusion semantic representation is denoised to form a motor denoised semantic representation, and based on the transmission semantic representation, the noise fusion semantic representation is denoised to form a transmission denoised semantic representation. The mean of the motor denoising semantic representation and the transmission denoising semantic representation is calculated to achieve target fusion and form a target fusion semantic representation. The target fusion semantic representation is decoded to form the device operation monitoring results.

[0014] This invention also provides a data analysis-based anaerobic tank mixing equipment operation monitoring system, including a processor and a memory. The memory is used to store computer programs, and the processor is used to execute the computer programs to implement the above-described data analysis-based anaerobic tank mixing equipment operation monitoring method.

[0015] The anaerobic tank stirring equipment operation monitoring method and system based on data analysis provided in this invention first acquires motor monitoring data and transmission monitoring data; second, it performs a first encoding to form a motor semantic representation corresponding to the motor monitoring data; then, it performs a second encoding to form a transmission semantic representation corresponding to the transmission monitoring data. The first encoding includes semantic encoding based on data segments formed by multi-time-length segmentation of the motor monitoring data, and / or, the second encoding includes repetitive semantic encoding based on the correlation between multiple data segments in the transmission monitoring data; finally, it decodes the motor semantic representation and the transmission semantic representation to form the equipment operation monitoring result. Based on the above method, on the one hand, through encoding and decoding, the potential semantic features in the motor monitoring data and transmission monitoring data can be fully utilized, thereby improving the reliability of operation monitoring (compared to conventional static rule judgment or threshold comparison). On the other hand, since the first encoding includes semantic encoding of data segments formed by multi-time-length segmentation of motor monitoring data, the granularity of the encoding is more extensive, resulting in a comprehensive capture of potential semantic features at multiple granularities. This ensures the richness of the resulting motor semantic representation. Furthermore, since the second encoding includes repetitive semantic encoding based on the correlation between multiple data segments in the transmission monitoring data, and the positions of the stirring blades and transmission rods are repetitive, repetitive semantic encoding can capture strong characteristic semantic information in the transmission monitoring data, resulting in higher accuracy of the resulting transmission semantic representation. This further improves the reliability of the decoded equipment operation monitoring results. Based on this, the improved solution of this invention can address the problem of relatively low reliability in the operation monitoring of anaerobic tank stirring equipment in the prior art.

[0016] 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

[0017] Figure 1 The structural block diagram of the anaerobic tank stirring equipment operation monitoring system based on data analysis provided in the embodiments of the present invention.

[0018] Figure 2 This is a flowchart illustrating the steps of the data analysis-based anaerobic tank mixing equipment operation monitoring method provided in this embodiment of the invention.

[0019] Figure 3 This is a schematic diagram of vibration data provided in an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram of sound data provided in an embodiment of the present invention.

[0021] Figure 5 This is a schematic diagram of the first encoding provided in an embodiment of the present invention.

[0022] Figure 6 This is a schematic diagram of attention fusion provided in an embodiment of the present invention.

[0023] Figure 7 This is a schematic diagram of the second encoding provided in an embodiment of the present invention.

[0024] Figure 8 This is a schematic diagram of repetitive semantic mining provided in an embodiment of the present invention.

[0025] Figure 9 This is a schematic diagram of decoding provided in an embodiment of the present invention. Detailed Implementation

[0026] 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.

[0027] 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.

[0028] like Figure 1 As shown in the figure, this embodiment of the invention provides an anaerobic tank stirring equipment operation monitoring system based on data analysis, which may include a memory and a processor.

[0029] 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 data analysis-based anaerobic tank stirring equipment operation monitoring method provided in this embodiment of the invention.

[0030] Optionally, 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.

[0031] Optionally, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may 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.

[0032] Optionally, the data analysis-based anaerobic tank mixing equipment operation monitoring system can be a server or server cluster with data processing capabilities.

[0033] Combination Figure 2 This invention also provides a data analysis-based method for monitoring the operation of an anaerobic tank mixing device, which can be applied to the aforementioned data analysis-based anaerobic tank mixing device operation monitoring system. The method steps defined in the relevant process of the data analysis-based anaerobic tank mixing device operation monitoring method can be implemented by the data analysis-based anaerobic tank mixing device operation monitoring system. The following will describe... Figure 2 The specific process shown will be explained in detail.

[0034] Step S110: Obtain motor monitoring data at the motor monitoring point and transmission monitoring data at the transmission structure monitoring point of the target anaerobic tank stirring equipment.

[0035] In this embodiment of the invention, the data analysis-based anaerobic tank mixing equipment operation monitoring system can acquire motor monitoring data at the motor monitoring points and transmission monitoring data at the transmission structure monitoring points of the target anaerobic tank mixing equipment. The motor monitoring data includes vibration or sound data of the motor. For example, sound or vibration sensors installed on the motor housing can collect the sound or vibration generated during motor operation to reflect the motor's operating status. Generally, abnormal sound or vibration may be generated when the motor malfunctions. The transmission monitoring data is the position data of the transmission structure, which includes stirring blades and a transmission rod connecting the stirring blades and the motor. That is, the position of the transmission structure, such as the position of the stirring blades, can be collected using position sensors installed at any location on the transmission structure.

[0036] Step S120: Perform the first encoding to form the motor semantic representation corresponding to the motor monitoring data.

[0037] In this embodiment of the invention, after obtaining the motor monitoring data, the anaerobic tank stirring equipment operation monitoring system based on data analysis can perform a first encoding to form a motor semantic representation corresponding to the motor monitoring data. That is, the motor monitoring data can be first encoded to capture the potential semantic features in the motor monitoring data, and then represented by a vector (which can be a one-dimensional vector, such as 1...). n can also be a two-dimensional vector, such as m. n can also be a three-dimensional vector, such as h. m The motor semantic representation is obtained by representing it in the form of n).

[0038] Step S130: Perform the second encoding to form the transmission semantic representation corresponding to the transmission monitoring data.

[0039] In this embodiment of the invention, after obtaining the transmission monitoring data, the anaerobic tank stirring equipment operation monitoring system based on data analysis can perform a second encoding to form a transmission semantic representation corresponding to the transmission monitoring data. That is, the transmission monitoring data can be second-encoded to capture the potential semantic features in the transmission monitoring data, and then encoded using a vector (which can be a one-dimensional vector, such as 1...). n can also be a two-dimensional vector, such as m. n can also be a three-dimensional vector, such as h. m The transmission semantic representation is obtained by representing the data in the form of n). The first encoding includes semantic encoding based on data segments formed by multi-time-length segmentation of the motor monitoring data to improve the richness of the semantic encoding, and / or the second encoding includes repetitive semantic encoding based on the correlation between multiple data segments in the transmission monitoring data to improve the accuracy of the semantic encoding, such as the repetitive and periodic characteristics of position data.

[0040] Step S140: Based on the motor semantic representation and the transmission semantic representation, decode to form the equipment operation monitoring result.

[0041] In this embodiment of the invention, after obtaining the motor semantic representation and the transmission semantic representation, the data analysis-based anaerobic tank mixing equipment operation monitoring system can decode and form equipment operation monitoring results based on the motor semantic representation and the transmission semantic representation. The equipment operation monitoring results are used to characterize whether there are any abnormalities in the operation of the target anaerobic tank mixing equipment.

[0042] Based on the above method, on the one hand, through encoding and decoding, the latent semantic features in motor monitoring data and transmission monitoring data can be fully utilized, thereby improving the reliability of operation monitoring (compared to conventional static rule judgment or threshold comparison). On the other hand, since the first encoding includes semantic encoding based on data segments formed by multi-time-length segmentation of motor monitoring data, the granularity of the encoding is more extensive, resulting in a comprehensive capture of multi-granular latent semantic features, thus ensuring the richness of the formed motor semantic representation. Furthermore, since the second encoding includes repetitive semantic encoding based on the correlation between multiple data segments in transmission monitoring data, and the positions of the stirring blade and transmission rod are repetitive, repetitive semantic encoding can capture strong characteristic semantic information in the transmission monitoring data, resulting in higher accuracy of the formed transmission semantic representation. This can further improve the reliability of the equipment operation monitoring results formed by decoding. Based on this, the improved solution of the embodiments of the present invention can improve the problem of relatively low reliability of anaerobic tank stirring equipment operation monitoring in the prior art.

[0043] Based on the above implementation method, further explanation is needed for step S110. The specific process of obtaining motor monitoring data of motor monitoring points and transmission monitoring data of transmission structure monitoring points is not limited and can be configured according to actual needs.

[0044] For example, in one specific implementation, to improve the timeliness of operation monitoring, motor monitoring data and transmission monitoring data collected by relevant sensors can be acquired in real time. After acquiring the data, subsequent encoding and decoding can be performed to obtain the current equipment operation monitoring results, i.e., to determine whether there are any anomalies, so as to facilitate timely maintenance. Furthermore, it should be noted that the acquired motor monitoring data and transmission monitoring data can be either raw data or pre-processed data, such as data after sampling processing.

[0045] Based on the above implementation method, further explanation is needed for step S120. That is, the specific process of performing the first encoding to form the motor semantic representation corresponding to the motor monitoring data is not limited and can be configured according to actual needs.

[0046] For example, in one specific implementation, the motor monitoring data can be time-series data, such as... Figure 3 The vibration data shown (i.e., the horizontal axis represents time and the vertical axis represents vibration amplitude) was obtained, as follows: Figure 4 The sound data shown is presented (i.e., the horizontal axis represents time, and the vertical axis represents sound amplitude). Thus, a one-dimensional convolution (e.g., a convolution kernel of 1) can be performed on the motor monitoring data. It should be noted that the motor monitoring data can be normalized before convolution to obtain the corresponding motor semantic representation. Specifically, to achieve segmentation across multiple time lengths and improve the richness of semantic encoding, the motor monitoring data can be segmented based on multiple time lengths. Then, a one-dimensional convolution is performed on the segmentation results for each time length to obtain the motor semantic representation.

[0047] For example, in another specific implementation, in order to ensure the semantic capture accuracy of the first encoding and to make the resulting motor semantic representation take into account both semantic richness and semantic accuracy, the above step S120 may further include sub-steps S121, S122, S123, S124 and S125.

[0048] Sub-step S121: The motor monitoring data is divided into A time segments to form a data segment sequence corresponding to each time segment.

[0049] In this embodiment of the invention, combined with Figure 5The motor monitoring data can be segmented based on A time lengths to form a data segment sequence corresponding to each time length. Each data segment sequence includes multiple motor monitoring data segments belonging to a given time length. For example, the data segment sequence corresponding to the first time length may include four motor monitoring data segments, and the data segment sequence corresponding to the second time length may include two motor monitoring data segments.

[0050] Sub-step S122 involves performing a convolution operation on each data segment sequence to form a motor semantic representation sequence corresponding to each data segment sequence.

[0051] In this embodiment of the invention, after obtaining the data segment sequence, each data segment sequence can be convolved to form a motor semantic representation sequence corresponding to each data segment sequence. The motor semantic representation sequence includes motor segment semantic representations corresponding to each motor monitoring data segment of a corresponding time length. For example, the four motor monitoring data segments included in the data segment sequence corresponding to the first time length can be convolved separately (as mentioned above, a one-dimensional convolution can be performed on the normalized motor monitoring data segments to obtain the corresponding motor segment semantic representations), and the obtained motor segment semantic representations can be combined to form the motor semantic representation sequence corresponding to the data segment sequence of the first time length (including four motor segment semantic representations). As another example, the two motor monitoring data segments included in the data segment sequence corresponding to the second time length can be convolved separately, and the obtained motor segment semantic representations can be combined to form the motor semantic representation sequence corresponding to the data segment sequence of the second time length (including two motor segment semantic representations).

[0052] Sub-step S123: For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, determine the multi-time length semantic set based on the motor semantic representation sequence corresponding to the x-th time length.

[0053] In this embodiment of the invention, after obtaining the motor semantic representation sequence, for the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, a multi-time-length semantic set is determined based on the motor semantic representation sequence corresponding to the x-th time length. The multi-time-length semantic set includes the semantic set corresponding to the y-th motor monitoring data segment from the x-th time length to the A-th (i.e., the last) time length. This semantic set includes the motor segment semantic representations corresponding to motor monitoring data segments preceding the y-th motor monitoring data segment (i.e., including the motor segment semantic representations corresponding to historical motor monitoring data segments, i.e., only focusing on historical semantic features), where x and y are both integers greater than or equal to 1. For example, for the first motor monitoring data segment in the data segment sequence corresponding to the first time length, the corresponding multi-time-length semantic set includes the semantic set of the first motor monitoring segment in the first time length and the semantic set of the first motor monitoring segment in the second time length. For example, for the first motor monitoring data segment in the data segment sequence corresponding to the second time length, the corresponding multi-time length semantic set includes the semantic set of the first motor monitoring segment in the second time length and the semantic set of the first motor monitoring segment in the second time length. Furthermore, it should be noted that the time lengths gradually increase; that is, the first time length is shorter than the second time length, or in other words, the number of motor monitoring data segments included in the data segment sequence corresponding to the first time length is greater than the number of motor monitoring data segments included in the data segment sequence corresponding to the second time length. Based on this, it should also be noted that different time lengths result in different coding granularities. That is, segmenting the motor monitoring data using different time lengths can obtain motor monitoring data segments of different granularities. Thus, a shorter time length can capture local details in the motor monitoring data, while a longer time length can effectively cover the global information of the motor monitoring data, or the overall information from a broader perspective. Furthermore, since a multi-time-length semantic set is formed, smaller time lengths can access the overall information provided by larger time lengths. This allows the receptive field to be expanded through the attention mechanism of multiple time lengths and historical semantic features, considering not only the historical semantic features of the current time length but also the historical semantic features of all larger time lengths, thereby maximizing the capture of temporal relationships.

[0054] Sub-step S124 involves performing attention fusion on the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, to form a multi-time-length semantic representation.

[0055] In this embodiment of the invention, after obtaining the multi-time-length semantic set, the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time-length can be fused with attention to form a multi-time-length semantic representation. Thus, each motor monitoring data segment can form a corresponding multi-time-length semantic representation.

[0056] Sub-step S125: Based on the multi-time-length semantic representation corresponding to each motor monitoring data segment in each time length, determine the motor semantic representation corresponding to the motor monitoring data.

[0057] In this embodiment of the invention, after obtaining the multi-time-length semantic representation, the motor semantic representation corresponding to the motor monitoring data can be determined based on the multi-time-length semantic representation corresponding to each motor monitoring data segment within each time length. That is, the multi-time-length semantic representations corresponding to each motor monitoring data segment within each time length can be aggregated (e.g., through splicing or averaging). This allows for the acquisition of semantic features of various granularities captured from multiple time lengths and multiple local data, resulting in a motor semantic representation that balances semantic richness and semantic accuracy, effectively improving the semantic representation capability of the motor semantic representation.

[0058] Based on the above implementation method, further explanation is needed for sub-step S123. That is, the specific process of determining the multi-time-length semantic set is not limited. For example, in a specific implementation method, in order to balance the semantic richness based on the determined multi-time-length semantic set and the computational efficiency of subsequent attention fusion, the above step S123 may further include sub-steps S123a, S123b, S123c and S123d.

[0059] Sub-step S123a: Based on the number of motor monitoring data segments included in the data segment sequence corresponding to the xth time length, determine the target screening parameters corresponding to the xth time length.

[0060] In this embodiment of the invention, the target screening parameter corresponding to the xth time length can be determined based on the number of motor monitoring data segments included in the data segment sequence corresponding to the xth time length. For example, the target screening parameter = B(b The number of motor monitoring data segments), where b is a coefficient greater than 0 and less than 1, and B() can represent rounding, such as rounding down. In other words, the determined target screening parameter needs to be less than the number of motor monitoring data segments.

[0061] Sub-step S123b: For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, when y-1 is less than or equal to the target screening parameter corresponding to the x-th time length, the semantic representation of the motor segment corresponding to the y-1 motor monitoring data segment is taken as the semantic set.

[0062] In this embodiment of the invention, after obtaining the target screening parameters, for the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, if y-1 is less than or equal to the target screening parameters corresponding to the x-th time length (the number of preceding motor monitoring data segments is equal to or less than the target screening parameters), the semantic representations of the motor segments corresponding to the y-1 motor monitoring data segments are taken as a semantic set. Here, the y-1 motor monitoring data segments are all motor monitoring data segments in the data segment sequence preceding the y-th motor monitoring data segment, i.e., the semantic representations of the motor segments corresponding to each preceding motor monitoring data segment are taken as a semantic set.

[0063] Sub-step S123c: For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, when y-1 is greater than the target screening parameter corresponding to the x-th time length, the semantic representation of the motor segment corresponding to the nearest target screening parameter before the y-th motor monitoring data segment is taken as the semantic set.

[0064] In this embodiment of the invention, after obtaining the target screening parameters, for the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, when y-1 is greater than the target screening parameters corresponding to the x-th time length (the number of preceding motor monitoring data segments is greater than the target screening parameters), the semantic representation of the motor segment corresponding to the nearest target screening parameters before the y-th motor monitoring data segment is taken as a semantic set, that is, the number of motor segment semantic representations in the semantic set is equal to the target screening parameters.

[0065] Sub-step S123d: For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, determine the semantic set of the y-th motor monitoring data segment from the x-th time length to the A-th time length as a multi-time length semantic set.

[0066] In this embodiment of the invention, after obtaining the semantic set, for the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, the semantic set of the y-th motor monitoring data segment from the x-th time length to the A-th time length is determined as a multi-time length semantic set, that is, it integrates the current time length and every subsequent time length, so that the semantic field of coverage is expanded to multiple, thereby improving the semantic representation capability.

[0067] Based on the above implementation method, further explanation is needed for sub-step S124. That is, the specific process of forming multi-time-length semantic representation is not limited. For example, in a specific implementation method, in order to further improve the accuracy of attention fusion, the above sub-step S124 may further include sub-step S124a, sub-step S124b and sub-step S124c.

[0068] Sub-step S124a involves deeply compressing the splicing result of the semantic representations of each motor segment in the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, to form a compressed semantic representation of the motor.

[0069] In this embodiment of the invention, combined with Figure 6 The concatenation result of the semantic representations of each motor segment in the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length can be deeply compressed to form a compressed motor semantic representation. Deep compression includes convolution and attention processing. That is, the concatenation result can first be processed by convolution to achieve size compression and extraction of high-level semantic features, capturing more complex semantic information. Then, after size compression, such as to the size of a motor segment semantic representation, self-attention processing can be applied to the convolution result to capture the correlation between various local features, thus achieving full interaction and fusion of different motor segment semantic representations.

[0070] Sub-step S124b involves pooling and compressing the result of splicing the motor compressed semantic representation and the motor segment semantic representation of the y-th motor monitoring data segment to form a motor pooled semantic representation.

[0071] In this embodiment of the invention, after obtaining the compressed semantic representation of the motor, the result of concatenating the compressed semantic representation of the motor and the semantic representation of the motor segment of the y-th motor monitoring data segment can be pooled and compressed to form a pooled semantic representation of the motor. For example, the size compression of the concatenated result can be achieved by means of average pooling or max pooling, so that the size of the formed pooled semantic representation of the motor is equal to the size of the semantic representation of the motor segment.

[0072] Sub-step S124c involves forming a weighted adjustment parameter based on the motor pooling semantic representation mapping, and then, based on the weighted adjustment parameter, performing weighted adjustment on the motor segment semantic representation of the y-th motor monitoring data segment to form a multi-time-length semantic representation.

[0073] In this embodiment of the invention, after obtaining the motor pooling semantic representation, weighted adjustment parameters can be formed based on the mapping of the motor pooling semantic representation (e.g., performing a linear mapping without changing the size, y=Wx+z, where W and z are the weight matrix and bias parameter formed during training, respectively; then, nonlinear activation is applied to the result of the linear mapping, such as through a sigmoid function). Based on the weighted adjustment parameters, the semantic representation of the motor segment of the y-th motor monitoring data segment is weighted (e.g., by bitwise multiplication to achieve weighting) to form a multi-time-length semantic representation. It should be noted that in traditional gating mechanisms, the compressed motor semantic representation is typically mapped directly to form weighted adjustment parameters, and then the semantic representation of the motor segment is weighted based on these parameters to obtain a multi-time-length semantic representation. The quality of the result obtained in this way is highly dependent on the stability of the compressed motor semantic representation. Therefore, considering the poor stability of the compressed motor semantic representation, a gating mechanism is not used. Instead, an attention mechanism with higher data processing accuracy is adopted, i.e., cross-attention processing is performed on the semantic representation of the motor segment based on the compressed motor semantic representation. This approach leads to a significant increase in computational load. However, based on the scheme in the embodiments of the present invention, before mapping to form weighted adjustment parameters, the motor compression semantic representation and the motor segment semantic representation are fused first. That is, the motor segment semantic representation is fused to perform corresponding constraints, which makes the stability of the motor pooling semantic representation higher. This ensures that the weighted adjustment parameters formed by the subsequent gating processing not only consider the relevant motor compression semantic representation, but also the motor segment semantic representation itself, making the reliability of the weighted adjustment parameters higher.

[0074] Based on the above implementation method, further explanation is needed for step S130. That is, the specific process of performing the second encoding to form the transmission semantic representation corresponding to the transmission monitoring data is not limited and can be configured according to actual needs.

[0075] For example, in one specific implementation, in order to further improve the semantic capture capability of the second encoding, such as to make the semantic representation capability of the transmission semantic representation better, the above step S130 may further include sub-steps S131, S132, S133 and S134.

[0076] Sub-step S131: Mine the positional semantics of each transmission monitoring data segment in the transmission monitoring data to form a transmission positional semantic representation of each transmission monitoring data segment.

[0077] In this embodiment of the invention, combined with Figure 7This method can mine the positional semantics of each transmission monitoring data segment in the transmission monitoring data to form a transmission positional semantic representation for each segment. It should be noted that the transmission monitoring data segments are formed by segmenting the transmission monitoring data; these segments can be either non-repeating or repetitive (i.e., adjacent transmission monitoring data segments have overlapping parts). Furthermore, the specific method for mining the positional semantics of the transmission monitoring data segments is not limited. For example, in an alternative implementation, the transmission monitoring data segments (i.e., position data segments) can be normalized, and then convolution processing can be performed to obtain the corresponding transmission positional semantic representation. It should also be noted that if the position data is in two-dimensional coordinates, two-dimensional convolution can be performed directly, or the two-dimensional coordinates can be mapped to form one-dimensional data (similar to converting multi-channel color values ​​to grayscale values), and then one-dimensional convolution can be performed. If the position data is in three-dimensional coordinates, three-dimensional convolution can be performed directly, or the three-dimensional coordinates can be mapped to form one-dimensional data, and then one-dimensional convolution can be performed.

[0078] Sub-step S132: Mining the related semantics between the transmission position semantic representations of each transmission monitoring data segment to form corresponding position-related semantic representations; mining repetitive semantics of different time lengths from each position-related semantic representation to form repetitive local semantic representations corresponding to each time length; and determining position repetitive semantic representations based on the repetitive local semantic representations corresponding to each time length.

[0079] In this embodiment of the invention, after obtaining the transmission position semantic representation, the related semantics between the transmission position semantic representations of each transmission monitoring data segment can be mined to form corresponding position-related semantic representations. Furthermore, repetitive semantics of different time lengths can be mined from each position-related semantic representation to form repetitive local semantic representations corresponding to each time length. Finally, based on the repetitive local semantic representations corresponding to each time length, the position repetitive semantic representation is determined. It should be noted that the movement of the stirring blade and the transmission rod generally involves repetitive or reciprocating actions. Therefore, the related semantics between the transmission position semantic representations of each transmission monitoring data segment can effectively characterize this repetitiveness. Additionally, by capturing repetitive semantics in different time lengths, i.e., different periods, the comprehensiveness of capturing repetitive semantics can be improved, i.e., capturing repetitive semantics across multiple time lengths.

[0080] Sub-step S133: Mine the abnormal semantics in the transmission position semantic representation of each transmission monitoring data segment to form a position abnormal semantic representation.

[0081] In this embodiment of the invention, abnormal semantics in the transmission position semantic representation of each transmission monitoring data segment can also be mined to form a position anomaly semantic representation. Specifically, for the first step of mining and determining position repetitive semantic representation and the second step of mining and forming position anomaly semantic representation, at least the first step exists. That is, only sub-step S132 can be executed, or both sub-steps S132 and S133 can be executed. Furthermore, it should be noted that in this embodiment of the invention, the purpose of encoding and decoding is to determine whether there are any anomalies in the operation. Therefore, by mining abnormal semantics, the encoding can be more targeted, and the accuracy of the semantic representation can be improved.

[0082] Sub-step S134: Based on the position repeatability semantic representation, or based on the position repeatability semantic representation and the position anomaly semantic representation, determine the transmission semantic representation corresponding to the transmission monitoring data.

[0083] In this embodiment of the invention, after obtaining the position repeatability semantic representation and the position anomaly semantic representation, the transmission semantic representation corresponding to the transmission monitoring data can be determined based on the position repeatability semantic representation, or based on the position repeatability semantic representation and the position anomaly semantic representation. For example, the position repeatability semantic representation can be directly determined as the transmission semantic representation. Alternatively, the result of aggregating the position repeatability semantic representation and the position anomaly semantic representation (such as through summation, averaging, or concatenation) can be used as the transmission semantic representation.

[0084] Based on the above implementation method, further explanation is needed for sub-step S132. That is, the specific process of determining the positional repetition semantic representation is not limited. For example, in a specific implementation method, in order to improve the accuracy of related semantic capture and make the reliability of the formed positional repetition semantic representation higher, the above-mentioned sub-step S132 may further include sub-steps S132a, S132b, S132c and S132d.

[0085] Sub-step S132a involves determining the transmission position semantic representation of the first transmission monitoring data segment and the second transmission monitoring data segment from the transmission position semantic representation of each transmission monitoring data segment, and determining the relevant relationship parameters corresponding to the transmission position semantic representation of the first transmission monitoring data segment and the transmission position semantic representation of the second transmission monitoring data segment.

[0086] In this embodiment of the invention, the semantic representations of the transmission positions of a first transmission monitoring data segment and a second transmission monitoring data segment are determined from the semantic representations of the transmission positions of each transmission monitoring data segment. Furthermore, the correlation parameters corresponding to the semantic representations of the transmission positions of the first and second transmission monitoring data segments are determined. It should be noted that the first and second transmission monitoring data segments refer to any two transmission monitoring data segments for which position-related semantic representations need to be determined. Additionally, the correlation parameters can be obtained by calculating the cosine similarity between the two transmission position semantic representations, or they can be pre-configured or trained for the corresponding two transmission monitoring data segments.

[0087] Sub-step S132b involves performing row and column transformation on the transmission position semantic representation of the first transmission monitoring data segment to form a position transformation semantic representation, and determining the position-related semantic representation between the transmission position semantic representation of the first transmission monitoring data segment and the transmission position semantic representation of the second transmission monitoring data segment based on the position transformation semantic representation, the relevant relational parameters, and the transmission position semantic representation of the second transmission monitoring data segment.

[0088] In this embodiment of the invention, the transmission position semantic representation of the first transmission monitoring data segment can be transformed (i.e., transposed) to form a position transformation semantic representation. Furthermore, based on the position transformation semantic representation, the correlation parameter, and the transmission position semantic representation of the second transmission monitoring data segment, a position-related semantic representation between the transmission position semantic representations of the first and second transmission monitoring data segments is determined. For example, the correlation parameter can be multiplied by the transmission position semantic representation of the second transmission monitoring data segment to achieve correlation weighting. Then, the position transformation semantic representation and the correlation weighting result can be multiplied (matrix multiplication) to obtain the corresponding position-related semantic representation. It should be noted that performing matrix multiplication of the position transformation semantic representation and the transmission position semantic representation of the second transmission monitoring data segment actually yields the distribution of the correlation parameter between the two transmission position semantic representations. However, in this embodiment of the invention, by adding the correlation parameter as a weight for adjustment, the accuracy of the formed position-related semantic representation in representing the correlation relationship can be further improved.

[0089] Sub-step S132c involves mining repetitive semantics of different time lengths from the location-related semantic representations of each location, forming repetitive local semantic representations corresponding to each time length.

[0090] In this embodiment of the invention, after obtaining the location-related semantic representations (obtained by pairwise combination calculation of each transmission monitoring data segment as described above), repetitive semantics of different time lengths can be mined from the location-related semantic representations to form repetitive local semantic representations corresponding to each time length. That is, although the transmission monitoring data segments have the same time length, making the time lengths corresponding to the location-related semantic representations the same, further mining can be performed according to different time lengths, resulting in different perspectives for further mining of the location-related semantic representations, thereby further improving the semantic mining capability, i.e., capturing as many potential semantic features as possible.

[0091] Sub-step S132d involves performing semantic aggregation processing based on the repetitive local semantic representations corresponding to each time length to form the corresponding positional repetitive semantic representation.

[0092] In this embodiment of the invention, after obtaining the repetitive local semantic features, semantic aggregation processing can be performed based on the repetitive local semantic representations corresponding to each time length to form corresponding positional repetitive semantic representations. For example, the repetitive local semantic representations corresponding to each time length can be aggregated by methods such as splicing, averaging, and summing to form positional repetitive semantic representations.

[0093] Based on the above implementation method, further explanation is needed for sub-step S132c. That is, the specific process of mining repetitive semantics of different time lengths from the position-related semantic representations is not limited. For example, in a specific implementation method, in order to effectively capture repetitive semantics of different time lengths, the above sub-step S132c may further include sub-steps c1, c2 and c3.

[0094] Sub-step c1: For each location-related semantic representation, determine the corresponding target window based on the time length pre-configured for that location-related semantic representation, and based on the target window, perform semantic compression on the location-related semantic representation to form a location-aware semantic representation of that time length.

[0095] In this embodiment of the invention, combined with Figure 8For each location-related semantic representation, a corresponding target window is determined based on a pre-configured time length for that location-related semantic representation. Then, based on this target window, semantic compression is performed on the location-related semantic representation to form a location-aware semantic representation for that time length. For example, different sized target windows can be pre-configured for different location-related semantic representations (because the window size is different, the time length corresponding to the semantic features of interest is different), enabling attention to local semantic features at different scales. The target window can serve as the window for pooling processing, thus enabling compression and perception of different location-related semantic representations from different perspectives; that is, semantic compression is achieved through pooling processing.

[0096] Sub-step c2: For each location-aware semantic representation, based on the correlation between the parameters in the location-aware semantic representation, mine the location mining semantic representation of the location-aware semantic representation.

[0097] In this embodiment of the invention, after obtaining the location-aware semantic representation, for each location-aware semantic representation, a location mining semantic representation can be mined based on the correlation between the parameters in the location-aware semantic representation. Specifically, self-attention processing can be applied to the location-aware semantic representation to capture the correlation between its internal semantics.

[0098] Sub-step c3 involves mapping the semantic representation mined at each location using a multilayer perceptron to form repetitive local semantic representations corresponding to each time length.

[0099] In this embodiment of the invention, after obtaining the location mining semantic representation, each location mining semantic representation can be mapped using a multilayer perceptron (MLP) (thus, more linear relationships can be captured), forming repetitive local semantic representations corresponding to each time length.

[0100] Based on the above implementation method, further explanation is needed for sub-step S133. That is, the specific process of mining abnormal semantics in the semantic representation of the transmission position of each transmission monitoring data segment is not limited. For example, in a specific implementation method, in order to achieve comprehensive capture of abnormal semantics, the above sub-step S133 may further include sub-steps S133a, S133b and S133c.

[0101] Sub-step S133a: For each transmission monitoring data segment, determine the associated semantic representation between the transmission position semantic representation and the mean semantic representation of each transmission position semantic representation centered on the transmission position semantic representation.

[0102] In this embodiment of the invention, for each transmission position semantic representation of the transmission monitoring data segment, the associated semantic representation between the transmission position semantic representation and the mean semantic representation of all transmission position semantic representations centered on the transmission position semantic representation can be determined. For example, a length can be determined, and a range can be defined centered on a certain transmission position semantic representation. Then, the mean of each transmission position semantic representation within this range is calculated to obtain the corresponding mean semantic representation. After that, the associated semantic representation between the central transmission position semantic representation and the mean semantic representation can be determined. Thus, the central transmission position semantic representation and the mean semantic representation are subjected to cross-attention processing to capture the corresponding associated semantic relationship and obtain the associated semantic representation. In addition, it should be noted that the arrangement relationship of the transmission position semantic representations can be determined based on the arrangement relationship between the corresponding transmission monitoring data segments.

[0103] Sub-step S133b involves performing a difference calculation based on the transmission position semantic representation and the associated semantic representation to determine the position difference semantic representation of the transmission position semantic representation.

[0104] In this embodiment of the invention, after obtaining the associated semantic representation, a difference processing operation can be performed based on the transmission position semantic representation and the associated semantic representation to determine the positional difference semantic representation of the transmission position semantic representation. That is, since the associated semantic representation is a feature extracted from the transmission position semantic representation that relates to other adjacent transmission position semantic representations, by performing a difference processing operation, unrelated semantic features can be captured, thus obtaining a positional difference semantic representation that can reflect some differing semantic features.

[0105] Sub-step S133c involves mining common semantics based on the semantic representations of each location difference to form a semantic representation of differences and common semantics, and performing deep compression based on the splicing result of the semantic representations of each location difference and the semantic representations of differences and common semantics to form a semantic representation of location anomalies.

[0106] In this embodiment of the invention, after obtaining the positional difference semantic representation, common semantic mining can be performed based on each positional difference semantic representation to form a difference-common semantic representation. Furthermore, deep compression can be performed on the concatenation result of each positional difference semantic representation and the difference-common semantic representation to form a positional anomaly semantic representation. It should be noted that since each positional difference semantic representation represents its own unique semantic features, these features include both locally unique features, such as anomalies occurring only once, and common features, such as anomalies occurring multiple times. Therefore, by performing common semantic mining, such common, multiple-occurrence anomalies can be captured. This provides a more comprehensive representation of the inherent anomalies of the target anaerobic tank stirring equipment. Therefore, these anomalies can be captured separately and then combined with each positional difference semantic representation as a positional anomaly semantic representation. Deep compression can be achieved through convolution and / or pooling, etc. Furthermore, common semantic mining can be achieved through an attention mechanism. For example, the positional difference semantic representations can be arranged according to the arrangement relationship between the corresponding transmission monitoring data segments to form a corresponding sequence. Then, based on the attention mechanism, the previous positional difference semantic representation is sequentially fused into the next positional difference semantic representation until the last positional difference semantic representation is fused, and then the difference common semantic representation is output. Alternatively, the positional difference semantic representations can be concatenated and then subjected to self-attention processing to obtain the difference common semantic representation.

[0107] Based on the above implementation method, further explanation is needed for step S140. That is, the specific process of decoding the motor semantic representation and the transmission semantic representation to form the equipment operation monitoring result is not limited and can be configured according to actual needs.

[0108] For example, in one specific implementation, the motor semantic representation and the transmission semantic representation can be concatenated, and then the concatenated result can be decoded to obtain the equipment operation monitoring result. In this way, the decoding efficiency can be improved and the computational overhead can be reduced.

[0109] For example, in another specific implementation, in order to improve the decoding accuracy and make the obtained device operation monitoring results more reliable, the above step S140 may further include sub-steps S141, S142, S143 and S144.

[0110] Sub-step S141 involves calculating the mean of the motor semantic representation and the transmission semantic representation to achieve preliminary fusion and form an initial fused semantic representation, and injecting random noise into the initial fused semantic representation to form a noise fused semantic representation.

[0111] In this embodiment of the invention, combined with Figure 9 The semantic representations of the motor and the transmission can be averaged to achieve preliminary fusion, forming an initial fused semantic representation. Random noise can then be injected into the initial fused semantic representation to form a noisy fused semantic representation. It should be noted that since the semantic representations of the motor and the transmission belong to different dimensions of semantic features, directly averaging them for preliminary fusion may result in poor fusion performance due to semantic mismatch. Therefore, in this embodiment, the result of the preliminary fusion can be used as an intermediate semantic feature, i.e., it has correlations with both the motor and transmission semantic representations. This allows for further fusion with both the motor and transmission semantic representations. However, considering issues such as overfitting, noise can be further introduced to ensure semantic diversity.

[0112] Sub-step S142: Based on the motor semantic representation, the noise fusion semantic representation is denoised to form a motor denoised semantic representation, and based on the transmission semantic representation, the noise fusion semantic representation is denoised to form a transmission denoised semantic representation.

[0113] In this embodiment of the invention, after obtaining the noise fusion semantic representation, on the one hand, the noise fusion semantic representation can be denoised based on the motor semantic representation to form a motor denoised semantic representation. On the other hand, the noise fusion semantic representation can be denoised based on the transmission semantic representation to form a transmission denoised semantic representation. That is, during the denoising process, further fusion of semantic features is achieved. Specifically, by capturing the relevant semantic features between the motor semantic representation and the noise fusion semantic representation, the noise injected into the noise fusion semantic representation can be suppressed or removed, thereby achieving a fusion between motor-related semantic features and transmission-related semantic features. Furthermore, by capturing the relevant semantic features between the transmission semantic representation and the noise fusion semantic representation, the noise injected into the noise fusion semantic representation can be suppressed or removed, thereby achieving another fusion between motor-related semantic features and transmission-related semantic features. The capture of relevant semantic features can be achieved through a cross-attention mechanism.

[0114] Sub-step S143 involves averaging the motor denoising semantic representation and the transmission denoising semantic representation to achieve target fusion and form a target fusion semantic representation.

[0115] In this embodiment of the invention, after obtaining the motor denoising semantic representation and the transmission denoising semantic representation, the average of the motor denoising semantic representation and the transmission denoising semantic representation can be calculated to achieve target fusion and form a target fused semantic representation. That is, since the motor denoising semantic representation and the transmission denoising semantic representation are formed through deep fusion of an intermediate noise fusion semantic representation, deep fusion of different dimensions has been achieved. Thus, they are adapted in the semantic feature space, allowing for further fusion through average calculation.

[0116] Sub-step S144: Decode the target fused semantic representation to form the device operation monitoring result.

[0117] In this embodiment of the invention, after obtaining the target fusion semantic representation, the target fusion semantic representation can be decoded to form a device operation monitoring result. The decoding process may include: performing a fully connected mapping on the target fusion semantic representation to obtain a mapping vector, such as (e, f); then, performing a probability mapping on the mapping vector (e.g., using a softmax function) to obtain a probability distribution, such as (g, l), where g can represent the probability of an anomaly existing and l can represent the probability of no anomaly existing; then, the type corresponding to the larger of the two probabilities (existence of an anomaly or absence of an anomaly) can be determined as the device operation monitoring result.

[0118] In summary, the data analysis-based anaerobic tank mixing equipment operation monitoring method and system provided by this invention first acquires motor monitoring data and transmission monitoring data; second, it performs a first encoding to form a motor semantic representation corresponding to the motor monitoring data; then, it performs a second encoding to form a transmission semantic representation corresponding to the transmission monitoring data. The first encoding includes semantic encoding based on data segments formed by multi-time-length segmentation of the motor monitoring data, and / or, the second encoding includes repetitive semantic encoding based on the correlation between multiple data segments in the transmission monitoring data; finally, based on the motor semantic representation and the transmission semantic representation, it decodes to form the equipment operation monitoring result. Based on the above method, on the one hand, through encoding and decoding, the potential semantic features in the motor monitoring data and transmission monitoring data can be fully utilized, thereby improving the reliability of operation monitoring (compared to conventional static rule judgment or threshold comparison). On the other hand, since the first encoding includes semantic encoding of data segments formed by multi-time-length segmentation of motor monitoring data, the granularity of the encoding is more extensive, resulting in a comprehensive capture of potential semantic features at multiple granularities. This ensures the richness of the resulting motor semantic representation. Furthermore, since the second encoding includes repetitive semantic encoding based on the correlation between multiple data segments in the transmission monitoring data, and the positions of the stirring blades and transmission rods are repetitive, repetitive semantic encoding can capture strong characteristic semantic information in the transmission monitoring data, resulting in higher accuracy of the resulting transmission semantic representation. This further improves the reliability of the decoded equipment operation monitoring results. Based on this, the improved solution of this invention can address the problem of relatively low reliability in the operation monitoring of anaerobic tank stirring equipment in the prior art.

[0119] 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 operation of an anaerobic tank mixing device based on data analysis, characterized in that, include: Acquire motor monitoring data at the motor monitoring point and transmission monitoring data at the transmission structure monitoring point of the target anaerobic tank mixing equipment. The motor monitoring data is vibration data or sound data of the motor, and the transmission monitoring data is position data of the transmission structure. The transmission structure includes stirring blades and a transmission rod connecting the stirring blades and the motor. Perform the first encoding to form the motor semantic representation corresponding to the motor monitoring data; A second encoding is performed to form a transmission semantic representation corresponding to the transmission monitoring data. The first encoding includes semantic encoding based on data segments formed by multi-time-length segmentation of the motor monitoring data, and / or the second encoding includes repetitive semantic encoding based on the correlation between multiple data segments in the transmission monitoring data. Based on the motor semantic representation and the transmission semantic representation, the equipment operation monitoring result is decoded, wherein the equipment operation monitoring result is used to characterize whether there is any abnormality in the operation of the target anaerobic tank stirring equipment.

2. The method for monitoring operation of an anaerobic tank mixing device based on data analysis according to claim 1, characterized by, The step of performing a second encoding to form a transmission semantic representation corresponding to the transmission monitoring data includes: The positional semantics of each transmission monitoring data segment in the transmission monitoring data are mined to form a transmission positional semantic representation of each transmission monitoring data segment; The process involves mining the semantic relationships between the transmission position semantic representations of each transmission monitoring data segment to form corresponding position-related semantic representations, mining repetitive semantics of different time lengths from each position-related semantic representation to form repetitive local semantic representations corresponding to each time length, and determining position repetitive semantic representations based on the repetitive local semantic representations corresponding to each time length. Mining anomalous semantics in the transmission position semantic representation of each transmission monitoring data segment to form a position anomalous semantic representation, wherein, for the first step of mining and determining position repetitive semantic representation and the second step of mining and forming position anomalous semantic representation, at least the first step exists; Based on the location repeatability semantic representation, or based on the location repeatability semantic representation and the location anomaly semantic representation, the transmission semantic representation corresponding to the transmission monitoring data is determined.

3. The data analysis-based operation monitoring method for anaerobic tank mixing equipment as described in claim 2, characterized in that, The steps of mining the correlation semantics between the transmission position semantic representations of each transmission monitoring data segment to form corresponding position-related semantic representations, mining repetitive semantics of different time lengths from each position-related semantic representation to form repetitive local semantic representations corresponding to each time length, and determining the position repetitive semantic representation based on the repetitive local semantic representations corresponding to each time length include: In the transmission position semantic representation of each transmission monitoring data segment, determine the transmission position semantic representation of the first transmission monitoring data segment and the transmission position semantic representation of the second transmission monitoring data segment, and determine the relevant relationship parameters corresponding to the transmission position semantic representation of the first transmission monitoring data segment and the transmission position semantic representation of the second transmission monitoring data segment. The transmission position semantic representation of the first transmission monitoring data segment is transformed into rows and columns to form a position transformation semantic representation. Based on the position transformation semantic representation, the relevant relation parameters and the transmission position semantic representation of the second transmission monitoring data segment, the position-related semantic representation between the transmission position semantic representation of the first transmission monitoring data segment and the transmission position semantic representation of the second transmission monitoring data segment is determined. Repetitive semantics of different time lengths are extracted from the location-related semantic representations to form repetitive local semantic representations corresponding to each time length; Semantic aggregation processing is performed based on the repetitive local semantic representations corresponding to each time length to form the corresponding positional repetitive semantic representations.

4. The data analysis-based operation monitoring method for anaerobic tank mixing equipment as described in claim 3, characterized in that, The step of mining repetitive semantics of different time lengths from the location-related semantic representations to form repetitive local semantic representations corresponding to each time length includes: For each location-related semantic representation, a corresponding target window is determined based on the time length pre-configured for that location-related semantic representation, and based on the target window, the location-related semantic representation is semantically compressed to form a location-aware semantic representation of that time length. For each location-aware semantic representation, based on the correlation between the parameters in the location-aware semantic representation, a location mining semantic representation is mined from the location-aware semantic representation. The semantic representation of each location is mapped using a multilayer perceptron to form a repetitive local semantic representation corresponding to each time length.

5. The method for monitoring operation of an anaerobic tank mixing device based on data analysis according to claim 2, wherein, The step of mining anomalous semantics in the transmission position semantic representation of each transmission monitoring data segment to form a position anomaly semantic representation includes: For each transmission position semantic representation of the transmission monitoring data segment, the correlation semantic representation between the transmission position semantic representation and the mean semantic representation of all transmission position semantic representations centered on the transmission position semantic representation is determined. Based on the difference between the transmission position semantic representation and the associated semantic representation, the position difference semantic representation of the transmission position semantic representation is determined. Based on the semantic representations of the location differences, common semantics are mined to form a semantic representation of differences and common semantics. Based on the splicing result of the semantic representations of the location differences and the semantic representations of differences and common semantics, deep compression is performed to form a semantic representation of location anomalies.

6. The method for monitoring operation of an anaerobic tank mixing device based on data analysis according to claim 1, wherein, The step of performing the first encoding to form the motor semantic representation corresponding to the motor monitoring data includes: The motor monitoring data is divided into A time lengths to form a data segment sequence corresponding to each time length. The data segment sequence includes multiple motor monitoring data segments belonging to the corresponding time length. Each data segment sequence is convolved to form a motor semantic representation sequence corresponding to each data segment sequence, wherein the motor semantic representation sequence includes motor segment semantic representations corresponding to each motor monitoring data segment of the corresponding time length. For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, a multi-time length semantic set is determined based on the motor semantic representation sequence corresponding to the x-th time length. The multi-time length semantic set includes the semantic set corresponding to the y-th motor monitoring data segment from the x-th time length to the A-th time length. The semantic set includes the motor segment semantic representation corresponding to the motor monitoring data segments before the y-th motor monitoring data segment. x and y are both integers greater than or equal to 1. Attention fusion is performed on the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length to form a multi-time-length semantic representation. Based on the multi-time-length semantic representation corresponding to each motor monitoring data segment in each of the aforementioned time lengths, the motor semantic representation corresponding to the motor monitoring data is determined.

7. The data analysis-based operation monitoring method for anaerobic tank mixing equipment as described in claim 6, characterized in that, The step of determining a multi-time-length semantic set based on the motor semantic representation sequence corresponding to the x-th time length in the y-th motor monitoring data segment sequence corresponding to the x-th time length includes: Based on the number of motor monitoring data segments included in the data segment sequence corresponding to the xth time length, the target screening parameters corresponding to the xth time length are determined; For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, when y-1 is less than or equal to the target filtering parameter corresponding to the x-th time length, the semantic representation of the motor segment corresponding to the y-1 motor monitoring data segment is taken as a semantic set, wherein the y-1 motor monitoring data segment is all motor monitoring data segments in the data segment sequence before the y-th motor monitoring data segment; For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, when y-1 is greater than the target screening parameter corresponding to the x-th time length, the semantic representation of the motor segment corresponding to the nearest target screening parameter before the y-th motor monitoring data segment is taken as the semantic set; For the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length, the semantic set of the y-th motor monitoring data segment from the x-th time length to the A-th time length is determined as a multi-time length semantic set.

8. The method for monitoring operation of an anaerobic tank mixing device based on data analysis according to claim 6, wherein, The step of performing attention fusion on the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length to form a multi-time-length semantic representation includes: The splicing result of the semantic representations of each motor segment in the multi-time-length semantic set corresponding to the y-th motor monitoring data segment in the data segment sequence corresponding to the x-th time length is deeply compressed to form a motor compressed semantic representation. The deep compression includes convolution and attention processing. The result of concatenating the motor compressed semantic representation and the motor segment semantic representation of the y-th motor monitoring data segment is pooled and compressed to form the motor pooled semantic representation; Based on the motor pooling semantic representation mapping, a weighted adjustment parameter is formed, and based on the weighted adjustment parameter, the motor segment semantic representation of the y-th motor monitoring data segment is weighted and adjusted to form a multi-time-length semantic representation.

9. The method for monitoring operation of an anaerobic tank mixing apparatus based on data analysis according to any one of claims 1 to 8, characterized by, The step of decoding and forming equipment operation monitoring results based on the motor semantic representation and the transmission semantic representation includes: The average value of the motor semantic representation and the transmission semantic representation is calculated to achieve preliminary fusion and form an initial fused semantic representation. Random noise is injected into the initial fused semantic representation to form a noise fused semantic representation. Based on the motor semantic representation, the noise fusion semantic representation is denoised to form a motor denoised semantic representation, and based on the transmission semantic representation, the noise fusion semantic representation is denoised to form a transmission denoised semantic representation. The mean of the motor denoising semantic representation and the transmission denoising semantic representation is calculated to achieve target fusion and form a target fusion semantic representation. The target fusion semantic representation is decoded to form the device operation monitoring results.

10. An anaerobic tank agitator operation monitoring system based on data analysis, characterized by, 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 data analysis-based anaerobic tank stirring equipment operation monitoring method according to any one of claims 1-9.

Images