A method, device and electronic equipment for monitoring vibration of the entire system of a compressor unit

Abstract

The present invention discloses a method, device, electronic device, storage medium and application of vibration monitoring of the whole system of a compressor unit based on a mapping network, which relates to the field of compressor monitoring technology and solves technical problems such as incomplete existing monitoring, including: obtaining M measuring point data of the whole system of the compressor unit; performing feature extraction and data processing on the obtained M measuring point data; analyzing the correlation of the obtained M measuring point data, and determining the N measuring point feature data with the highest correlation as key measuring point data; training a data association model based on an attention mechanism; obtaining the N measuring point feature data at the current moment, and inputting the data association model after processing to obtain M data outputs. The present invention utilizes a few existing vibration measuring points and other monitoring parameters of the compressor unit to realize real-time online monitoring and evaluation of the vibration of the whole system of the compressor unit, ensure the safe operation of the unit, reduce workload, reduce costs, and has high promotion value and application value.

Description

Technical Field

[0001] The present invention relates to the field of compressor monitoring technology, and more specifically to a method, device, electronic equipment, storage medium and application thereof for monitoring the vibration of a compressor unit system based on a mapping network. Background Art

[0002] The compressor is a driven fluid machine that can boost low-pressure gas to high-pressure gas. It is the heart of the refrigeration system. It sucks low-temperature and low-pressure refrigerant gas from the suction pipe, compresses it by driving the piston through the motor, and discharges high-temperature and high-pressure refrigerant gas to the exhaust pipe, providing power for the refrigeration cycle, thereby realizing the refrigeration cycle of compression → condensation (heat release) → expansion → evaporation (heat absorption).

[0003] The vibration index standards at different positions of the compressor unit are different. Taking the API618 standard as an example, the standard value of the vibration of the compressor unit lever head shall not exceed 18mm / s, and the buffer tank pipeline shall not exceed 32mm / s. In the prior art, there are many real-time vibration monitoring works for some parts of the compressor unit body, such as the lever head, cylinder head, crankcase, etc. However, at present, important parts such as buffer tanks and process pipelines cannot be monitored in real time. These parts are usually intermittently monitored through regular inspections, and it is impossible to evaluate the vibration of the entire compressor system in real time from the monitoring values. If the entire system needs to be monitored, sensors need to be arranged in various parts of the compressor, and complete and complex wiring work and remote online system design need to be carried out. This is not only time-consuming, labor-intensive and costly, but also the sensors and related lines for monitoring in these parts are not conducive to the normal operation of the compressor. Summary of the invention

[0004] The purpose of the present invention is to solve the above technical problems, and provide a method for monitoring the vibration of the whole system of a compressor unit based on a mapping network. The present invention uses a grey correlation analysis method to analyze the correlation between the data of each measuring point, and on this basis uses a seq2seq model based on an attention mechanism to learn and train a mapping model from a small number of measuring points to a large number of measuring points, so that the whole system vibration monitoring and evaluation can be completed through a small number of measuring point data, which can not only improve the efficiency of data analysis, but also monitor important parts that cannot be monitored in real time through the model.

[0005] Compared with other methods, the present invention only needs a small amount of key measuring point data after model learning and training, and utilizes the mapping relationship between the key measuring points and all measuring points to complete the online real-time vibration monitoring and evaluation of the entire compressor unit system. There is no need to install sensors on the entire compressor unit system, which greatly reduces the cost and monitoring workload, and meets the requirements of the compressor being able to operate normally while monitoring the vibration of important parts such as buffer tanks and process pipelines.

[0006] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions:

[0007] In a first aspect, the present invention discloses a method for monitoring vibration of a compressor unit system based on a mapping network, comprising:

[0008] Obtain M measuring point data of the entire compressor system;

[0009] Perform feature extraction and data processing on the acquired M measuring point data;

[0010] Analyze the correlation of the M measuring point data obtained, and arrange them from high to low according to the correlation, and successively determine N measuring point feature data starting from the highest, wherein the N measuring point feature data are key measuring point data;

[0011] The data association model based on the attention mechanism is trained with key measurement point data as input and M data as output;

[0012] Obtain the feature data of N measuring points at the current moment, extract the features and process the data, then input them into the data association model to obtain M data outputs.

[0013] Furthermore, the types of the M measuring point data include working condition data, pressure pulsation data, measuring point vibration acceleration values ​​and angular velocity values;

[0014] The feature extraction and data processing of the acquired M measuring point data includes the following contents:

[0015] (1) Extract the root mean square eigenvalue of the vibration acceleration value and angular velocity value of the M measuring point data of the whole compressor system:

[0016]

[0017] (2) The operating condition data, pressure pulsation data and root mean square characteristic value data are subjected to the non-quantitative tempering process using the following formula:

[0018]

[0019] Where:

[0020] X e Indicates working condition data / pressure pulsation data / root mean square value data;

[0021] μ is X e Data mean;

[0022] σ is X e The variance of

[0023] is the dimensionless processed data, X eThe data obtained after dimensionless processing There are G parameters in total.

[0024] Furthermore, the analysis of the correlation of the M measurement point data includes: using the grey correlation analysis method to analyze the dimensionless processed data That is, the correlation between G parameters is arranged from high to low according to the correlation, and N measurement point data with high correlation are continuously determined starting from the highest.

[0025] Furthermore, the relevance of the data is determined in the following manner:

[0026] Calculate the grey correlation coefficient ξ between the input data at time k ij (k):

[0027]

[0028] In the formula,

[0029] is the second-level minimum difference;

[0030] is the second-level maximum difference;

[0031] ρ is the resolution coefficient, and its value range is (0, 1);

[0032]

[0033] and The grey relational degree r ij for:

[0034]

[0035] The grey relational degree r ij Sort by size, and select N measuring point feature data with high correlation between the data corresponding to the G parameters in order from large to small.

[0036] Furthermore, the data association model is a seq2seq model.

[0037] Furthermore, the seq2seq model includes:

[0038] Input layer: Input the working condition data after the tempering treatment, pressure pulsation data and root mean square eigenvalue data, a total of F types of data, the selected data length is K, and the input layer dimension is K×F;

[0039] Encoding layer: The encoding layer uses a 2-layer RNN based on the attention mechanism, with 128 LSTM neural units in each layer; given an input sequence x = (x1, x2, ... x f,…x F ), where x f (f∈{1,2,…,F}) is a column vector of length K The encoding formula input to the hidden layer is: f =b(h f-1 ,x f ), where b is the activation function of the LSTM network, h f represents the vector of the hidden layer when f; RNN reads all the sequences in sequence and obtains the vector C by weighted summation of the hidden layer vectors based on the attention mechanism f , vector C f The calculation method is as follows:

[0040]

[0041]

[0042]

[0043] Among them, V a , W, U are the parameters to be learned; h f-1 、s f-1 is the hidden layer state and neuron state of the previous step; is the weight of the encoder output vector at time f;

[0044] Decoding layer: the final output vector C of the encoding layer f As the input of the decoding layer, the decoding layer uses a 2-layer RNN, with 128 LSTM neural units in each layer;

[0045] Output layer: The final output state of the decoding layer as the output layer is the following formula:

[0046] y m =g(y m-1 ,s m ,C f ) (8)

[0047] Among them, s m =f(s m-1 ,y m-1 ,C f ), g and f are the activation functions of LSTM, y m-1 is the output of the previous moment m-1.

[0048] In a second aspect, the present invention discloses a compressor unit full system vibration monitoring device based on a mapping network, comprising:

[0049] Data acquisition module: a module used to acquire characteristic data of N measuring points which are key measuring point data;

[0050] Extraction and processing module: extract features and process data from the acquired data;

[0051] Model acquisition module: used to obtain a good data association model based on the attention mechanism for training;

[0052] Data input module: used to input the measurement point data processed by the extraction and processing module into the data association model;

[0053] Data output module: used to display and store the output of the data association model.

[0054] In a third aspect, the present invention discloses an electronic device, which includes a processor and a storage device for storing programs. When the program is executed by the processor, the processor implements any of the monitoring methods described above.

[0055] In a fourth aspect, the present invention discloses a computer-readable storage medium having a computer program stored thereon, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the monitoring method described in any one of the above is implemented.

[0056] In a fifth aspect, the present invention discloses an application of the method described in any one of the above items in vibration monitoring.

[0057] The beneficial effects of the present invention are as follows:

[0058] 1. The present invention requires a small amount of data in actual monitoring and the data is accurate, and can obtain monitoring of the entire compressor unit system based on a small amount of input data. Compared with other methods, the present invention only requires a small amount of key measuring point data, and uses the mapping relationship between key measuring points and all measuring points to complete the online real-time vibration monitoring and evaluation of the entire compressor unit system. There is no need to install sensors on the entire compressor unit system, which greatly reduces the monitoring cost and workload, and effectively avoids the maintenance trouble and high cost of electronic devices used for monitoring after failure.

[0059] 2. In the early stage or in the laboratory, data is collected for the entire system of the compressor unit. The data of important parts such as existing buffer tanks and process pipelines that cannot be monitored in real time are collected together with the data of other existing parts, and data correlation analysis is performed to train a data correlation model based on the attention mechanism; then the obtained model is directly put into the production site to monitor the entire system of the compressor unit, and then there is no need to arrange sensors in important parts such as buffer tanks and process pipelines of the entire compressor unit system that has been put into use, and the monitoring data of these parts can be obtained. This not only improves the comprehensive monitoring of the entire compressor unit system, is not limited to structures such as buffer tanks and process pipelines on the compressor unit where sensors cannot be arranged, but also can obtain data from all measuring points through fewer key measuring points, which saves time and effectively controls costs, and will not affect the normal operation of the compressor. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the process of the present invention;

[0061] Figure 2 This is a schematic diagram of the seq2seq model based on the attention mechanism;

[0062] Figure 3 It is a schematic diagram of the structure of the device. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0064] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] Example 1

[0066] like Figure 1 As shown, this embodiment provides a method for monitoring vibration of the entire compressor system based on a mapping network, including:

[0067] Obtain M measuring point data of the entire compressor system;

[0068] Perform feature extraction and data processing on the acquired M measuring point data;

[0069] Analyze the correlation of the M measuring point data obtained, and arrange them from high to low according to the correlation, and successively determine N measuring point feature data starting from the highest, wherein the N measuring point feature data are key measuring point data;

[0070] The data association model based on the attention mechanism is trained with key measurement point data as input and M data as output;

[0071] Obtain the feature data of N measuring points at the current moment, extract the features and process the data, then input them into the data association model to obtain M data outputs.

[0072] The present invention is based on analyzing the correlation between the data that need to be monitored for the entire compressor unit system, analyzing the correlation between the data of each measuring point, and on this basis using a model based on the attention mechanism to learn and train to complete the mapping from a few measuring points to a large number of measuring points, so that the vibration monitoring and evaluation of the entire system can be completed through the data of a few measuring points.

[0073] The present invention requires a small amount of data in actual monitoring and the data is accurate, and can obtain monitoring of the entire compressor unit system based on a small amount of input data. Compared with other methods, the present invention only requires a small amount of key measuring point data, and utilizes the mapping relationship between key measuring points and all measuring points to complete the online real-time vibration monitoring and evaluation of the entire compressor unit system. There is no need to install sensors on the entire compressor unit system, which greatly reduces the monitoring cost and monitoring workload, and effectively avoids the maintenance trouble and high cost of electronic devices used for monitoring after failure.

[0074] At the same time, during implementation, data collection can be carried out on the entire system of the compressor unit in the early stage or in the laboratory, and the data of important parts such as existing buffer tanks and process pipelines that cannot be monitored in real time can be collected together with the data of other existing parts, and data correlation analysis can be carried out to train a data correlation model based on the attention mechanism; then the obtained model can be directly put into the production site to monitor the entire system of the compressor unit, and then there is no need to arrange sensors on important parts such as buffer tanks and process pipelines of the entire compressor unit system that has been put into use, so as to obtain the monitoring data of these parts. This not only improves the comprehensive monitoring of the entire compressor unit system, is not limited to structures such as buffer tanks and process pipelines on the compressor unit where sensors cannot be arranged, but also can obtain data of all measuring points through fewer key measuring points, which saves time and effectively controls costs, and will not affect the normal operation of the compressor.

[0075] Example 2

[0076] This embodiment further illustrates the present invention based on the embodiment 1.

[0077] like Figure 1As shown, this embodiment provides a method for monitoring vibration of the entire compressor system based on a mapping network, including:

[0078] S1. Acquire M measuring point data of the entire compressor system, wherein the types of the M measuring point data include operating condition data, pressure pulsation data, measuring point vibration acceleration values ​​and angular velocity values;

[0079] S2. Feature extraction and data processing are performed on the acquired M measuring point data. The specific contents are as follows:

[0080] (1) Extract the root mean square eigenvalue of the vibration acceleration value and angular velocity value of the M measuring point data of the whole compressor system:

[0081]

[0082] Here, the vibration acceleration value and the angular velocity value at each point are measured by a sensor, and based on spectrum analysis, m vibration acceleration values ​​and m angular velocity values ​​at this point are obtained.

[0083] (2) The operating condition data, pressure pulsation data and root mean square characteristic value data are subjected to the non-quantitative tempering process using the following formula:

[0084]

[0085] Where:

[0086] X e Indicates working condition data / pressure pulsation data / root mean square value data;

[0087] μ is X e Data mean;

[0088] σ is X e The variance of

[0089] is the dimensionless processed data, X e The data obtained after dimensionless processing There are G parameters in total.

[0090] S3. Analyze the correlation of the M measurement point data obtained. The specific analysis content is:

[0091] The dimensionless data were analyzed using grey correlation analysis. That is, the correlation between G parameters is arranged from high to low according to the correlation, and N measurement point data with high correlation are continuously determined starting from the highest;

[0092] The relevance of the data is determined in the following manner:

[0093] Calculate the grey correlation coefficient ξ between the input data at time k ij (k):

[0094]

[0095] In the formula,

[0096] is the second-level minimum difference;

[0097] is the second-level maximum difference;

[0098] ρ is the resolution coefficient, and its value range is (0, 1);

[0099]

[0100] and The grey relational degree r ij for:

[0101]

[0102] The grey relational degree r ij Sort by size, and select N measuring point feature data with high correlation between data corresponding to G parameters in order from large to small, that is, arrange from high to low according to the correlation, and continuously determine N measuring point feature data from the highest, the N measuring point feature data are key measuring point data, the N measuring point feature data belong to M measuring point data, and the N value is much smaller than the M value;

[0103] S4, using the key measurement point data as input and M data as output to train a data association model based on the attention mechanism; the data association model is a seq2seq model, which is specifically as follows Figure 2 shown.

[0104] Specifically, the seq2seq model is Figure 2 As shown, it includes:

[0105] Input layer: Input the working condition data, pressure pulsation data and RMS eigenvalue data after the infinite tempering treatment, a total of F types of data, that is, the maximum dimension of the input layer is F, the selected data length is K, and the input layer dimension is K×F;

[0106] Encoding layer: The encoding layer uses a 2-layer RNN based on the attention mechanism, with 128 LSTM neural units in each layer; given an input sequence x = (x1, x2, ... x f ,…x F ), where x f (f∈{1,2,…,F}) is a column vector of length K The encoding formula input to the hidden layer is: f =b(h f-1 ,x f ), where b is the activation function of the LSTM network, h f represents the vector of the hidden layer when f; RNN reads all the sequences in sequence and obtains the vector C by weighted summation of the hidden layer vectors based on the attention mechanism f , vector C f The calculation method is as follows:

[0107]

[0108]

[0109]

[0110] Among them, V a , W, U are the parameters to be learned; h f-1 、s f-1 is the hidden layer state and neuron state of the previous step; is the weight of the encoder output vector at time f;

[0111] Decoding layer: the final output vector C of the encoding layer f As the input of the decoding layer, the decoding layer uses a 2-layer RNN, with 128 LSTM neural units in each layer;

[0112] Output layer: The final output state of the decoding layer as the output layer is the following formula:

[0113] y n =g(y m-1 ,s m ,C f ) (8)

[0114] Among them, s m =f(s m-1 ,y m-1 ,C f ), g and f are the activation functions of LSTM, y m-1 is the output of the previous moment m-1; the final output is the output y of the previous moment m-1 m-1 , hidden state neuron s at time m m And the vector C calculated by the attention mechanism f It is determined by three factors; the dimension of the output layer is M×P, where M is the number of data types of all measurement points and P is the selected data length.

[0115] Compressor vibration monitoring model training: The collected data is divided into training set and test set in proportion, for example, divided into training set and test set in a ratio of 7:3. In the model training stage, the N measurement point feature data, operating condition data, and pressure pulsation data selected based on correlation in the training set are used as model input, and the M measurement point data of the compressor unit system are used as model output. This is used as a training sample to train the model, and the loss function is L2. Finally, the compressor vibration monitoring model is obtained, and then verified with the test set data.

[0116] S5. When monitoring the entire compressor system to be monitored, obtain N measuring point feature data of the entire compressor system at the current moment, extract features and perform data processing on them, and then input them into the data association model to obtain M data outputs. The positions of the N measuring point feature data are consistent with the positions of the N measuring point feature data determined in step S3.

[0117] In analyzing the compressor vibration monitoring data, the present invention analyzes the correlation between the monitoring data based on the grey correlation analysis method, selects a few key measuring points, and then uses the seq2seq model based on the attention mechanism to complete the mapping of a few measuring points to all measuring points, and finally realizes the online monitoring of the whole system vibration of the compressor unit. It not only solves the problem that the inspection cannot evaluate the vibration of the compressor unit in real time, but also greatly reduces the workload and cost of comprehensive monitoring of the vibration of the compressor unit, and has high promotion value and application value.

[0118] The present invention utilizes a few existing vibration measuring points and other monitoring parameters of the compressor unit to achieve real-time online monitoring and evaluation of the vibration of the entire compressor unit system, thereby ensuring the safe operation of the unit, reducing workload and lowering costs.

[0119] In the above, the feature extraction and data processing part can also adopt or add other time-frequency domain feature extraction methods, such as effective value, kurtosis, peak-to-peak value, etc. The correlation analysis between data can also be changed to other distance measurement-based methods, such as density clustering method, dynamic time warping and other distance measurement-based methods. A mapping model from a few measuring points to all measuring points can also be established, which can also be based on neural networks or other deep learning models. That is, the feature extraction and data processing methods of the data of the present invention are not limited, the data association analysis method is not limited, and the model is not uniquely limited.

[0120] Example 3

[0121] This embodiment provides a compressor unit full system vibration monitoring device based on a mapping network, such as Figure 3As shown, it includes: a data acquisition module: a module for acquiring feature data of N measuring points that are key measuring point data; an extraction and processing module: performing feature extraction and data processing on the acquired data; a model acquisition module: used to acquire a good data association model based on the attention mechanism for training; a data input module: used to input the measuring point data processed by the extraction and processing module into the data association model; a data output module: used to display and store the output of the data association model.

[0122] The compressor unit system-wide vibration monitoring device provided by the present invention requires a small amount of data and accurate data during actual monitoring, and can obtain monitoring of the entire compressor unit system based on a small amount of input data. Compared with other methods, the present invention only requires a small amount of key measuring point data, and utilizes the mapping relationship between the key measuring points and all measuring points to complete the online real-time vibration monitoring and evaluation of the compressor unit system-wide. There is no need to install sensors on the compressor unit system-wide, which greatly reduces the monitoring cost and monitoring workload, and effectively avoids the maintenance troubles and high costs of electronic devices used for monitoring after failure.

[0123] Example 4

[0124] This embodiment provides an electronic device, the electronic device includes a processor and a storage device for storing a program, when the program is executed by the processor, the processor implements any of the monitoring methods described above. The processor may be one or more, and the storage device may also be one or more. The memory and the processor are preferably connected in a CAN bus manner.

[0125] Example 5

[0126] A computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the monitoring method described in any one of the above is implemented.

[0127] Example 6

[0128] The application of the method described above in vibration monitoring is specifically to collect data of all monitoring points required for the system / equipment / device to be monitored, then extract features and dimensionless process the data of the monitoring points, then perform correlation analysis on the data obtained by dimensionless processing, and then sort the analysis results from high to low correlation, and continuously determine a certain number of highly correlated measurement point data starting from the highest correlation as measurement point feature data, and then use all the measurement point feature data as input data and the data of the monitoring points on the system / equipment / device as input for model training to obtain a data association model. Then, the monitoring point where the measurement point feature data is located is the monitoring point on the system / equipment / device, and then these points are used as input data and input into the data association model, so as to obtain data of all monitoring points required for such system / equipment / device, and then based on this method, the monitoring of systems / equipment / devices of the same model / type can be satisfied, thereby reducing the monitoring cost of the system / equipment / device.

[0129] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for monitoring the vibration of the entire compressor system based on a mapping network, characterized in that: include: Obtain M measuring point data of the entire compressor system; Perform feature extraction and data processing on the acquired M measuring point data; Analyze the correlation of the M measuring point data obtained, and arrange them from high to low according to the correlation, and successively determine N measuring point feature data starting from the highest, wherein the N measuring point feature data are key measuring point data; The data association model based on the attention mechanism is trained with key measurement point data as input and M data as output; Obtain feature data of N measuring points at the current moment, extract features and process the data, then input into the data association model to obtain M data outputs; The types of the M measuring point data include working condition data, pressure pulsation data, measuring point vibration acceleration values ​​and angular velocity values; The feature extraction and data processing of the acquired M measuring point data includes the following contents: (1) Extract the root mean square eigenvalue of the vibration acceleration value a and angular velocity value ω of the M measurement point data of the whole compressor system: (2) The operating condition data, pressure pulsation data and root mean square eigenvalue data are dimensionlessly processed using the following formula: Where: Xe represents operating condition data / pressure pulsation data / root mean square value data; μ is the mean value of Xe data; σ is the variance of Xe; X e (0) Xe is the data after dimensionless processing. e (0) There are G parameters in total; The analysis of the correlation of the M measurement point data includes: using the grey correlation analysis method to analyze the dimensionless processed data X e (0) , that is, the correlation between G parameters, arranged from high to low according to the correlation, and N measurement point data with high correlation are continuously determined starting from the highest; The relevance of the data is determined in the following manner: Calculate the grey correlation coefficient ξ between the input data at time k ij (k): In the formula, is the second-level minimum difference; is the second-level maximum difference; ρ is the resolution coefficient, and its value range is (0, 1); x i (0) and x j (0) The grey relational degree r ij for: The grey relational degree r ij Sort by size, and select N measuring point feature data with high correlation between the data corresponding to the G parameters in order from large to small.

2. A method for monitoring the vibration of the entire compressor system based on a mapping network according to claim 1, characterized in that: The data association model is a seq2seq model.

3. A method for monitoring the vibration of the entire compressor system based on a mapping network according to claim 2, characterized in that: The seq2seq model includes: Input layer: Input the dimensionless processed working condition data, pressure pulsation data and RMS eigenvalue data, a total of F types of data, the selected data length is K, and the input layer dimension is K×F; Encoding layer: The encoding layer uses a 2-layer RNN based on the attention mechanism, with 128 LSTM neural units in each layer; Given an input sequence x = (x1, x2, ...x f , …x F ), where x f (f∈{1, 2, ..., F}) is a column vector of length K The encoding formula input to the hidden layer is: f =b(h f-1 , x f ), where b is the activation function of the LSTM network, h f represents the vector of the hidden layer when f; RNN reads all the sequences in sequence and obtains the vector C by weighted summation of the hidden layer vectors based on the attention mechanism f , vector C f The calculation method is as follows: Among them, V a , W, U are the parameters to be learned; h f-1 、s f-1 is the hidden layer state and neuron state of the previous step; is the weight of the encoder output vector at time f; Decoding layer: the final output vector C of the encoding layer f As the input of the decoding layer, the decoding layer uses a 2-layer RNN, with 128 LSTM neural units in each layer; Output layer: The final output state of the decoding layer as the output layer is the following formula: y m =g(y m-1 ,s m ,C f ) (8) Among them, s m =f(s m-1 ,y m-1 ,C f ), g and f are the activation functions of LSTM, y m-1 is the output of the previous moment m-1.

4. A vibration monitoring device for the compressor unit full system vibration monitoring method based on mapping network as described in any one of claims 1 to 3, characterized in that: include: Data acquisition module: a module used to acquire characteristic data of N measuring points which are key measuring point data; Extraction and processing module: extract features and process data from the acquired data; Model acquisition module: used to obtain a good data association model based on the attention mechanism for training; Data input module: used to input the measurement point data processed by the extraction and processing module into the data association model; Data output module: used to display and store the output of the data association model.

5. An electronic device, characterized in that: The electronic device comprises a processor and a storage device for storing a program. When the program is executed by the processor, the processor implements the monitoring method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, the computer program comprising program instructions, characterized in that: When the program instructions are executed by a processor, the monitoring method according to any one of claims 1 to 3 is implemented.

Images