CN114037026B - Air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence - Google Patents
Air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence Download PDFInfo
- Publication number
- CN114037026B CN114037026B CN202210024089.XA CN202210024089A CN114037026B CN 114037026 B CN114037026 B CN 114037026B CN 202210024089 A CN202210024089 A CN 202210024089A CN 114037026 B CN114037026 B CN 114037026B
- Authority
- CN
- China
- Prior art keywords
- air compressor
- opening angle
- air
- starting
- similarity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
Abstract
The invention relates to the technical field of artificial intelligence and air compressors, in particular to an air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence. The method comprises the following steps: acquiring the joint similarity of the starting processes of any two air compressors, and clustering the air compressors to obtain a plurality of clusters; combining clusters of which the opening angle difference value of the anti-surge valve is smaller than a preset opening angle difference value threshold; pre-training the twin network to obtain a feature descriptor of each air compressor; continuing to train the twin network, and determining the maximum likelihood estimation opening angle of the air compressor to be self-checked according to the similarity of the characteristic descriptor of the air compressor to be self-checked and the characteristic descriptors of other air compressors; and adjusting the opening angle of the anti-surge valve, and trying to start the air compressor under pressure. According to the method, the opening range and the extreme value of the anti-surge valve for successfully starting the air compressor under pressure can be accurately determined according to the maximum likelihood estimation opening angle while industrial big data of users are protected, the air compressor can be safely started under pressure, and the average failure-free time of the air compressor is prolonged.
Description
Technical Field
The invention relates to the technical field of artificial intelligence and air compressors, in particular to an air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence.
Background
The air compressor is widely applied to the industrial fields of petroleum, metallurgy, chemical industry and the like, and along with the rapid development of scientific technology, the air compressor unit tends to be large-sized and complex more and more. Because the operation condition of the air compressor unit is more and more complex, the balance pressure of the air compressor unit is no longer a single condition when the air compressor unit is started, and more rigorous requirements are provided for the driving equipment and the starting scheme of the air compressor unit. Taking a certain large ethane refrigeration air compressor driven by a motor as an example, because the process gas is actually at the dew point temperature and is influenced by the environmental temperature factor, the balance pressure of the air compressor unit can reach a very high degree after the air compressor unit is stopped. At this time, if the air compressor is directly started, namely, the air compressor unit is started with high pressure, the motor needs large starting power to drive the air compressor to the rated rotating speed. Once the starting load exceeds the maximum starting power which can be borne by the motor, the air compressor unit stops due to the motor current overload protection, the driving progress is influenced, and the subsequent process and the motor are greatly damaged.
In order to reduce the pressure starting load in the pressure starting process, the method which is effective and does not need extra cost is to reduce the starting power by properly closing an anti-surge valve so as to achieve the aim of safe high-pressure starting of the air compressor. The pressurized starting process of the air compressor is a common working condition, and the air compressor can be started smoothly by continuously determining the opening value of the surge valve in some processes because the working condition of the pressurized starting of the air compressor at each time is different. In a large-scale industrial process, if the pressurized starting of the air compressor during each driving needs the participation of professional maintenance personnel, and the anti-surge valve is continuously finely adjusted to realize smooth starting, the efficiency can be greatly reduced, and the manual arrival time and the debugging time can cause various economic losses to modern production.
The industrial big data face the privacy problem, and the information of the industrial big data which can be collected in engineering is rich and the quantity is large. Therefore, an air compressor self-checking system based on industrial big data is needed, and the air compressor self-checking system and the air compressor working condition self-checking system analyze the characteristics of the air compressor during starting under pressure, so that a proper opening interval of a surge valve is obtained based on a large amount of operation data, and the air compressor is automatically tried in a safer opening interval of the surge valve. Thereby improving the mean time between failures of the air compressor.
Disclosure of Invention
In order to solve the technical problems, the invention aims to provide an air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence, and the adopted technical scheme is as follows:
one embodiment of the invention provides an air compressor under-pressure starting self-checking method based on industrial big data and artificial intelligence, which comprises the following steps: acquiring an overload power fluctuation characteristic sequence and an opening angle of an anti-surge valve in the starting process of the air compressor which is successfully started under pressure; acquiring the joint similarity of the starting processes of any two air compressors according to the correlation degree and the opening angle difference of the overload power fluctuation characteristic sequences of the starting processes of the two air compressors; clustering air compressors according to the joint similarity of any two air compressors to obtain a plurality of clusters;
combining clusters of which the opening angle difference value of the anti-surge valve is smaller than a preset opening angle difference value threshold;
pre-training the twin network: the input of the twin network is an overload power fluctuation characteristic sequence and an opening angle of any two air compressors in the starting process, the input is labeled according to whether the two air compressors belong to the same cluster, and the output of the twin network is a characteristic descriptor of each air compressor;
continuing to train the twin network: if the input of the twin network is from the same cluster, the label value is a first set value, otherwise, the label value is the joint similarity of the two air compressors;
determining the maximum likelihood estimation opening angle of the air compressor to be self-checked according to the similarity of the characteristic descriptor of the air compressor to be self-checked and the characteristic descriptors of other air compressors;
and adjusting the opening angle of the anti-surge valve according to the maximum likelihood estimation opening angle, and trying to start the air compressor under pressure.
Preferably, the overload power fluctuation characteristic sequence in the starting process of the air compressor for obtaining the successful starting under pressure comprises the following steps: determining the acquisition duration of real-time load power in the process of starting the air compressor under pressure; when the starting load power of the air compressor exceeds a nominal value, starting to collect real-time load power, and ending the collection time; and taking the ratio of the real-time load power to the power nominal value as an overload power fluctuation characteristic data point in the acquisition time length, and forming an overload power fluctuation characteristic sequence by all the overload power fluctuation characteristic data points in the acquisition time length.
Preferably, the opening angle of the anti-surge valve in the starting process of the air compressor which is successfully started under pressure is a proportional value, and the larger the value is, the larger the opening angle of the anti-surge valve is.
Preferably, the obtaining of the joint similarity of the starting processes of any two air compressors according to the correlation degree and the opening angle difference of the overload power fluctuation characteristic sequences of the starting processes of the two air compressors includes: and obtaining the difference value between the normalized cross correlation coefficient of the overload power fluctuation characteristic sequences of the two air compressors and a set value and the absolute value of the difference value of the opening angles of the two air compressors, wherein the product of the two values is the joint similarity of the starting processes of the two air compressors.
Preferably, the clustering the air compressors according to the joint similarity of any two air compressors to obtain a plurality of clusters comprises: determining the cluster number of clusters to be clustered according to the number of the air compressors as a sample; acquiring the joint similarity of any two air compressors as the distance between samples; and clustering the air compressor according to the number of the clusters and the distance between the samples to obtain a plurality of clusters.
Preferably, determining the maximum likelihood estimation opening angle of the air compressor to be self-tested according to the similarity between the air compressor to be self-tested feature descriptor and the other air compressor feature descriptors comprises: obtaining the similarity between the characteristic descriptor of the air compressor to be self-tested and other characteristic descriptors of the air compressor; selecting the similarity larger than a preset similarity threshold value and the corresponding opening angles of other air compressors; and taking the selected similarity as the weight corresponding to the opening angles of other air compressors, and obtaining a weighted average value as the maximum likelihood estimation opening angle of the air compressor to be self-checked.
Preferably, the anti-surge valve opening angle is adjusted according to the maximum likelihood estimation opening angle, and the attempt of starting the air compressor under pressure comprises the following steps: forming a valve opening angle adjustment conservative interval according to the maximum likelihood estimation opening angle and the maximum opening angle of the anti-surge valve of the air compressor; and adjusting the opening angle of the valve in the conservative interval, if the opening angle of the valve is the maximum likelihood estimation opening angle, the starting of the air compressor is still unsuccessful, and stopping trying to protect the air compressor.
Preferably, the opening angle of the anti-surge valve is adjusted according to the maximum likelihood estimation opening angle, and the attempting of the pressurized start of the air compressor further comprises: if the attempt of starting the air compressor under pressure is successful in the conservative interval, updating a preset similarity threshold; a more accurate maximum likelihood estimate opening angle is obtained.
Preferably, the preset similarity threshold is 0.7.
Preferably, the preset similarity threshold t:
t=Min(α+β*D*∑(dmin,k)/[(d+r)*(K-1)],1)
wherein alpha and beta are constant parameters, D is the density of the cluster to which the air compressor belongs to be self-tested, and Dmin,kThe minimum distance between the air compressor to be self-inspected and other clusters K is defined, K is the number of clusters, d is the distance between the air compressor to be self-inspected and the center of the cluster to which the air compressor belongs, and r is the minimum circumscribed circle radius of the minimum convex polygon corresponding to all data points during clustering.
The embodiment of the invention at least has the following beneficial effects: the method comprises the steps of obtaining a normalized cross correlation coefficient of an overload power fluctuation sequence of the air compressors, obtaining combined similarity of the two air compressors according to the cross correlation coefficient and an anti-surge valve opening angle, clustering the air compressors according to the combined similarity to obtain a plurality of clusters, pre-training a twin network, combining the clusters with the anti-surge valve opening angle difference value smaller than a preset opening angle difference value threshold value, continuously training the twin network, obtaining the similarity of a feature descriptor of the air compressor to be self-detected and feature descriptors of other air compressors so as to obtain a maximum likelihood estimation opening angle, trying the under-pressure starting of the air compressor, accurately determining an anti-surge valve opening interval and an extreme value of the air compressor which is successfully started under pressure according to the maximum likelihood estimation opening angle while protecting industrial big data of users, ensuring that the air compressor can be started under pressure safely, and improving the average failure-free time of the air compressors. The invention adopts pre-training and continuous training for the twin network, and different labels are respectively used for two times of training, thereby improving the training efficiency compared with one time of training and leading the network to be more quickly converged. Compared with direct clustering, the method has the advantages that the twin network is adopted to obtain the characteristic descriptors of the starting process of the air compressors, so that the data privacy of air compressor users in a big data environment can be effectively protected, the obtained characteristic descriptors can represent the tiny difference among the starting processes of the air compressors, more accurate similarity comparison results can be obtained, and the precision of the maximum likelihood estimation opening angle of the anti-surge valve is further improved; and a dynamic preset similarity threshold is designed by combining with the scene of the air compressor, and the preset similarity threshold is updated according to whether the starting is successful, so that the estimation precision of the opening angle of the anti-surge valve is further improved, and the risk of the pressurized starting process is reduced.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of the method of the present invention.
Detailed Description
In order to further explain the technical means and effects of the present invention adopted to achieve the predetermined invention purpose, the following detailed description, with reference to the accompanying drawings and preferred embodiments, describes specific implementation, structure, features and effects of a self-checking method for under-pressure start of an air compressor based on industrial big data according to the present invention. In the following description, different "one embodiment" or "another embodiment" refers to not necessarily the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The following specifically describes a specific scheme of the air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence, which is provided by the invention, with reference to the accompanying drawings.
Example 1
The main application scenarios of the invention are as follows: the method is applied to the pressurized starting process of the air compressor, the required starting power in the pressurized starting process is very large and may exceed the maximum starting power which can be borne by the motor, and the air compressor can be stopped due to the overload protection of the current of the motor. At the moment, the opening angle of the anti-surge valve is adjusted, the starting power of the air compressor during starting under pressure is reduced, and the compressor can be started smoothly.
Referring to fig. 1, a flowchart of a method for self-checking an air compressor under pressure starting based on industrial big data and artificial intelligence according to an embodiment of the present invention is shown, where the method includes the following steps:
firstly, acquiring an overload power fluctuation characteristic sequence and an opening angle of an anti-surge valve in the starting process of an air compressor which is successfully started under pressure; acquiring the joint similarity of the starting processes of any two air compressors according to the correlation degree and the opening angle difference of the overload power fluctuation characteristic sequences of the starting processes of the two air compressors; and clustering the air compressors according to the joint similarity of any two air compressors to obtain a plurality of clusters.
Whether the air compressor is in the pressurized starting process or not can be judged by judging the pipeline pressure of the air compressor, and when the air compressor receives a starting instruction and is determined to be in the pressurized starting process through the pipeline pressure, the real-time load power of the air compressor is recorded. The acquisition process in this embodiment takes a certain 3200KW air compressor as an example, acquires the real-time load power in the air compressor starting process of successful starting with pressure for describing the load characteristics when the air compressor is successfully started with pressure to drive the fluid, the load power is easy to fluctuate in the successful starting process with pressure of some air compressors, and the load power is relatively stable in the successful starting process with pressure of some air compressors.
The load characteristic of the air compressor is an important index, and the load change of the air compressor during the starting under pressure mainly depends on the fluid type, the process of the air compressor and other factors, so that the starting under pressure loads of different fluids and different air compressor processes can be abstracted by the real-time load power of the air compressor which is successfully started under pressure during the starting under pressure to form a sample cluster, and further other information under the load characteristic is described in a unified manner.
And setting the acquisition time length of the load power implemented in the process of starting the air compressor under pressure, wherein the acquisition time length is preferably set to 60s in the embodiment. In the process of successful starting of the air compressor under pressure, when the power reading of the motor exceeds the power nominal value PtypicalThen, real-time load power acquisition begins until the maximum duration is reached for 60s, in this embodiment, the sample acquisition frequency is 10 Hz. Therefore, 600 real-time load power data points are generated in the successful under-pressure starting process of each air compressor, and a real-time load power sequence with time sequence information is formed and used as power change characteristic data. It should be noted that when the power of the motor exceeds the nominal value, the equilibrium pressure of the pressurized start exceeds a certain value, and the power can be reduced through the regulation of the anti-surge valve, so that the compressor can be normally started under pressure.
The collected real-time load power is processed as follows to obtain normalized fluctuation characteristics: p = Pn/Ptypical，PnFor the acquired airAnd (5) real-time power of the press. And taking the ratio of the real-time load power to the power nominal value as an overload power fluctuation characteristic data point in the acquisition time length to represent the proportion exceeding the nominal value, wherein an overload power fluctuation characteristic sequence formed by all the overload power fluctuation characteristic data points in the acquisition time length represents the fluctuation condition of the real-time load power value of the air compressor near the power nominal value. Wherein n is the nth point in the sequence, n is the [1,600 ]]And taking the overload power fluctuation characteristic power point sequence as an overload power fluctuation characteristic sequence P of the air compressor. Simultaneously recording the opening angle Q of the anti-surge valve, wherein Q is a proportional value and belongs to [0,1 ]]Wherein Q =0 represents full-off, Q =1 represents full-on, and the open angle is in its value range, and its value is bigger, and anti-surge valve aperture is bigger.
Although the fluid property of the air compressor cannot be known by counting the overload power characteristics of the air compressor, the real-time load power of the air compressor is the easiest to obtain and the wrong data is the most difficult to occur in the actual information acquisition process. Further, the fluctuation characteristic of the overload power can well represent the size of the load, the compression characteristic of the fluid and the characteristic of the process in the starting process of the air compressor.
And repeating the steps, acquiring the successful pressurized starting information of the air compressors, and obtaining the record { P, Q } of the successful pressurized starting of each air compressor.
For convenience of understanding, the present embodiment describes a method for calculating the similarity between the overload power fluctuation characteristic sequences of the air compressor a and the air compressor B: for A, B overload power fluctuation characteristic sequences P of two air compressors, a normalized cross-correlation method is used for calculating a normalized cross-correlation coefficient C of the two sequences: c (a, B) = NCC (P)A,PB) Wherein the value range of the normalized correlation coefficient is located at [ -1,1 [)]The normalized cross-correlation coefficient can be in the range of-1, 1]And measuring the similarity of the overload power fluctuation characteristic sequences of the two air compressors in an absolute scale range. The correlation coefficient represents the linear description of the approximation degree between the fluctuation sequences of the starting power of the air compressor under pressure. Generally, the closer to 1, the more closely the two have a linear relationship.
Constructing a combined similarity measurement model to obtain a combined phase embodied by the two air compressors due to factors such as load, fluid, process and the likeSimilarity: l = (1-C (A, B)) × | QA－QBAn | element. Wherein C is a normalized cross-correlation coefficient, 1 is a set value, and 1-C has the meaning that when the change between two overload power fluctuation sequences is in a positive correlation linear relationship, C tends to 1, and the term in L tends to 0, otherwise, the term tends to 1 when no correlation exists, and the term tends to 2 when the correlation is negative, and the overall linear relationship is the more positive correlation is similar, and the L is considered to be smaller. | QA－QB| is the absolute value of the difference between the valve opening angle settings between the two air compressors, reflecting the difference between the two valve opening angles.
And determining clustering information based on L, and calculating L by using a K-Means algorithm on the high-dimensional vector P and the data point Q of each air compressor to serve as the distance between samples. The implementer gives the cluster number K of the clusters, and can select the K value according to the number of the air compressors, so that over-segmentation among samples is realized, and the K value does not need to be finely adjusted. Preferably, in this embodiment, the number of samples of the compressor is 600, and K is 50. Finally, a plurality of clusters are obtained and used for distinguishing the pressurized starting working conditions of different air compressors and the opening angle of the anti-surge valve.
Then, combining clusters of which the opening angle difference value of the anti-surge valve is smaller than a preset opening angle difference value threshold value; pre-training the twin network: the input of the twin network is an overload power fluctuation characteristic sequence and an opening angle of any two air compressors in the starting process, the input is labeled according to whether the two air compressors belong to the same cluster, and the output of the twin network is a characteristic descriptor of each air compressor; continuing to train the twin network: if the input of the twin network is from the same cluster, the label value is a first set value, otherwise, the label value is the joint similarity of the two air compressors; and determining the maximum likelihood estimation opening angle of the air compressor to be self-tested according to the similarity of the air compressor to be self-tested characteristic descriptor and other air compressor characteristic descriptors.
And pre-training the twin network on the clustering result to ensure that the twin network model can distinguish the air compressors to be most similar to a certain cluster of samples based on P and Q sequences between the air compressors.
The twin network model used in this embodiment is a multilayer perceptron (MLP) with 5 hidden layers, and the size of the input tensor is 601 data points, which correspond to P and Q data of one air compressor. And outputting a 256-dimensional high-dimensional vector serving as a characteristic descriptor of the air compressor. In the training process, if two groups of samples come from the same cluster, the label value is 1, otherwise, if the two groups of samples do not come from the same cluster, the label value is 0, and the loss function uses the mean square error loss. The cosine similarity between the feature descriptors of the same cluster tends to 1, and the cosine similarity between the feature descriptors of different clusters tends to 0.
The method for training the twin network is a classic comparative learning strategy, and the method for training the twin network is well known in the field of artificial intelligence, so that the detailed description is omitted.
Given a preset opening angle difference threshold q, preferably q =0.005 in this embodiment. The cluster combination method is used for combining clusters, wherein in the air compressor { P, Q } data in two clusters, the absolute value of the difference value of the opening angles Q of the anti-surge valves between every two clusters is smaller than Q. Preferably, when merging clusters, the difference of Q-means of each cluster is compared with a threshold Q.
The purpose of merging the clusters obtained by clustering again is as follows: because the combined similarity L considers the overload power fluctuation characteristic and the anti-surge valve opening angle of the two air compressors, the performances of the air compressors with similar opening angles are more similar, but because the similar clusters obtained directly according to the opening angles can ignore the characteristic of the air compressor load power fluctuation, the combined similarity L is firstly clustered to realize over-segmentation, and then partial clusters are merged based on the preset opening angle difference threshold value to obtain the final clustering result.
And continuing training the twin network based on the joint similarity to obtain the similarity of the air compressor and obtain the fine-grained subclass. Continuing to train the twin network, wherein the label value at this time is no longer 1 or 0, the label value in the same cluster is a first set value 1, and the label value of the sample between different clusters is the joint similarity of two samples: l = (1-C (A, B)) × | QA－QB｜。
The purpose of continuing to train the twin network is: simple clustering can obtain the similarity between the air compressors, and need obtain data when all air compressors are started in the area successfully like this, contrast one by one, can't directly obtain the similarity according to the data when user's air compressor is started in the area successfully, can lead to other users ' air compressor to take the data when starting to reveal in the area simultaneously, can't protect all air compressor user's industry big data.
The precision can be guaranteed and a small deep neural network model can be obtained by directly pre-training and distance learning the information by using the twin network, and the number of hidden layers of the MLP is 5, so that the MLP is suitable for mapping high-dimensional vectors onto the feature descriptors, therefore, a user can obtain the corresponding feature descriptors only by sending data into the twin network, and the precision of an estimated interval is higher (the similarity between the outside of a cluster is 1-L/Max (L), the similarity between the inside of the cluster is close to 1) by comparing the data of all the feature descriptors, and the network training speed is faster.
After the training is finished, inputting data of all air compressors of the big data platform in the process of starting under pressure into the twin network, and obtaining the corresponding feature descriptors of all the air compressors of the big data platform.
Obtaining the maximum likelihood estimation opening angle of the anti-surge valve based on the similarity of the air compressor feature descriptors: and obtaining the similarity S between the air compressor feature descriptor to be self-tested and other air compressor feature descriptors. Presetting a similarity threshold t, and calculating the opening angles corresponding to m samples with the similarity greater than t to obtain a maximum likelihood estimation opening angle Qm：
Wherein the content of the first and second substances,
for the similarity between the ith sample in the m samples and the characteristic descriptor of the air compressor to be self-tested,
the opening angle of the ith sample of the m samples. If air compressor data are newly added to the large data platform and pressurized starting is not attempted, the default opening angle is 1 when the characteristic descriptor of the air compressor is obtained.
Preferably, the preset similarity threshold t may be fixed, and the preset similarity threshold t in this embodiment is 0.7.
Preferably, the preset similarity threshold t can be customized according to the air compressor to be self-checked: determining the cluster to which the air compressor to be self-inspected belongs according to the characteristic descriptor of the air compressor to be self-inspected, and determining the distance D between the air compressor to be self-inspected and the center of the cluster to which the air compressor to be self-inspected belongs, the density D of the cluster to which the air compressor to be self-inspected belongs and the shortest distance D between the air compressor to be self-inspected and other clusters kmin,kDetermining a preset similarity threshold:
t=Min(α+β*D*∑(dmin,k)/[(d+r)*(K-1)],1)
wherein, alpha and beta are constant parameters, preferably, alpha takes 0.6, beta takes 0.37, K is the number of clusters, sigma (d)min,k) And D is normalized density which is the sum of the shortest distances between the air compressor to be self-inspected and other clusters, after the densities of all the clusters are obtained, the densities of all the clusters are normalized by utilizing the maximum density, and r is the minimum circumscribed circle radius of the minimum convex polygon corresponding to all the sample points.
And finally, adjusting the opening angle of the anti-surge valve according to the maximum likelihood estimation opening angle, and trying to start the air compressor under pressure.
In the process of starting and self-checking the air compressor under pressure, the opening range of the valve is tried from large to small, namely the opening angle of the anti-surge valve is gradually closed, so that the load of the air compressor is reduced, and at the moment, the opening angle of the valve of the air compressor is in a conservative range [ Q [m，Qmax]Estimating the opening angle Q from the maximum likelihoodmAnd the maximum opening angle of the anti-surge valve of the air compressor. If Q is reachedmIf the operation is unsuccessful, the air compressor cannot be started in a conservative interval, and at the moment, maintenance personnel are reported, and are enabled to arrive in the field in advance or the starting attempt is terminated. If the attempt of starting the air compressor under pressure is successful in the conservative interval, the following operations are carried out: if the opening angle of successful start is larger than the opening angle corresponding to the characteristic descriptor of the air compressor to be self-checked, updating the characteristic descriptor of the air compressor to be self-checked according to the starting process (inputting the overload power fluctuation characteristic sequence and the opening angle of successful start into a twin network to obtain the characteristic descriptor of the air compressor to be self-checked); and updating the preset similarity threshold to obtain a more accurate maximum likelihood estimation opening angle. The purpose of doing this is to get the ideal opening angle gradually after many iterations, making the start-up process faster. In particular, the default phaseThe similarity threshold is updated as follows: obtaining the opening angle and Q of successful start-upmThe difference delta, t + delta/Q ofmWherein t is a preset threshold, δ is a constant coefficient, and preferably δ is 0.15.
The strategy for attempting the pressurized start of the air compressor according to the maximum likelihood estimation opening angle can be set by an implementer, and only the corresponding action of each process is illustrated in the embodiment. If the process of trying to start the air compressor under pressure is unsuccessful, the process means that the load of starting under pressure is overlarge, and the air compressor is protected. In addition, the surge risk interval [ Q ] can be setmin，Qm]The present invention will not be described in detail herein.
Therefore, self-checking of the air compressor based on industrial big data is realized, and the opening angle Q is estimated through the maximum likelihoodmAnd (4) carrying out pressurized starting attempt, and reminding related maintenance personnel in time when the attempt fails so as to quicken personnel response and facilitate the attempt in a surge risk interval under the supervision condition.
It should be noted that: the precedence order of the above embodiments of the present invention is only for description, and does not represent the merits of the embodiments. And specific embodiments thereof have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (7)
1. The air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence is characterized by comprising the following steps of: acquiring an overload power fluctuation characteristic sequence and an opening angle of an anti-surge valve in the starting process of the air compressor which is successfully started under pressure; acquiring a difference value between a normalized correlation coefficient of an overload power fluctuation characteristic sequence and a set value in the starting process of the two air compressors and an absolute value of an opening angle difference value of the two air compressors, wherein the product of the normalized correlation coefficient and the set value is the joint similarity of the starting process of the two air compressors; clustering air compressors according to the joint similarity of any two air compressors to obtain a plurality of clusters; wherein, the overload power fluctuation characteristic sequence in the starting process of the air compressor which is successfully started under pressure comprises the following steps: determining the acquisition duration of real-time load power in the process of starting the air compressor under pressure; when the starting load power of the air compressor exceeds a nominal value, starting to collect real-time load power, and ending the collection time; taking the ratio of the real-time load power to the power nominal value as an overload power fluctuation characteristic data point in the acquisition time length, and forming an overload power fluctuation characteristic sequence by all the overload power fluctuation characteristic data points in the acquisition time length;
combining clusters of which the opening angle difference value of the anti-surge valve is smaller than a preset opening angle difference value threshold;
pre-training the twin network: the input of the twin network is an overload power fluctuation characteristic sequence and an opening angle of any two air compressors in the starting process, the input is labeled according to whether the two air compressors belong to the same cluster, and the output of the twin network is a characteristic descriptor of each air compressor;
continuing to train the twin network: if the input of the twin network is from the same cluster, the label value is a first set value, otherwise, the label value is the joint similarity of the two air compressors;
acquiring the similarity of the characteristic descriptor of the air compressor to be self-checked and the characteristic descriptors of other air compressors, and selecting the similarity larger than a preset similarity threshold value and the corresponding opening angles of other air compressors; taking the selected similarity as the weight corresponding to the opening angles of other air compressors, and obtaining a weighted average value as the maximum likelihood estimation opening angle of the air compressor to be self-checked;
and adjusting the opening angle of the anti-surge valve according to the maximum likelihood estimation opening angle, and trying to start the air compressor under pressure.
2. The method according to claim 1, wherein the opening angle of the anti-surge valve during the starting process of the air compressor which is successfully started under pressure is a proportional value, and the larger the value is, the larger the opening angle of the anti-surge valve is.
3. The method of claim 1, wherein the clustering the air compressors according to the joint similarity of any two air compressors to obtain a plurality of clusters comprises: determining the cluster number of clusters to be clustered according to the number of the air compressors as a sample; acquiring the joint similarity of any two air compressors as the distance between samples; and clustering the air compressor according to the number of the clusters and the distance between the samples to obtain a plurality of clusters.
4. The method of claim 1, wherein adjusting an anti-surge valve opening angle based on a maximum likelihood estimation opening angle, wherein attempting an air compressor start-up with pressure comprises: forming a valve opening angle adjustment conservative interval according to the maximum likelihood estimation opening angle and the maximum opening angle of the anti-surge valve of the air compressor; and adjusting the opening angle of the valve in the conservative interval, if the opening angle of the valve is the maximum likelihood estimation opening angle, the starting of the air compressor is still unsuccessful, and stopping trying to protect the air compressor.
5. The method of claim 1, wherein adjusting an anti-surge valve opening angle based on a maximum likelihood estimate opening angle, wherein attempting an air compressor start-up with pressure further comprises: if the attempt of starting the air compressor under pressure is successful in the conservative interval, updating a preset similarity threshold; a more accurate maximum likelihood estimate opening angle is obtained.
6. The method of claim 5, wherein the predetermined similarity threshold is 0.7.
7. The method according to claim 5, characterized in that the preset similarity threshold t:
t=Min(α+β*D*∑(dmin,k)/[(d+r)*(K-1)],1)
wherein alpha and beta are constant parameters, D is the density of the cluster to which the air compressor belongs to be self-tested, and Dmin,kThe minimum distance between the air compressor to be self-inspected and other clusters K is defined, K is the number of clusters, d is the distance between the air compressor to be self-inspected and the center of the cluster to which the air compressor belongs, and r is the minimum circumscribed circle radius of the minimum convex polygon corresponding to all data points during clustering.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210024089.XA CN114037026B (en) | 2022-01-11 | 2022-01-11 | Air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210024089.XA CN114037026B (en) | 2022-01-11 | 2022-01-11 | Air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114037026A CN114037026A (en) | 2022-02-11 |
| CN114037026B true CN114037026B (en) | 2022-03-25 |
Family
ID=80147458
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210024089.XA Active CN114037026B (en) | 2022-01-11 | 2022-01-11 | Air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114037026B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102297120A (en) * | 2011-08-04 | 2011-12-28 | 长春工业大学 | Air compressor surge-preventing generalized predictive control system and method |
| CN110067768A (en) * | 2019-05-16 | 2019-07-30 | 济南大陆机电股份有限公司 | A kind of compressor anti-surge control method and system |
| CN110878759A (en) * | 2018-09-06 | 2020-03-13 | 新特能源股份有限公司 | Anti-surge control method for high-rotation-speed centrifugal compressor |
| CN112580267A (en) * | 2021-01-13 | 2021-03-30 | 南京航空航天大学 | Aero-engine surge prediction method based on multi-branch feature fusion network |
| WO2021248746A1 (en) * | 2020-06-10 | 2021-12-16 | 大连理工大学 | Axial flow compressor stall and surge prediction method based on deep learning |
-
2022
- 2022-01-11 CN CN202210024089.XA patent/CN114037026B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102297120A (en) * | 2011-08-04 | 2011-12-28 | 长春工业大学 | Air compressor surge-preventing generalized predictive control system and method |
| CN110878759A (en) * | 2018-09-06 | 2020-03-13 | 新特能源股份有限公司 | Anti-surge control method for high-rotation-speed centrifugal compressor |
| CN110067768A (en) * | 2019-05-16 | 2019-07-30 | 济南大陆机电股份有限公司 | A kind of compressor anti-surge control method and system |
| WO2021248746A1 (en) * | 2020-06-10 | 2021-12-16 | 大连理工大学 | Axial flow compressor stall and surge prediction method based on deep learning |
| CN112580267A (en) * | 2021-01-13 | 2021-03-30 | 南京航空航天大学 | Aero-engine surge prediction method based on multi-branch feature fusion network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114037026A (en) | 2022-02-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110852426B (en) | Pre-training model integration acceleration method and device based on knowledge distillation | |
| CN109766950B (en) | Industrial user short-term load prediction method based on morphological clustering and LightGBM | |
| US20210233299A1 (en) | Speech-driven facial animation generation method | |
| Tscherepanow | TopoART: A topology learning hierarchical ART network | |
| CN110941734A (en) | Depth unsupervised image retrieval method based on sparse graph structure | |
| CN114037026B (en) | Air compressor pressurized starting self-checking method based on industrial big data and artificial intelligence | |
| CN115657265A (en) | Zoom control method and system for lens | |
| CN116503676A (en) | Picture classification method and system based on knowledge distillation small sample increment learning | |
| CN116258281A (en) | Internet of things fire control monitoring and regulating system based on cloud platform management | |
| CN111508475A (en) | Robot awakening voice keyword recognition method and device and storage medium | |
| Lu et al. | Unsupervised speaker segmentation and tracking in real-time audio content analysis | |
| CN116992246B (en) | Intelligent sensing method and system for underground airflow parameters | |
| CN113157918A (en) | Commodity name short text classification method and system based on attention mechanism | |
| CN112101659A (en) | Complex equipment service life prediction method based on stacking denoising autoencoder | |
| CN115596654B (en) | Reciprocating compressor fault diagnosis method and system based on state parameter learning | |
| CN111080088A (en) | Method for quickly judging product quality based on clustered hypersphere model | |
| CN113792540B (en) | Method for updating intention recognition model and related equipment | |
| CN116504230A (en) | Data closed-loop method, device, computer equipment and computer readable storage medium | |
| CN115862103A (en) | Method and system for identifying face of thumbnail | |
| CN112966108B (en) | Method, apparatus, device and storage medium for detecting data and training classification model | |
| US11373657B2 (en) | System and method for speaker identification in audio data | |
| CN114067803A (en) | Speaker confirmation method based on distance correlation metric learning | |
| Chen et al. | An evolutionary based clustering algorithm applied to dada compression for industrial systems | |
| CN117370870B (en) | Knowledge and data compound driven equipment multi-working condition identification and performance prediction method | |
| CN111191689B (en) | Sample data processing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |