Ore pulp pipeline flow velocity detection method, device and equipment and readable storage medium

Abstract

The invention discloses an ore pulp pipeline flow velocity detection method, device and equipment and a readable storage medium, and relates to the technical field of automation. The method comprises the following steps: acquiring a plurality of turbulent vibration signals corresponding to the positions of sensors on the pipe wall of an ore pulp pipeline, and constructing a feature engineering sample set; obtaining a plurality of flow velocity sample sets according to each feature engineering sample set; according to each feature engineering sample set and each flow velocity sample set, training a gradient boosting decision tree model to obtain a target prediction model; obtaining a to-be-detected feature engineering sequence; predicting each feature engineering sequence through each target prediction model to obtain a flow velocity prediction parameter; and obtaining a target flow velocity detection parameter according to the flow velocity prediction parameter. Therefore, on the basis of accurately detecting the flow velocity of the ore pulp pipeline, the problems that the ore pulp pipeline needs to be installed after being damaged in a contact type detection method, the installation difficulty is large, and maintenance is difficult are solved, and meanwhile the problem that a contact type detection device is damaged is solved.

Description

Technical Field

[0001] The present invention relates to the field of automation technology, and in particular to a method, device, equipment and readable storage medium for detecting flow velocity in a slurry pipeline. Background Art

[0002] In the mineral processing process, the slurry flow rate has a direct impact on the processing capacity of the flotation machine. At present, the methods for detecting the flow rate of slurry pipelines at home and abroad mainly include non-contact detection and contact detection. Among the non-contact detection methods, the Doppler effect is not suitable in principle. The contact detection method requires the slurry pipeline to be destroyed before installation, which is difficult to install and maintain. In some mineral processing processes, the slurry is prone to scaling and strong corrosion, which will damage the contact detection device when used for a long time. Summary of the invention

[0003] In view of this, the purpose of the present invention is to overcome the shortcomings of the prior art and provide a slurry pipeline flow velocity detection method, device, equipment and readable storage medium, which are used to model the turbulent vibration signal value obtained by a polyvinylidene fluoride (PVDF) sensor through machine learning, and then perform slurry pipeline flow velocity detection.

[0004] The present invention provides the following technical solutions:

[0005] In a first aspect, the present invention provides a method for detecting flow velocity in a slurry pipeline, comprising:

[0006] Acquire multiple turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and construct a feature engineering sample set corresponding to each sensor position according to the multiple turbulent vibration signals corresponding to each sensor position;

[0007] Perform flow velocity calibration according to each of the characteristic engineering sample sets to obtain multiple flow velocity sample sets;

[0008] The gradient boosting decision tree model is trained according to each of the feature engineering sample sets and each of the flow velocity sample sets to obtain a target prediction model corresponding to each of the sensor positions;

[0009] Acquire the turbulent vibration signal to be detected corresponding to each of the sensor positions, and obtain the feature engineering sequence corresponding to each of the sensor positions according to each of the turbulent vibration signals to be detected;

[0010] Each of the feature engineering sequences is predicted by each of the target prediction models to obtain a flow velocity prediction parameter corresponding to each of the sensor positions;

[0011] The target flow rate detection parameters are obtained according to the flow rate weights and flow rate prediction parameters corresponding to each of the sensor positions.

[0012] In one embodiment, the step of acquiring a plurality of turbulent vibration signals corresponding to each sensor position on the wall of a slurry pipeline, and constructing a feature engineering sample set corresponding to each sensor position according to each turbulent vibration signal, comprises:

[0013] For each of the sensor positions, based on a preset time interval, a plurality of turbulent vibration signals at a plurality of slurry flow rates are acquired within a preset acquisition time period to obtain the turbulent vibration signal set;

[0014] The turbulent vibration signal set is subjected to dimension reduction to obtain the feature engineering sample set.

[0015] In one embodiment, the dimensionality reduction of the turbulent vibration signal set to obtain the feature engineering sample set includes:

[0016] Filtering the turbulent vibration signal set based on a moving average filtering algorithm;

[0017] The filtered turbulent vibration signal set is subjected to dimension reduction to obtain the feature engineering sample set.

[0018] In one embodiment, the gradient boosting decision tree model is trained according to each of the feature engineering sample sets and each of the flow rate sample sets to obtain a target prediction model corresponding to each of the sensor positions, including:

[0019] For each of the sensor positions, determining a training sample set and a test sample set according to the feature engineering sample set and the flow velocity sample set;

[0020] Based on K-fold cross validation, the gradient boosting decision tree model is trained according to the training sample set to obtain K candidate prediction models, and a first prediction model is obtained from the K candidate prediction models based on preset model indicators;

[0021] Based on the preset model indicators, model tests are performed on each of the candidate prediction models according to the test sample set to obtain a second prediction model;

[0022] If the first prediction model and the second prediction model are the same, the first prediction model or the second prediction model is used as the target prediction model.

[0023] In one embodiment, the method further comprises:

[0024] If the first prediction model and the second prediction model are different, the model parameters of the gradient boosting decision tree model are adjusted, and the step of re-executing the K-fold cross-validation to train the gradient boosting decision tree model according to the training sample set to obtain K candidate prediction models, and obtaining the first prediction model from the K candidate prediction models based on preset model indicators.

[0025] In one embodiment, the step of acquiring the turbulent vibration signal to be detected corresponding to each of the sensor positions, and obtaining the feature engineering sequence corresponding to each of the sensor positions according to each of the turbulent vibration signals to be detected, comprises:

[0026] For each of the sensor positions, based on the preset time interval, the turbulent vibration signal to be detected is acquired to obtain a sequence to be detected;

[0027] The dimension of the sequence to be detected is reduced to obtain the feature engineering sequence.

[0028] In one embodiment, the step of reducing the dimension of the sequence to be detected to obtain the feature engineering sequence includes:

[0029] Filtering the sequence to be detected based on a moving average filtering algorithm;

[0030] The filtered sequence to be detected is subjected to dimension reduction to obtain the feature engineering sequence.

[0031] In a second aspect, the present invention provides a slurry pipeline flow velocity detection device, comprising:

[0032] A first acquisition module is used to acquire a plurality of turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and to construct a feature engineering sample set corresponding to each sensor position according to the plurality of turbulent vibration signals corresponding to each sensor position;

[0033] A calibration module, used for performing flow velocity calibration according to each of the feature engineering sample sets to obtain multiple flow velocity sample sets;

[0034] A training module, used to train the gradient boosting decision tree model according to each of the feature engineering sample sets and each of the flow velocity sample sets, to obtain a target prediction model corresponding to each of the sensor positions;

[0035] A second acquisition module is used to acquire the turbulent vibration signal to be detected corresponding to each of the sensor positions, and obtain a feature engineering sequence corresponding to each of the sensor positions according to each of the turbulent vibration signals to be detected;

[0036] A prediction module, used to predict each of the feature engineering sequences respectively through each of the target prediction models to obtain a flow velocity prediction parameter corresponding to each of the sensor positions;

[0037] The detection module is used to obtain the target flow rate detection parameter according to the flow rate weight and flow rate prediction parameter corresponding to each of the sensor positions.

[0038] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for detecting flow velocity in a slurry pipeline as described in the first aspect is implemented.

[0039] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the method for detecting flow velocity in a slurry pipeline as described in the first aspect is implemented.

[0040] The present invention discloses a method, device, equipment and readable storage medium for detecting flow velocity in a slurry pipeline. The method obtains a plurality of turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and constructs a feature engineering sample set corresponding to each sensor position according to the plurality of turbulent vibration signals corresponding to each sensor position; performs flow velocity calibration according to each feature engineering sample set to obtain a plurality of flow velocity sample sets; trains a gradient boosting decision tree model according to each feature engineering sample set and each flow velocity sample set, respectively, to obtain a target prediction model corresponding to each sensor position; obtains a turbulent vibration signal to be detected corresponding to each sensor position, and obtains a feature engineering sequence corresponding to each sensor position according to each turbulent vibration signal to be detected; predicts each feature engineering sequence by each target prediction model, respectively, to obtain a flow velocity prediction parameter corresponding to each sensor position; obtains a target flow velocity detection parameter according to a flow velocity weight and a flow velocity prediction parameter corresponding to each sensor position. The turbulent vibration signals obtained by PVDF sensors at different positions are modeled and trained through the gradient boosting decision tree model to obtain the target prediction model corresponding to each sensor. The target prediction model is further used to predict the flow velocity prediction parameters of different sensor positions. Finally, each flow velocity prediction parameter is weightedly calculated based on the flow velocity weights of different sensor positions to obtain the final target flow velocity detection parameters. This solves the problem that the contact detection method requires the slurry pipeline to be destroyed before installation, which is difficult to install and maintain. At the same time, it solves the problem that in some mineral processing processes, the slurry is prone to scaling and strong corrosion, which will damage the contact detection device when used for a long time. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope of protection of the present invention. In each of the drawings, similar components are numbered similarly.

[0042] Figure 1 A schematic diagram of a flow chart of a slurry pipeline flow velocity detection method proposed in this embodiment is shown;

[0043] Figure 2 A schematic diagram showing the PVDF sensor proposed in this embodiment installed on a slurry pipeline is shown;

[0044] Figure 3 Another schematic diagram of the flow chart of the slurry pipeline flow velocity detection method proposed in this embodiment is shown;

[0045] Figure 4 A schematic diagram of the moving average filtering algorithm proposed in this embodiment is shown;

[0046] Figure 5 A schematic diagram of the feature engineering proposed in this embodiment is shown;

[0047] Figure 6 Another schematic diagram of the flow chart of the slurry pipeline flow velocity detection method proposed in this embodiment is shown;

[0048] Figure 7 A schematic structural diagram of a slurry pipeline flow velocity detection device proposed in this embodiment is shown.

[0049] Description of the accompanying drawings:

[0050] 700 - slurry pipeline flow rate detection device; 701 - first acquisition module; 702 - calibration module; 703 - training module; 704 - second acquisition module; 705 - prediction module; 706 - detection module. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0052] The components of the embodiments of the present invention generally described and shown in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0053] Hereinafter, the terms "including", "having" and their cognates, which may be used in various embodiments of the present invention, are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the foregoing items or adding the possibility of one or more features, numbers, steps, operations, elements, components or combinations of the foregoing items.

[0054] Furthermore, the terms “first”, “second”, “third”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0055] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meanings as those generally understood by those skilled in the art to which the various embodiments of the present invention belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meanings as the contextual meanings in the relevant technical field and will not be interpreted as having idealized meanings or overly formal meanings unless clearly defined in the various embodiments of the present invention.

[0056] Example 1

[0057] The fluid in the slurry pipeline is a liquid mixture formed by mixing solid raw materials such as ore and mineral soil with water and other auxiliary agents. It is a multiphase fluid mixed with gas, liquid and solid. Among the non-contact detection methods, the Doppler effect is not suitable in principle. The contact detection method requires the slurry pipeline to be destroyed before installation, which is difficult to install and maintain. In some mineral processing processes, the slurry is prone to scaling and strong corrosion, which will damage the contact detection device when used for a long time.

[0058] At present, the common way to obtain pipe wall vibration is to use a PVDF sensor wrapped around the outer wall of the pipe to listen to the vibration caused by turbulence when the fluid passes through the pipe wall. Among them, the PVDF sensor is a piezoelectric sensor made of PVDF (polyvinylidene fluoride) material, which has flexible and piezoelectric properties. PVDF sensors have the advantages of high sensitivity, fast response time, light weight, thin thickness, and foldability. They are widely used in robotics, automobiles, aerospace and other fields to detect physical quantities such as pressure, vibration, and displacement.

[0059] The disclosed embodiment provides a method for detecting flow velocity in a slurry pipeline, which is used to model the turbulent vibration signal value obtained by a PVDF sensor through machine learning, and then perform flow velocity detection in the slurry pipeline.

[0060] See also Figure 1 A method for detecting flow velocity in a slurry pipeline includes steps S101 to S106, and each step is described in detail below.

[0061] Step S101, obtaining a plurality of turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and constructing a feature engineering sample set corresponding to each sensor position according to the plurality of turbulent vibration signals corresponding to each sensor position.

[0062] Please note that, see Figure 2 PVDF sensors are installed in the middle of multiple relatively stable long straight slurry pipelines to ensure that the turbulent vibration signal data collection and slurry pipeline flow velocity detection results are representative. Usually 4-8 PVDF sensors are installed. For the long straight pipeline where the PVDF sensor is installed, the length of the long straight pipeline is generally greater than or equal to 10 to 50 times the pipe diameter before the slurry reaches the PVDF sensor, and the length of the long straight pipeline is generally greater than or equal to 5 times the pipe diameter after the slurry passes through the PVDF sensor.

[0063] In this embodiment, a plurality of turbulent vibration signals collected from each sensor position on the wall of the slurry pipeline are obtained, and a feature engineering sample set corresponding to each sensor position is constructed according to the plurality of turbulent vibration signals corresponding to each sensor position.

[0064] It should be noted that the signal obtained by the fixed acquisition frequency of the PVDF sensor is characterized. If the acquisition frequency is too low, the signal information obtained is very little, and it is impossible to reflect that the vibration signal changes with the flow rate. When the subsequent modeling based on the gradient boosting decision tree (CatBoost) model is used, it is impossible to establish a suitable model to realize the slurry pipeline flow rate detection.

[0065] A high acquisition frequency is needed to obtain more signal information. However, if all the data collected in one second is directly used as model input for modeling and slurry pipeline velocity detection, it will cause the ultra-high-dimensional training data to be unable to load during model training, and the model will run on an industrial computer with insufficient hardware resources or slow detection speed during velocity detection. Therefore, it is necessary to maintain a high acquisition frequency and build low-dimensional feature engineering at the same time so that the CatBoost model can be trained and velocity detected normally.

[0066] See also Figure 3 In a specific embodiment, step S101 includes steps S1011 to S1012, and each step is described in detail below.

[0067] Step S1011, for each of the sensor positions, based on a preset time interval, a plurality of turbulent vibration signals at a plurality of slurry flow rates are acquired within a preset acquisition time length to obtain the turbulent vibration signal set.

[0068] In this embodiment, for sensors at various positions, multiple turbulent vibration signals at multiple slurry flow rates are acquired within a preset acquisition time period based on a preset time interval to obtain a turbulent vibration signal set.

[0069] Exemplarily, the acquisition frequency of the PVDF sensor is set to F, and then at different slurry flow rates, the turbulent vibration signal of the PVDF sensor within a preset acquisition time is acquired at a preset time interval of 1 / F.

[0070] Define the sample space of the turbulent vibration signal set corresponding to each PVDF sensor as A, and there exists: A = (v1, v2, v3, ..., v n ).

[0071] The turbulent vibration signal data of multiple slurry flow rates under n types are collected cumulatively. i (i=1,2,3,…,n) is the sample set under the i-th flow rate, and there exists: v i =(D1,D2,D3,…,D T ).

[0072] At each slurry flow rate, T seconds of data are collected cumulatively. j (j=1,2,3,…,T) is the turbulent vibration signal data of the jth second, and there exists: D j =[p1,p2,p3,…,p F ].

[0073] Data is collected every second at a time interval of 1 / F, that is, the interval between p1 and p2 is 1 / F, and so on, F turbulent vibration signals are collected every second.

[0074] Step S1012, reducing the dimension of the turbulent vibration signal set to obtain the feature engineering sample set.

[0075] In this embodiment, the turbulent vibration signal set is reduced in dimension to obtain a feature engineering sample set. By converting high-dimensional data into low-dimensional data through feature engineering, while retaining the key information in the original data as much as possible, it can not only reduce the computational complexity, but also improve the efficiency of data processing.

[0076] In a specific embodiment, step S1012 includes: filtering the turbulent vibration signal set based on a moving average filtering algorithm; and reducing the dimension of the filtered turbulent vibration signal set to obtain the feature engineering sample set.

[0077] It should be noted that when the PVDF sensor obtains the turbulent vibration signal of the slurry in the pipeline, the signal value will change with the change of flow rate. Since the PVDF sensor collects the vibration signal of the slurry pipeline wall, and the slurry pipeline is usually spread throughout the entire beneficiation plant, the vibration generated by the operation of large-scale mechanical equipment in the beneficiation plant, as well as the vibration generated by knocking, cutting and other actions during operations around the slurry pipeline, will interfere with the signal value collected by the PVDF sensor. Therefore, when performing slurry pipeline flow rate detection, it is necessary to remove the abnormal data when the PVDF sensor signal is collected.

[0078] In this embodiment, the turbulent vibration signal set is filtered by a moving average filtering algorithm.

[0079] For example, see Figure 4 , for the F turbulent vibration signals collected every second, there are:

[0080] Where N is the window size of the moving average filter, q t is the filtered output value of the t-th turbulent vibration signal, p t-m is the input value of the t-th turbulence vibration signal before filtering at the m-th moment. After the moving average filtering algorithm, the turbulence vibration signal per second becomes F-N+1, and there are: D j =[q N ,q N+1 ,q N+2 ,…,q F ].

[0081] Furthermore, the filtered turbulent vibration signal set is reduced in dimension to obtain a feature engineering sample set.

[0082] For example, see Figure 5 , for each second of filtered turbulent vibration data, split it into W segments at equal intervals, and take continuous (F-N+1) / W data in each segment. There exists: D j =[D j1 ,D j2 ,D j3 ,…D jW ],

[0083] Every second of data at each flow rate is feature engineered in this way, and there are: i =(D 11 ,D 12 ,D 13 ,…,D 1W ,D 21 ,D 22 ,D 23 ,…,D 2W,…,D T1 ,D T2 ,D T3 ,…,D TW ).

[0084] Among them, v i (i=1,2,3,…,n) is the sample set under the i-th slurry flow rate, D jc (j=1,2,3,…,T;c=1,2,3,…,W) is the data of the cth segment of the turbulent vibration signal of the jth second. Under each slurry flow rate, the turbulent vibration signal set of T seconds and F vibration signals per second is converted into a feature engineering sample set of T seconds and W segments per second, each segment (F-N+1) / W vibration signals, through feature engineering. When training and testing the model, as well as when detecting the flow rate, the segments of (F-N+1) / W vibration signals are used as features as the input of the model, and the dimension is lower than directly using F vibration signals as features.

[0085] Step S102, performing flow velocity calibration according to each of the feature engineering sample sets to obtain a plurality of flow velocity sample sets.

[0086] In this embodiment, the velocity is calibrated on the feature engineering sample set to obtain the velocity sample space Y, which exists: Y = (y1, y2, y3, ... y n ). Among them, y m ,m=1,2,3,…,n is the slurry flow rate.

[0087] The feature engineering sample set A(v1,v2,v3,…,v n ) The corresponding velocity label is R, and there exists: R = (l1,l2,l3,…l n ).

[0088] Among them, l s =y m ,s=1,2,3,…,n;m=1,2,3,…,n are the flow rate labels corresponding to the feature engineering sample set. The number of features under each slurry flow rate is (F-N+1) / W feature D jc , the flow rate labels are all the current flow rate l s .

[0089] Step S103, training the gradient boosting decision tree model according to each of the feature engineering sample sets and each of the flow velocity sample sets, to obtain a target prediction model corresponding to each of the sensor positions.

[0090] In this embodiment, each feature engineering sample set is used as a feature, and each flow rate sample set is used as a label. The feature engineering sample set corresponding to the same sensor and its corresponding flow rate sample set are input into a gradient boosting decision tree model for training. Similarly, a target prediction model corresponding to each sensor position can be obtained.

[0091] See also Figure 6 In a specific embodiment, step S103 includes steps S1031 to S1034, and each step is described in detail below.

[0092] Step S1031: for each of the sensor positions, determine a training sample set and a test sample set according to the feature engineering sample set and the flow rate sample set.

[0093] In this embodiment, the target prediction model training process corresponding to the sensor at each sensor position is specifically as follows: the feature engineering sample set and the flow rate sample set corresponding to the sensor position need to be divided into a training sample set and a test sample set.

[0094] Define the sample space B of the training sample set and the sample space C of the test sample set. The sample space B of the training sample set is for each slurry flow rate v i Randomly select trainPercent from the sample set, that is, v i =(D 11 ,D 12 ,D 13 ,…,D 1W ,D 21 ,D 22 ,D 23 ,…,D 2W ,…,D T1 ,D T2 ,D T3 ,…,D TW ), randomly select trainPercent samples, and the sample space B also includes the corresponding flow rate labels; the sample space C of the test sample set is the 1-trainPercent samples not selected in the sample space B, and the sample space C also includes the corresponding flow rate labels. There exists: B = (b1, b2, b3, ..., b nTW*trainPercent ), C=(c1,c2,c3,…c nTW*(1-trainPercent) ). Wherein, trainPercent may be 20%, which is not limited in this embodiment.

[0095] Among them, b f ,f=1,2,3,…,nTW*trainPercent is the training data of CatBoost model. v,v=1,2,3,…,nTW*(1-trainPercent) is the test data of CatBoost model. f and c v D jc (j=1,2,3,…,T;c=1,2,3,…,W),namely b f and c v Represents the data of the cth segment of the jth second turbulent vibration signal in the feature engineering sample set.

[0096] Step S1032, based on K-fold cross validation, the gradient boosting decision tree model is trained according to the training sample set to obtain K candidate prediction models, and a first prediction model is obtained from the K candidate prediction models based on preset model indicators.

[0097] In this embodiment, a CatBoost model is constructed and model parameters are set, including: maximum depth of decision tree, learning rate, number of trees used in training, and loss function.

[0098] Furthermore, the original training sample set is divided into K subsets, called "folds", using the K-fold cross-validation method. Then, the CatBoost model is trained and validated K times on K different subsets. In each training, one fold is used as a validation set, and the remaining K-1 folds are used as training sets, thereby obtaining K candidate prediction models and corresponding K validation set indicator test results. The first prediction model is obtained from the K candidate prediction models based on the preset model indicators and K validation scores.

[0099] For example, the original training sample set is divided into K subsets (folds), each with the same number of samples, which is nTW×trainPercent / K. There exists: B=(Data1,Data2,Data3,…,Data K ), By analogy, there exists:

[0100]

[0101] According to Data1,Data2,Data3,…,Data K In order, one of the samples is selected for validation each time, and the remaining data is used for training. The training is performed K times in total. After each training, the model is validated. There are:

[0102] in, For CatBoost model M vs b fThe flow rate detection result is: t z =M(Data h ).

[0103] Compare the test results T = (t1, t2, t3, ..., t nTW×trainPercent / K ) and the corresponding label R h (l1,l2,l3,…l nTW×trainPercent / K ), the preset model indicators include mean absolute error MAE, root mean square error RMSE and determination coefficient R 2 , so it is necessary to calculate each candidate prediction model in turn and in, Is the data used for current verification h Average of all flow rate labels.

[0104] After K trainings, K candidate prediction models and their validation set indicator test results are obtained, and there are:

[0105] ALL Model =(M1,M2,M3,…M K ), VAL_MAE=(MAE val1 ,MAE val2 ,MAE val3 ,…MAE valK ),

[0106] VAL_RMSE = (RMSE val1 ,RMSE val2 ,RMSE val3 ,…RMSE valK ), VAL_R2=(R 2 val1 ,R 2 val2 ,R 2 val3 ,…R 2 valK ).

[0107] Comparing the validation set indicator test results of different candidate prediction models, the mean absolute error MAE, root mean square error RMSE are the smallest, and the determination coefficient R 2 The largest model is the optimal model, and the optimal model is used as the first prediction model.

[0108] The test results of the validation set indicators of this embodiment are as follows in Table 1:

[0109] MAE RMSE <![CDATA[R 2 ]]> 1 0.0539 0.0691 0.9613 2 0.0568 0.0726 0.9559 3 0.0559 0.0710 0.9617 4 0.0544 0.0696 0.9595 5 0.0539 0.0696 0.9652 6 0.0546 0.0698 0.9626 7 0.0540 0.0700 0.9628 8 0.0558 0.0714 0.9611 9 0.0555 0.0701 0.9614 10 0.0539 0.0685 0.9666

[0110] Step S1033: Based on the preset model indicators and according to the test sample set, each of the candidate prediction models is tested to obtain a second prediction model.

[0111] In this embodiment, the test sample set C = (c1, c2, c3, ... c nTW×(1-trainPercent) ) Test all candidate prediction models, and select one candidate prediction model M g ,g=1,2,3,…,K, when testing, there exists: T=(t1,t2,t3,…,t nTW×(1-trainPercent) )=M g (C). Where, t z ,z=1,2,3,…,nTW×(1-trainPercent) is the model M g C v The flow rate detection result is: t z =M g (c v ).

[0112] Compare the test results T = (t1, t2, t3, ..., t nTW×(1-trainPercent) ) and the corresponding label R c (l1,l2,l3,…l nTW×(1-trainPercent) ), calculate the mean absolute error MAE, root mean square error RMSE and determination coefficient R 2 :

[0113]

[0114] in, It is the average value of all flow rate labels in the test sample set. After each candidate prediction model calculates the index for the test sample set, there is: TEST_MAE = (MAE C1 ,MAE C2 ,MAE C3 ,…MAE CK ), TEST_RMSE = (RMSE C1 ,RMSE C2 ,RMSE C3 ,…RMSE CK ), TEST_R2=(R 2 C1 ,R 2 C2 ,R 2 C3 ,…R 2 CK ).

[0115] Comparing the test results of the test set indicators of different models, the mean absolute error MAE, root mean square error RMSE are the smallest, and the determination coefficient R2 The largest model is the best model, i.e. the second most predictive model.

[0116] The test results of the test set indicators of this embodiment are as follows in Table 2:

[0117]

[0118]

[0119] Step S1034: if the first prediction model and the second prediction model are the same, use the first prediction model or the second prediction model as the target prediction model.

[0120] In this embodiment, if the first prediction model selected according to the validation set index test result and the second prediction model selected according to the test set index test result are the same candidate prediction model, then the candidate prediction model is used as the target prediction model M. best .

[0121] In a specific embodiment, if the first prediction model and the second prediction model are different, the model parameters of the gradient boosting decision tree model are adjusted, and the step of re-executing the K-fold cross-validation to train the gradient boosting decision tree model according to the training sample set to obtain K candidate prediction models, and obtaining the first prediction model from the K candidate prediction models based on preset model indicators.

[0122] In this embodiment, if the first prediction model selected according to the validation set index test result and the second prediction model selected according to the test set index test result are not the same candidate prediction model, starting from step S1032, the CatBoost model parameters are adjusted, and the model training, verification, and model testing are performed again until the first prediction model selected according to the validation set index test result is the same as the second prediction model selected according to the test set index test result, and the CatBoost model construction, training, verification, and testing are completed to obtain the target prediction model M. best .

[0123] It should be noted that the feature engineering sample sets corresponding to sensors at different positions and their corresponding flow rate sample sets are all trained and tested with the CatBoost model according to the process of step S103 to obtain the target prediction models at their respective positions for flow rate detection.

[0124] Step S104, acquiring the turbulent vibration signal to be detected corresponding to each of the sensor positions, and obtaining a feature engineering sequence corresponding to each of the sensor positions according to each of the turbulent vibration signals to be detected.

[0125] In this embodiment, the turbulent vibration signal to be detected corresponding to each sensor position is acquired, and the feature engineering sequence corresponding to each sensor position is obtained according to each turbulent vibration signal to be detected.

[0126] In a specific embodiment, step S104 includes: for each of the sensor positions, based on the preset time interval, acquiring the turbulent vibration signal to be detected to obtain a sequence to be detected; and performing dimensionality reduction on the sequence to be detected to obtain the feature engineering sequence.

[0127] In this embodiment, the sequence of turbulent vibration signals to be detected for one second is defined as d j , exists: d j =[p1,p2,p3,…,p F ]. Among them, d j The PVDF sensor collects one second of data at 1 / F intervals, and collects F turbulent vibration signals to be detected every second. Furthermore, the sequence to be detected is reduced in dimension to obtain a feature engineering sequence.

[0128] In a specific embodiment, reducing the dimension of the sequence to be detected to obtain the feature engineering sample set includes: filtering the sequence to be detected based on a moving average filtering algorithm; and reducing the dimension of the filtered sequence to be detected to obtain the feature engineering sequence.

[0129] In this embodiment, the F turbulent vibration signals to be detected collected in one second are used to remove abnormal signals through a moving average filtering algorithm. Specifically: Where N is the window size of the moving average filter, q t is the filtered output value of the tth signal, p t-m is the input value of the tth turbulent vibration signal to be detected before filtering, counted forward to the mth moment. After the moving average filtering algorithm, the turbulent vibration signal to be detected in one second becomes F-N+1, and there are: d j =[q N ,q N+1 ,q N+2 ,…,q F ].

[0130] Furthermore, the turbulent vibration data to be detected after one second of filtering is divided into W segments at equal intervals, and each segment takes continuous (F-N+1) / W data.

[0131] There is a feature engineering sequence d j =[d j1 ,d j2 ,d j3 ,…d jW ],

[0132] Among them, d jc (c=1,2,3,…,W) is the data of the cth segment feature.

[0133] Step S105 , predicting each of the feature engineering sequences respectively through each of the target prediction models to obtain flow velocity prediction parameters corresponding to each of the sensor positions.

[0134] In this embodiment, based on the correspondence between each sensor position and each target detection model, the feature engineering sequence corresponding to each sensor position is input into the corresponding target prediction model to obtain the flow velocity prediction parameters corresponding to each sensor position output by each target prediction model.

[0135] Specifically, for each sensor position, the filtered feature engineering sequence d corresponding to one sensor position for one second is j , the W segment feature d jc (c=1,2,3,…,W) is input into the corresponding target detection model for prediction, and W flow rates are obtained. There are: speed c =M best (d jc ). Among them, speed c (c=1,2,3,…,W) is the flow velocity prediction parameter obtained after the c-th segment feature is input into the target detection model.

[0136] The flow rate at this sensor location in this second final is the average value of W velocity prediction parameters, that is:

[0137] Step S106, obtaining target flow rate detection parameters according to the flow rate weights and flow rate prediction parameters corresponding to the sensor positions.

[0138] In this embodiment, different sensor positions have corresponding different flow rate weights. The target flow rate detection parameter speed for this second is all , exists: speed all =weight1×speed final1 +weight2×speed final2 +…+weightL×speed finalL .

[0139] Among them, weightx, x = 1, 2, 3, ..., L represents the flow velocity weight of the x-th sensor position, speed finalx ,x=1,2,3,…L represents the flow velocity prediction parameter of the xth sensor position, and L represents that a total of L positions are installed with PVDF sensors.

[0140] The slurry pipeline flow velocity detection method proposed in this embodiment obtains multiple turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and constructs a feature engineering sample set corresponding to each sensor position according to the multiple turbulent vibration signals corresponding to each sensor position; calibrates the flow velocity according to each feature engineering sample set to obtain multiple flow velocity sample sets; trains a gradient boosting decision tree model according to each feature engineering sample set and each flow velocity sample set to obtain a target prediction model corresponding to each sensor position; obtains a turbulent vibration signal to be detected corresponding to each sensor position, and obtains a feature engineering sequence corresponding to each sensor position according to each turbulent vibration signal to be detected; predicts each feature engineering sequence by each target prediction model to obtain a flow velocity prediction parameter corresponding to each sensor position; obtains a target flow velocity detection parameter according to a flow velocity weight and a flow velocity prediction parameter corresponding to each sensor position. The turbulent vibration signals obtained by PVDF sensors at different positions are modeled and trained through the gradient boosting decision tree model to obtain the target prediction model corresponding to each sensor. The target prediction model is further used to predict the flow velocity prediction parameters of different sensor positions. Finally, each flow velocity prediction parameter is weightedly calculated based on the flow velocity weights of different sensor positions to obtain the final target flow velocity detection parameters. This solves the problem that the contact detection method requires the slurry pipeline to be destroyed before installation, which is difficult to install and maintain. At the same time, it solves the problem that in some mineral processing processes, the slurry is prone to scaling and strong corrosion, which will damage the contact detection device when used for a long time.

[0141] Example 2

[0142] In addition, the present disclosure provides a slurry pipeline flow velocity detection device 700, see Figure 7 , the device comprises:

[0143] The first acquisition module 701 is used to acquire a plurality of turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and construct a feature engineering sample set corresponding to each sensor position according to the plurality of turbulent vibration signals corresponding to each sensor position;

[0144] A calibration module 702 is used to perform flow velocity calibration according to each of the feature engineering sample sets to obtain multiple flow velocity sample sets;

[0145] A training module 703 is used to train the gradient boosting decision tree model according to each of the feature engineering sample sets and each of the flow rate sample sets to obtain a target prediction model corresponding to each of the sensor positions;

[0146] The second acquisition module 704 is used to acquire the turbulent vibration signal to be detected corresponding to each of the sensor positions, and obtain the feature engineering sequence corresponding to each of the sensor positions according to each of the turbulent vibration signals to be detected;

[0147] A prediction module 705 is used to predict each of the feature engineering sequences using each of the target prediction models to obtain a flow velocity prediction parameter corresponding to each of the sensor positions;

[0148] The detection module 706 is used to obtain the target flow rate detection parameter according to the flow rate weight and flow rate prediction parameter corresponding to each of the sensor positions.

[0149] Optionally, the first acquisition module 701 is used to obtain multiple turbulent vibration signals at multiple slurry flow rates within a preset collection time period for each sensor position, based on a preset time interval, to obtain the turbulent vibration signal set; and perform dimensionality reduction on the turbulent vibration signal set to obtain the feature engineering sample set.

[0150] Optionally, the first acquisition module 701 is used to filter the turbulent vibration signal set based on a moving average filtering algorithm; and perform dimension reduction on the filtered turbulent vibration signal set to obtain the feature engineering sample set.

[0151] Optionally, the training module 703 is used to determine, for each sensor position, a training sample set and a test sample set based on the feature engineering sample set and the flow velocity sample set; based on K-fold cross validation, perform model training on the gradient boosting decision tree model according to the training sample set to obtain K candidate prediction models, and obtain a first prediction model from the K candidate prediction models based on preset model indicators; based on the preset model indicators, perform model testing on each of the candidate prediction models according to the test sample set to obtain a second prediction model; if the first prediction model and the second prediction model are the same, use the first prediction model or the second prediction model as the target prediction model.

[0152] Optionally, the training module 703 is used to adjust the model parameters of the gradient boosting decision tree model if the first prediction model and the second prediction model are different, and re-execute the step of training the gradient boosting decision tree model based on the K-fold cross-validation to obtain K candidate prediction models, and obtain the first prediction model from the K candidate prediction models based on preset model indicators.

[0153] Optionally, the second acquisition module 704 is used to acquire the turbulent vibration signal to be detected based on the preset time interval for each sensor position to obtain a sequence to be detected; and perform dimensionality reduction on the sequence to be detected to obtain the feature engineering sequence.

[0154] Optionally, the second acquisition module 704 is used to filter the sequence to be detected based on a moving average filtering algorithm; and perform dimension reduction on the filtered sequence to be detected to obtain the feature engineering sequence.

[0155] The device provided in the embodiment of the present disclosure can execute the steps of the slurry pipeline flow velocity detection method provided in Example 1, which will not be described again to avoid repetition.

[0156] The slurry pipeline flow velocity detection device proposed in this embodiment obtains multiple turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and constructs a feature engineering sample set corresponding to each sensor position according to the multiple turbulent vibration signals corresponding to each sensor position; performs flow velocity calibration according to each feature engineering sample set to obtain multiple flow velocity sample sets; trains a gradient boosting decision tree model according to each feature engineering sample set and each flow velocity sample set, respectively, to obtain a target prediction model corresponding to each sensor position; obtains a turbulent vibration signal to be detected corresponding to each sensor position, and obtains a feature engineering sequence corresponding to each sensor position according to each turbulent vibration signal to be detected; predicts each feature engineering sequence by each target prediction model, and obtains a flow velocity prediction parameter corresponding to each sensor position; obtains a target flow velocity detection parameter according to a flow velocity weight and a flow velocity prediction parameter corresponding to each sensor position. The turbulent vibration signals obtained by PVDF sensors at different positions are modeled and trained through the gradient boosting decision tree model to obtain the target prediction model corresponding to each sensor. The target prediction model is further used to predict the flow velocity prediction parameters of different sensor positions. Finally, each flow velocity prediction parameter is weightedly calculated based on the flow velocity weights of different sensor positions to obtain the final target flow velocity detection parameters. This solves the problem that the contact detection method requires the slurry pipeline to be destroyed before installation, which is difficult to install and maintain. At the same time, it solves the problem that in some mineral processing processes, the slurry is prone to scaling and strong corrosion, which will damage the contact detection device when used for a long time.

[0157] Example 3

[0158] In addition, an embodiment of the present disclosure provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for detecting the flow velocity of a slurry pipeline described in Example 1 is implemented.

[0159] The device provided in the embodiment of the present disclosure can execute the steps of the slurry pipeline flow velocity detection method provided in Example 1, which will not be described again to avoid repetition.

[0160] Example 4

[0161] The embodiment of the present disclosure provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for detecting the flow velocity of a slurry pipeline described in Embodiment 1 is implemented.

[0162] In this embodiment, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0163] The computer-readable storage medium provided in this embodiment can implement the slurry pipeline flow velocity detection method provided in Example 1, and will not be described again here to avoid repetition.

[0164] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limiting, and thus other examples of the exemplary embodiments may have different values.

[0165] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0166] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A method for detecting flow velocity in a slurry pipeline, characterized in that: include: Acquire multiple turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and construct a feature engineering sample set corresponding to each sensor position according to the multiple turbulent vibration signals corresponding to each sensor position; Perform flow velocity calibration according to each of the characteristic engineering sample sets to obtain multiple flow velocity sample sets; The gradient boosting decision tree model is trained according to each of the feature engineering sample sets and each of the flow velocity sample sets to obtain a target prediction model corresponding to each of the sensor positions; Acquire the turbulent vibration signal to be detected corresponding to each of the sensor positions, and obtain the feature engineering sequence corresponding to each of the sensor positions according to each of the turbulent vibration signals to be detected; Each of the feature engineering sequences is predicted by each of the target prediction models to obtain a flow velocity prediction parameter corresponding to each of the sensor positions; The target flow rate detection parameters are obtained according to the flow rate weights and flow rate prediction parameters corresponding to each of the sensor positions.

2. The method for detecting flow velocity in a slurry pipeline according to claim 1, characterized in that: The method of acquiring a plurality of turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline and constructing a feature engineering sample set corresponding to each sensor position according to each turbulent vibration signal includes: For each of the sensor positions, based on a preset time interval, a plurality of turbulent vibration signals at a plurality of slurry flow rates are acquired within a preset acquisition time period to obtain the turbulent vibration signal set; The turbulent vibration signal set is subjected to dimension reduction to obtain the feature engineering sample set.

3. The method for detecting flow velocity in a slurry pipeline according to claim 2, characterized in that: The step of reducing the dimension of the turbulent vibration signal set to obtain the feature engineering sample set includes: Filtering the turbulent vibration signal set based on a moving average filtering algorithm; The filtered turbulent vibration signal set is subjected to dimension reduction to obtain the feature engineering sample set.

4. The method for detecting flow velocity in a slurry pipeline according to claim 1, characterized in that: The gradient boosting decision tree model is trained according to each of the feature engineering sample sets and each of the flow rate sample sets to obtain a target prediction model corresponding to each of the sensor positions, including: For each of the sensor positions, determining a training sample set and a test sample set according to the feature engineering sample set and the flow velocity sample set; Based on K-fold cross validation, the gradient boosting decision tree model is trained according to the training sample set to obtain K candidate prediction models, and a first prediction model is obtained from the K candidate prediction models based on preset model indicators; Based on the preset model indicators, model tests are performed on each of the candidate prediction models according to the test sample set to obtain a second prediction model; If the first prediction model and the second prediction model are the same, the first prediction model or the second prediction model is used as the target prediction model.

5. The method for detecting flow velocity in a slurry pipeline according to claim 4, characterized in that: The method further comprises: If the first prediction model and the second prediction model are different, the model parameters of the gradient boosting decision tree model are adjusted, and the step of re-executing the K-fold cross-validation to train the gradient boosting decision tree model according to the training sample set to obtain K candidate prediction models, and obtaining the first prediction model from the K candidate prediction models based on preset model indicators.

6. The method for detecting flow velocity in a slurry pipeline according to claim 2, characterized in that: The step of acquiring the turbulent vibration signal to be detected corresponding to each of the sensor positions, and obtaining the feature engineering sequence corresponding to each of the sensor positions according to each of the turbulent vibration signals to be detected, comprises: For each of the sensor positions, based on the preset time interval, the turbulent vibration signal to be detected is acquired to obtain a sequence to be detected; The dimension of the sequence to be detected is reduced to obtain the feature engineering sequence.

7. The method for detecting flow velocity in a slurry pipeline according to claim 6, characterized in that: The step of reducing the dimension of the sequence to be detected to obtain the feature engineering sequence includes: Filtering the sequence to be detected based on a moving average filtering algorithm; The filtered sequence to be detected is subjected to dimension reduction to obtain the feature engineering sequence.

8. A slurry pipeline flow velocity detection device, characterized in that: include: A first acquisition module is used to acquire a plurality of turbulent vibration signals corresponding to each sensor position on the wall of the slurry pipeline, and to construct a feature engineering sample set corresponding to each sensor position according to the plurality of turbulent vibration signals corresponding to each sensor position; A calibration module, used for performing flow velocity calibration according to each of the feature engineering sample sets to obtain multiple flow velocity sample sets; A training module, used to train the gradient boosting decision tree model according to each of the feature engineering sample sets and each of the flow velocity sample sets, to obtain a target prediction model corresponding to each of the sensor positions; A second acquisition module is used to acquire the turbulent vibration signal to be detected corresponding to each of the sensor positions, and obtain a feature engineering sequence corresponding to each of the sensor positions according to each of the turbulent vibration signals to be detected; A prediction module, used to predict each of the feature engineering sequences respectively through each of the target prediction models to obtain a flow velocity prediction parameter corresponding to each of the sensor positions; The detection module is used to obtain the target flow rate detection parameter according to the flow rate weight and flow rate prediction parameter corresponding to each of the sensor positions.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method for detecting flow velocity in a slurry pipeline according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: It stores a computer program, and when the computer program is executed by a processor, the method for detecting flow velocity in a slurry pipeline according to any one of claims 1 to 7 is implemented.

Images