Intelligent Recognition Method for Typical Excavator Operating Conditions Based on HBA-KNN
Through the HBA-KNN method, considering the pressure signal characteristics of the two pumps of the hydraulic excavator, the KNN algorithm is optimized, and the accuracy and accuracy of the excavator operating conditions in the existing technology are solved, and efficient operating conditions are achieved Identification and segmented energy-saving control.
Patent Information
- Application Number
- CN202210376488.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-11
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-04-11
AI Technical Summary
In the prior art, in the excavator operating conditions identification, the pressure relationship between the two pumps was not considered based on the main pump pressure, and the SVM identification classification algorithm used was low accuracy, which could not effectively support segmented energy-saving control and construction site supervision. .
Using the HBA-KNN method, the two pump outlet pressure signals of the hydraulic excavator are collected, wavelet noise reduction processing and time-frequency domain feature extraction are performed, and the KNN algorithm is optimized to identify the excavator operating conditions.
It improves the accuracy and accuracy of identification, can quickly and accurately identify the working conditions of the excavator, and provides important guarantees for construction site supervision and segmented energy-saving control.
Smart Images

Figure CN114841197B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of engineering machinery, and in particular to an intelligent identification method for typical working conditions of an excavator based on HBA-KNN. Background Art
[0002] As a typical construction machinery, excavators are widely used in various engineering projects due to their flexible operation and strong adaptability. Due to the two-way demand for high completion efficiency and low energy consumption, the related research on excavator operation fee statistics, construction site monitoring, excavator power matching and segmented energy-saving control has been promoted. Among them, the high-precision identification of typical operating conditions of hydraulic excavators is an important guarantee for the above research.
[0003] At present, the most direct method for identifying the working conditions of excavators is manual on-site identification, but this is a time-consuming and labor-intensive method, and the reliability of the results is not high. Therefore, it is of great theoretical and practical significance to accurately identify them through artificial intelligence technology. The current technology can be mainly divided into two aspects: computer vision and sensor technology according to the objects to be identified and the technical means of implementation. Although the identification of the working conditions of the excavator through computer vision uses some more advanced machine learning and deep learning algorithms, it cannot avoid problems such as occlusion, dark light and proportional deviation, and the amount of calculation is large, and the identification results cannot be directly used for the signals of segmented energy-saving control; secondly, there are many mature studies on sensor technology, but some identification objects are difficult to obtain, and the identification results cannot be used for the signals of segmented energy-saving control. However, based on the identification of the main pump pressure signal, the above problems can be avoided, and the characteristics of different working conditions are obviously different. The main pump pressure is also directly related to the load. For the power matching control and segmented energy-saving control of the engine-pump, the pump pressure signal is also more representative.
[0004] However, the existing research based on the main pump pressure is not mature enough. The pressure relationship between the two pumps (i.e., pressure difference) is not considered when extracting features. In addition, the recognition and classification algorithms are all based on SVM, with low accuracy. The recognition results are inaccurate and cannot perform intelligent recognition of the working conditions of the excavator very well. Summary of the invention
[0005] In order to solve the shortcomings of the above-mentioned prior art, the present invention provides an intelligent identification method for typical operating conditions of excavators based on HBA-KNN, which proposes a new intelligent identification method for typical operating conditions of excavators, based on the characteristics that the main pump pressure signal and the pressure difference signal of the two pumps have obvious characteristic differences between different operating conditions, and uses the K nearest neighbor optimization algorithm (KNN) optimized by the honey badger algorithm (HBA) to identify the operating conditions. It can ensure the accuracy of recognition and the accuracy of the results on the basis of replacing manual recognition, and has fast recognition efficiency and high accuracy of recognition results, thereby providing important guarantee for construction site supervision and segmented energy-saving control.
[0006] Specifically, the present invention provides an intelligent identification method for typical working conditions of an excavator based on HBA-KNN, which comprises the following steps:
[0007] S1, respectively collecting pressure signals at two pump outlets of the hydraulic excavator;
[0008] S2, preprocessing the pressure signals at the outlets of the two pumps and the pressure difference data of the two pumps collected in step S1, specifically including the following sub-steps:
[0009] S21, using wavelet threshold denoising to perform denoising filtering on the outlet pressure signal data collected in step S1, with the noise reduction ratio SNR as large as possible and the root mean square error RMSE as small as possible as constraints, selecting the wavelet basis function that meets the constraints for denoising;
[0010] The specific calculation formulas for signal-to-noise ratio SNR and root mean square error RMSE are as follows:
[0011]
[0012]
[0013] Wherein, b is the pump pressure signal before noise reduction, b1 is the pump pressure signal after noise reduction, n is the number of samples, i is the sample individual, and lg is the logarithmic operator;
[0014] S22, sampling the pressure signal data after noise reduction in step S21 in real time according to the set sample size and update rate, to obtain n groups of pressure signals of two pumps and corresponding pressure difference data. The update rate formula is as follows:
[0015]
[0016] Where η is the sample update rate, ΔT is the sample size, ΔT′1 is the end time of the first sample, and ΔT′2 is the end time of the second sample;
[0017] S23, extracting time-frequency domain features from the n groups of data sampled in step S22, specifically comprising the following sub-steps:
[0018] S231, extracting the time domain characteristic signal of the sampling data in step S22, extracting a total of 7 dimensioned indicators and 4 dimensionless indicators, as follows:
[0019] Dimensional indices f1-f7:
[0020] Maximum value: f1 = max(b1); Minimum value: f2 = min(b1);
[0021] Mean: Peak-to-peak value: f4 = f1-f2;
[0022] Square root amplitude: variance:
[0023] Mean Square Value:
[0024] Dimensionless index f8-f11:
[0025] Kurtosis: Pulse Factor:
[0026] Form Factor: Margin Factor:
[0027] In the formula, n is the number of samples, i is the sample individual, and b1 is the pump pressure signal after noise reduction;
[0028] S232, perform fast Fourier transform on the sampled data of step S22, and extract frequency domain characteristic signals, and extract 4 frequency domain characteristic signals f12-f15 in total, as follows:
[0029] Average frequency: Center of gravity frequency:
[0030] Frequency RMS: Frequency variance:
[0031] Where K is the number of spectral lines in the amplitude spectrum; s k is the amplitude spectrum, k = 1, 2, ... K; f k is the frequency value of the kth amplitude spectrum line;
[0032] S24, performing PCA dimensionality reduction on the time-frequency domain feature signal extracted in step S23;
[0033] S25, dividing the dimension reduction data set obtained in step S24 into a training set and a test set in a ratio of 7:3, and performing normalization processing;
[0034] S3, building an HBA-KNN classification model, taking the feature data obtained in step S25 as input, and performing identification and judgment of typical working conditions of the excavator, which specifically includes the following sub-steps:
[0035] S31, using the training data sets with different sample sizes and update rates obtained in step S25 to optimize the k value of KNN using the HBA algorithm, specifically including the following sub-steps:
[0036] S311, initialize the honey badger population size N and k value, and define the maximum number of iterations t max , the specific formula for each k value is as follows:
[0037] k i =lk i +r1×(uk i -lk i )
[0038] In the formula, k i is the position of the ith honey badger, i.e., the ith k value of the present invention, r1 is a random number between 0 and 1, uk i is the upper limit of the k value to be searched, lk i is the lower limit of the k value to be searched;
[0039] S312, the initial k value obtained in step S311 is simulated and verified one by one through the training set, and the k value corresponding to the highest accuracy is obtained, i.e., k max , and define the search intensity. The specific formula is as follows:
[0040]
[0041] S=(k i -k i+1 ) 2
[0042] d i =k max -k i
[0043] In the formula, I i is the search intensity, r2 is a random number between 0 and 1, S is the k-value concentration, k i is the position of the i-th k value, k i+1 is the position of the i+1th k value, d i is the i-th k value and k max The distance, k max The position of the k value corresponding to the highest accuracy;
[0044] S313. Update the density factor α. The specific formula is as follows:
[0045]
[0046] In the formula, C is a constant greater than 1, the default value is 2, e is the Euler number, t is the number of training iterations, and t max is the maximum number of training iterations;
[0047] S314, update k by mining or following the bee bird max The formulas for jumping out of the local optimum, finding the global optimum, and exploring the two ways of following the bee-bird are as follows:
[0048] k d =k max +F×β×I×k max +F×r3×α×d i
[0049] ×|(cos 2πr4)×[1-cos(2πr5)]|
[0050] k h =k max +F×r6×α×d i
[0051]
[0052] In the formula, k d The k value updated for the mining search part, k h The k value is updated to follow the search part of the bee-bird. β represents the search ability and is 6 by default. I is the search intensity. F is the search direction. r3, r4, r5, and r6 are random numbers between 0 and 1 respectively.
[0053] S315, through step S314, continuously search for a k value with a higher accuracy, and perform t max After iterations, multiple k global optimal values are obtained. max ;
[0054] S32, using the k obtained by HBA optimization in step S315 max The KNN is used to identify and predict the typical working conditions of the excavator on the test set obtained by S25, which specifically includes the following sub-steps:
[0055] S321. Use the Euclidean distance to calculate the similarity between the test data and the training sample. The specific formula is as follows:
[0056]
[0057] In the formula, d is the distance between two data points x and y, and x and y are two data points respectively;
[0058] S322, according to k obtained in step S315 max value, extract the k values arranged from large to small in similarity max d value;
[0059] S323, the top k extracted by detection max The similarity is calculated, and the proportion of each working condition category is calculated. It is predicted that the working condition with the largest proportion is the typical working condition of the excavator. The specific calculation method is as follows:
[0060] output=max(k1,k2,k3,k4)
[0061] In the formula, output is the corresponding working condition category obtained by testing the test data, max is the maximum value symbol, and k1~4 represent the first k max The number of each of the four working conditions in each similarity.
[0062] Preferably, in step S25, the input features of the training set and the test set are normalized to the range of 0 to 1. The specific formula is as follows:
[0063]
[0064] In the formula, f * is the normalized result, f is the main pump pressure data feature, f min is the minimum characteristic value of the main pump pressure data, f max It is the maximum value of the main pump pressure data characteristics.
[0065] Preferably, in step S32, typical working conditions of the excavator include an excavation condition, a lifting and rotating condition, an unloading condition and an empty bucket return condition.
[0066] Preferably, in step S21, the wavelet basis function is coif4, including 3 wavelet decomposition layers, the threshold function is a soft threshold, and the threshold calculation method is a maximum and minimum threshold method.
[0067] Preferably, in step S21, the final values of the signal-to-noise ratio SNR and the root mean square error RMSE are the average values of the two pump pressure noise reduction evaluation indicators.
[0068] Preferably, in step S24, when the PCA dimensionality reduction process is performed on the initial features, when the cumulative contribution rate of the first several principal components of any one of the four working conditions reaches more than 85%, the first several principal components are taken as the output results.
[0069] Compared with the prior art, the present invention has the following beneficial effects:
[0070] (1) The present invention performs wavelet noise reduction filtering on the extracted main pump pressure data during data processing, thereby reducing the interference of vibration, electromagnetic and other factors on the pressure, ensuring the accuracy of the collected pressure data, and laying the foundation for the subsequent recognition process.
[0071] (2) When extracting data features, the present invention not only extracts the main pump pressure data features of the two pumps, but also considers the influence of the pressure relationship between the two pumps, namely the pressure difference, and simultaneously extracts the main pump pressure difference signal features to ensure the integrity of the signal features.
[0072] (3) Based on the fact that the main pump pressure signal and the pressure difference signals of the two pumps have obvious characteristic differences under different working conditions, the present invention uses the K nearest neighbor optimization algorithm (KNN) optimized by the honey badger algorithm (HBA) to identify the working conditions. It can ensure the accuracy of recognition and the accuracy of the results on the basis of replacing manual recognition, with fast recognition efficiency and high accuracy of recognition results, which can provide important guarantee for construction site supervision and segmented energy-saving control.
[0073] (4) The present invention can identify four typical working conditions of hydraulic excavators, namely, excavation, lifting and rotation, unloading and empty bucket return, through the proposed identification method, so as to perform segmented energy-saving control of the excavator or construction efficiency supervision, and can be well applied in engineering practice. BRIEF DESCRIPTION OF THE DRAWINGS
[0074] Figure 1 It is a schematic diagram of the process of the present invention;
[0075] Figure 2 It is a flow chart of the method for identifying typical working conditions of an excavator based on HBA-KNN of the present invention;
[0076] Figure 3 This is a wavelet noise reduction principle diagram for noise reduction filtering of the main pump pressure of the present invention;
[0077] Figure 4 This is a noise reduction effect diagram of different wavelet basis functions of the present invention;
[0078] Figure 5 It is a statistical diagram of the cumulative contribution rate of the principal elements of PCA dimension reduction under four working conditions of the present invention;
[0079] Figure 6 It is a schematic diagram of the working condition identification model HBA-KNN of the present invention;
[0080] Figure 7 This is a fitness curve diagram of optimizing KNN using HBA in the present invention;
[0081] Figure 8 This is a diagram of the working condition identification result of the present invention. DETAILED DESCRIPTION
[0082] Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0083] The present invention provides an intelligent method suitable for identifying typical load conditions of hydraulic excavators. The method identifies four typical conditions of hydraulic excavators, namely, digging, lifting and rotating, unloading, and empty bucket return, so as to perform segmented energy-saving control on the excavator or supervise construction efficiency.
[0084] The method as a whole includes three modules: data collection, data preprocessing, and working condition identification. The data collection mainly collects the pressure signals of the two main pumps of the excavator, and obtains the pressure difference signals of the two pumps at the same time. Data preprocessing includes wavelet denoising of the main pump pressure, extracting pressure data segments based on sample size and update rate, extracting the time-frequency domain features of the two main pump pressures and pressure differences, and performing PCA dimension reduction on the extracted time-frequency domain feature signals. The working condition identification module includes dividing the preprocessed data into training sets and test sets in proportion and normalizing them. After optimizing the K value of KNN through the Honey Badger Algorithm (HBA), the optimized K value is used to identify the working condition.
[0085] Specifically, Figure 1 and Figure 2 As shown, the present invention provides an intelligent identification method for typical working conditions of an excavator based on HBA-KNN, which specifically includes the following steps:
[0086] S1. Collect the pressure data of the two main pumps of the hydraulic excavator through the pressure sensor, and calculate the pressure difference signal of the two main pumps.
[0087] S2, by performing wavelet denoising filtering on the main pump pressure data collected by S1, and then intercepting the pressure data segment according to a certain sample size and update rate for the denoised data, and then extracting the time-frequency domain features of the intercepted pressure data segment and performing PCA dimensionality reduction processing.
[0088] S3, divide the data features processed by S2 into training sets and test sets according to a certain ratio and normalize them, then use HBA to optimize the parameters, combine the obtained optimal K value, and use KNN to identify the working conditions.
[0089] In step S2, the wavelet denoising algorithm is used to denoise the collected data. The specific principle is as follows: Figure 3 As shown. In the specific operation, different wavelet basis functions are used for noise reduction. Finally, according to the evaluation index, the larger the noise reduction ratio (SNR), the smaller the root mean square error RMSE, and the better the noise reduction effect, it is concluded that coif4 has the best effect, and coif4 is selected to perform noise reduction filtering on the main pump pressure. Specifically, the noise reduction performance indicators of different wavelet basis functions are as follows: Figure 4As shown. The principal component analysis (PCA) dimensionality reduction is performed on the extracted time-frequency domain features. If the cumulative contribution rate of the first few principal components of any of the four typical working conditions after PCA processing reaches more than 85%, the first few principal components are taken for analysis. Finally, it is verified that the cumulative contribution rate of the first five principal components can basically reach more than 85%, so the first five principal components are selected as the input of the working condition identification model, as shown in Figure 5 shown.
[0090] In step S3, the preprocessed data features in S2 are first divided into a training set and a test set in a ratio of 7:3, and then normalized to between 0 and 1 according to the same principle. Finally, the HBA-KNN model is used for working condition identification, and the specific typical working condition of the excavator is input. The schematic diagram is as follows Figure 6 As shown in the figure, HBA is used to optimize the K value of the hyperparameter K of KNN, and find the parameter K with the best generalization performance at a faster speed. The optimized fitness curve is shown in the figure. Figure 7 As shown. Finally, KNN uses the best k value obtained by HBA optimization to identify the working conditions of the input features and output the typical working conditions of the specific hydraulic excavator. The results of the final test set are as follows Figure 8 shown.
[0091] The following is a further description of an intelligent identification method for typical working conditions of an excavator according to the present invention in combination with an embodiment:
[0092] S1, respectively collecting the outlet pressures of two pumps of the hydraulic excavator through pressure sensors;
[0093] S2, pre-process the two pump pressure signals and their pressure difference data collected by S1:
[0094] S21, use wavelet threshold denoising to perform denoising filtering on the data collected by S1. According to the principle that the larger the noise reduction ratio SNR and the smaller the RMSE, the better the noise reduction, the wavelet basis function with the best effect is selected to perform denoising. The specific formulas of signal-to-noise ratio SNR and root mean square error RMSE are as follows:
[0095]
[0096]
[0097] Where b is the pump pressure signal before noise reduction, b1 is the pump pressure signal after noise reduction, n is the number of samples, i is the sample individual, and lg is the logarithmic operator.
[0098] S22, sampling the pressure signal data after noise reduction in step S21 in real time according to the set sample size and update rate, to obtain n groups of pressure signals of two pumps and corresponding pressure difference data, wherein the calculation formula of the update rate is as follows:
[0099]
[0100] Where η is the sample update rate, ΔT is the sample size, ΔT′1 is the end time of one sample, and ΔT′2 is the end time of the second sample;
[0101] S23, extract the time-frequency domain features of the pressure and pressure difference of the two pumps from the 60 sets of data sampled in S22:
[0102] S231, extract the time domain characteristic signal of the sampling data of S22, and extract a total of 7 dimensional indicators and 4 dimensionless indicators, as follows:
[0103] Dimensional indicators:
[0104] Maximum value: f1 = max(b1); Minimum value: f2 = min(b1);
[0105] Mean: Peak-to-peak value: f4 = f1-f2;
[0106] Square root amplitude: variance:
[0107] Mean Square Value:
[0108] Dimensionless index:
[0109] Kurtosis: Pulse Factor:
[0110] Form Factor: Margin Factor:
[0111] Where n is the number of samples, i is the sample individual, and b1 is the pump pressure after noise reduction.
[0112] S232, perform fast Fourier transform on the S22 sampling data, and extract its frequency domain characteristic signals, a total of 4 are extracted, as follows:
[0113] Average frequency: Center of gravity frequency:
[0114] Frequency RMS: Frequency variance:
[0115] Where K is the number of spectral lines in the amplitude spectrum; s k is the amplitude spectrum, k = 1, 2, ... K; f k is the frequency value of the kth amplitude spectrum line.
[0116] S24, performing PCA dimension reduction on a total of 45 time-frequency domain feature signals of the pressure and pressure difference of the two pumps extracted in S23;
[0117] S25, dividing the dimension reduction data set obtained in S24 into a training set and a test set in a ratio of 7:3, and performing normalization processing;
[0118] S3: Build an HBA-KNN classification model, use the feature data processed by S25 as input, and identify and judge the typical working conditions of the excavator:
[0119] S31, using the training data sets with different sample sizes and update rates obtained in S25 to optimize the k value of KNN using the HBA algorithm;
[0120] S311, initialize the population size N of honey badgers and their respective locations, and define the maximum number of iterations t for optimization max , the specific formula for the position of each honey badger is as follows:
[0121] x i =lb i +r1×(uk i -lk i )
[0122] In the formula, x i is the position of the ith honey badger, i.e., the ith k value of the present invention, r1 is a random number between 0 and 1, uk i is the upper limit of the k value to be searched, lk i The lower limit of the k value to search.
[0123] S312, the initial k values obtained in S311 are simulated and verified one by one through the training set, and the k value position k corresponding to the highest accuracy is obtained. max , and define the search intensity. The specific formula is as follows:
[0124]
[0125] S=(k i -k i+1 ) 2
[0126] d i =k max -k i
[0127] In the formula, I i is the search intensity, r2 is a random number between 0 and 1, S is the concentration of honey badgers, and k i is the position of the i-th k value, k i+1 is the position of the i+1th k value, d iis the i-th k value and k max The distance, k max is the position of the k value corresponding to the highest accuracy.
[0128] S313. Update the density factor. The specific formula is as follows:
[0129]
[0130] In the formula, C is a constant greater than 1, the default value is 2, e is the Euler number, t is the number of iterations, and t max is the maximum number of iterations.
[0131] S314, update k by mining or following the bee bird max The formulas for jumping out of the local optimum, finding the global optimum, and exploring the two ways of following the bee-bird are as follows:
[0132] k d =k max +F×β×I×k max +F×r3×α×d i
[0133] ×|(cos 2πr4)×[1-cos(2πr5)]|
[0134] k h =k max +F×r6×α×d i
[0135] In the formula, k d The k value updated for the mining search part, k h The k value is updated to follow the honey badger search part, F is the search direction flag, β is the search ability of the honey badger, the default value is 6, and r3, r4, r5, and r6 are random numbers between 0 and 1.
[0136] S315, through S314, continuously search for a k value with a higher accuracy, t max After iterations, the global optimal value k is obtained. max ;
[0137] S32, using the k obtained by HBA optimization in S315 max The KNN is used to identify and predict the typical working conditions of the excavator on the test set obtained by S25. The specific steps are as follows:
[0138] S321. Use the Euclidean distance to calculate the similarity between the test data and the training sample. The specific formula is as follows:
[0139]
[0140] In the formula, d is the distance between two data points x and y, and x and y are two data points respectively;
[0141] S322, according to k obtained in step S315 max Value, extract the top k similarities arranged from large to small max d value;
[0142] S323, the top k extracted by detection max The similarity is calculated, and the proportion of each working condition category is calculated. It is predicted that the working condition with the largest proportion is the typical working condition of the excavator. The specific calculation method is as follows:
[0143] output=max(k1,k2,k3,k4)
[0144] In the formula, output is the corresponding working condition category obtained by testing the test data, max is the maximum value symbol, and k1~4 represent the first k max The number of each of the four working conditions in each similarity.
[0145] In the above identification method, the hyperparameters of KNN are optimized through HBA, and the best K value is used to identify the working condition. Figure 7 It can be seen that HBA found the global optimal K value in the first iteration, and the accuracy of working condition recognition was 98.6111%. Figure 8 In the final analysis, the accuracy rates of the four working conditions corresponding to KNN recognition, namely excavation, lifting and rotation, unloading, and empty bucket return, were 100%, 98.75%, 100%, and 100%, respectively. Only in the lifting and rotation working condition, one was identified as the unloading working condition, and the other data features were all identified correctly, ensuring a very high accuracy rate.
[0146] In the existing related research, the recognition results based on computer vision cannot be directly used for segmented energy-saving control. The recognition results based on multi-sensor information fusion technology are all based on the support vector machine algorithm (SVM) for working condition recognition, and the correctness and accuracy are not high. The recognition method of the present invention solves the problems existing in the prior art very well. First, in terms of data processing: wavelet noise reduction is performed; the update rate is used for real-time sampling and updating of the main pump pressure data; secondly, in terms of classification and recognition: the recognition object is not only the characteristics of the two pumps, but also the characteristics of the relationship between the pressures of the two pumps, that is, the pressure difference; in terms of the algorithm, the honey badger optimization algorithm (HBA) is used to optimize the classification algorithm (KNN). In this way, the best model can be obtained more quickly, and the working condition can be identified with fast recognition speed and high accuracy. The recognition results can be directly used as the input signal of the segmented energy-saving control system, and the construction party can also be supervised in real time, which provides an important guarantee for the related research on construction site supervision and segmented energy-saving control.
[0147] The embodiments described above are only descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.
Claims
1. An intelligent identification method for typical working conditions of an excavator based on HBA-KNN, characterized by: It includes the following steps: S1, respectively collecting pressure signals at two pump outlets of the hydraulic excavator; S2, preprocessing the pressure signals at the outlets of the two pumps and the pressure difference data of the two pumps collected in step S1, specifically including the following sub-steps: S21, using wavelet threshold denoising to perform denoising filtering on the outlet pressure signal data collected in step S1, with the noise reduction ratio SNR as large as possible and the root mean square error RMSE as small as possible as constraints, selecting the wavelet basis function that meets the constraints for denoising; The specific calculation formulas for signal-to-noise ratio SNR and root mean square error RMSE are as follows: Wherein, b is the pump pressure signal before noise reduction, b1 is the pump pressure signal after noise reduction, n is the number of samples, i is the sample individual, and lg is the logarithmic operator; S22, sampling the pressure signal data after noise reduction in step S21 in real time according to the set sample size and update rate, to obtain n groups of pressure signals of two pumps and corresponding pressure difference data. The update rate formula is as follows: Where η is the sample update rate, ΔT is the sample size, ΔT′1 is the end time of the first sample, and ΔT′2 is the end time of the second sample; S23, extracting time-frequency domain features from the n groups of data sampled in step S22, specifically comprising the following sub-steps: S231, extracting the time domain characteristic signal of the sampling data in step S22, extracting a total of 7 dimensioned indicators and 4 dimensionless indicators, as follows: Dimensional indexes f1-f7: Maximum value: f1 = max(b1); Minimum value: f2 = min(b1); Mean: Peak-to-peak value: f4 = f1-f2; Square root amplitude: variance: Mean Square Value: Dimensionless index f8-f11: Kurtosis: Pulse Factor: Form Factor: Margin Factor: In the formula, n is the number of samples, i is the sample individual, and b1 is the pump pressure signal after noise reduction; S232, perform fast Fourier transform on the sampled data of step S22, and extract frequency domain characteristic signals, and extract 4 frequency domain characteristic signals f12-f15 in total, as follows: Average frequency: Center of gravity frequency: Frequency RMS: Frequency variance: Where K is the number of spectral lines in the amplitude spectrum; s k is the amplitude spectrum, k = 1, 2, ... K; f k is the frequency value of the kth amplitude spectrum line; S24, performing PCA dimensionality reduction on the time-frequency domain feature signal extracted in step S23; S25, dividing the dimension reduction data set obtained in step S24 into a training set and a test set in a ratio of 7:3, and performing normalization processing; S3, building an HBA-KNN classification model, taking the feature data obtained in step S25 as input, and performing identification and judgment of typical working conditions of the excavator, which specifically includes the following sub-steps: S31, using the training data sets with different sample sizes and update rates obtained in step S25 to optimize the k value of KNN using the HBA algorithm, specifically including the following sub-steps: S311, initialize the honey badger population size N and k value, and define the maximum number of iterations t max , the specific formulas for each k value are as follows: k i =lk i +r1×(uk i -lk i ) In the formula, k i is the position of the ith honey badger, i.e., the ith k value, r1 is a random number between 0 and 1, uk i is the upper limit of the k value to be searched, lk i is the lower limit of the k value to be searched; S312, the initial k value obtained in step S311 is simulated and verified one by one through the training set, and the k value corresponding to the highest accuracy is obtained, i.e., k max , and define the search intensity. The specific formula is as follows: S=(k i -k i+1 ) 2 d i =k max -k i In the formula, I i is the search intensity, r2 is a random number between 0 and 1, S is the k-value concentration, k i is the position of the i-th k value, k i+1 is the position of the i+1th k value, d i is the i-th k value and k max The distance, k max The position of the k value corresponding to the highest accuracy; S313. Update the density factor α. The specific formula is as follows: In the formula, C is a constant greater than 1, the default value is 2, e is the Euler number, t is the number of training iterations, and t max is the maximum number of training iterations; S314, update k by mining or following the bee bird max The formulas for jumping out of the local optimum, finding the global optimum, and exploring the two ways of following the bee-bird are as follows: k d =k max +F×β×I×k max +F×r3×α×d i ×|(cos 2πr4)×[1-cos(2πr5)]| k h =k max +F×r6×α×d i In the formula, k d The k value updated for the mining search part, k h The k value is updated to follow the search part of the bee-bird. β represents the search ability and is 6 by default. I is the search intensity. F is the search direction. r3, r4, r5, and r6 are random numbers between 0 and 1 respectively. S315, through step S314, continuously search for a k value with a higher accuracy, and perform t max After iterations, multiple k global optimal values are obtained. max ; S32, using the k obtained by HBA optimization in step S315 max The KNN is used to identify and predict the typical working conditions of the excavator on the test set obtained by S25, which specifically includes the following sub-steps: S321. Use the Euclidean distance to calculate the similarity between the test data and the training sample. The specific formula is as follows: In the formula, d is the distance between two data points x and y, and x and y are two data points respectively; S322, according to k obtained in step S315 max value, extract the k values arranged from large to small in similarity max d value; S323, the top k extracted by detection max The similarity is calculated, and the proportion of each working condition category is calculated. It is predicted that the working condition with the largest proportion is the typical working condition of the excavator. The specific calculation method is as follows: output=max(k1,k2,k3,k4) In the formula, output is the corresponding working condition category obtained by testing the test data, max is the maximum value symbol, and k1~4 represent the first k max The number of each of the four working conditions in each similarity.
2. The method for intelligently identifying typical working conditions of an excavator based on HBA-KNN according to claim 1 is characterized in that: In step S25, the input features of the training set and the test set are normalized to the range of 0 to 1. The specific formula is as follows: In the formula, f * is the normalized result, f is the main pump pressure data feature, f min is the minimum characteristic value of the main pump pressure data, f max It is the maximum value of the main pump pressure data characteristics.
3. The method for intelligently identifying typical working conditions of an excavator based on HBA-KNN according to claim 1 is characterized in that: In step S32, typical working conditions of the excavator include excavation condition, lifting and rotating condition, unloading condition and empty bucket return condition.
4. The method for intelligently identifying typical working conditions of an excavator based on HBA-KNN according to claim 1 is characterized in that: In step S21, the wavelet basis function is coif4, which includes 3 wavelet decomposition layers, the threshold function is a soft threshold, and the threshold calculation method is the maximum and minimum threshold method.
5. The method for intelligently identifying typical working conditions of an excavator based on HBA-KNN according to claim 4 is characterized in that: In step S21, the final values of the signal-to-noise ratio SNR and the root mean square error RMSE are the average values of the two pump pressure noise reduction evaluation indicators.
6. The method for intelligently identifying typical working conditions of an excavator based on HBA-KNN according to claim 1 is characterized in that: In step S24, when the PCA dimensionality reduction process is performed on the initial features, when the cumulative contribution rate of the first several principal components of any one of the four working conditions reaches more than 85%, the first several principal components are taken as the output results.
