Elevator safety protection method and system based on big data

Through big data analysis of elevator operation sounds, and using technologies such as bandpass filtering and convolutional neural networks, the elevator operation risks are judged in real time, solving the problems of lag and one-sided judgment in traditional elevator maintenance, and improving maintenance accuracy and efficiency.

CN118965137AInactive Publication Date: 2024-11-15BEIJING SHICHUANG ELEVATOR ENG CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411042490.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional elevator maintenance and risk assessment rely on manual inspections, making it difficult to fully and accurately discover potential risks, resulting in lagging judgments and one-sidedness.

Method used

By monitoring the sound of elevator operation, using big data technology to perform bandpass filtering, time slice division, convolutional neural network endpoint detection, fast Fourier transform and MFCC feature extraction, calculate the sound wave feature spatial curve and Hausdorff distance, and judge the elevator operation risks in real time .

Benefits of technology

Real-time prediction and risk identification of elevator faults are realized, the accuracy and efficiency of maintenance work are improved, and maintenance costs are reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965137A_ABST
    Figure CN118965137A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of big data risk analysis, in particular to an elevator safety protection method and system based on big data, and the method comprises the steps: monitoring and reading elevator operation sound to obtain an elevator operation sound sequence, and carrying out the scribing of the elevator operation sound sequence through a band-pass filtering technology and a time slice division algorithm, performing feature extraction processing on the first interval sound slices to obtain an original MFCC feature vector set, extracting an energy feature set and a dynamic feature set according to the original MFCC feature vector set, and drawing a sound wave feature space curve of the energy feature and the dynamic feature on a coordinate axis through a mapping algorithm; the real-time operation risk of the elevator is judged by judging the relation value between the sound wave feature space curve and the Hausdorff distance of the preset risk sound wave feature space curve of the system, safety evaluation can be comprehensively and deeply conducted on the elevator, and the method has universality and good application prospects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of big data risk analysis, and in particular to an elevator safety protection method and system based on big data. Background Art

[0002] Traditional elevator maintenance and risk assessment mainly rely on manual inspections and regular testing. This method can only provide limited observations of the elevator's operating status and is difficult to fully and accurately detect potential risks.

[0003] As an important monitoring indicator, the sound of elevator operation can directly reflect the operating status and potential problems of the elevator. The introduction of sound monitoring technology can capture the sound generated by the elevator in real time and continuously during operation, so as to obtain more information about the elevator's operating status. By analyzing and processing massive sound data through big data technology, an elevator fault prediction model is established to determine whether the elevator has abnormal conditions such as abnormal noise and friction noise, so as to predict the risk of elevator failure, thereby improving the accuracy and efficiency of maintenance work. Summary of the invention

[0004] (1) Technical issues to be solved

[0005] The purpose of the present invention is to provide an elevator safety protection method and system based on big data to solve the lag and one-sidedness of elevator risk judgment and processing during elevator inspection.

[0006] (2) Technical solution

[0007] To achieve the above object, on the one hand, the present invention provides an elevator safety protection method based on big data, the method comprising:

[0008] An elevator operation sound sequence is obtained by monitoring and reading the elevator operation sound, and the elevator operation sound sequence is subjected to bandpass filtering technology to obtain a first enhanced sound sequence; the first enhanced sound sequence is sliced ​​by a time slice division algorithm to obtain a first interval sound slice, and the first interval sound slice is separated end-to-end by a convolutional neural network endpoint detection technology to obtain a second interval sound slice.

[0009] The second interval sound piece is subjected to the frame windowing method to obtain a sound signal sequence and a discrete power spectrum group is obtained by fast Fourier transform. The discrete power spectrum group is subjected to logarithmic operation to obtain a logarithmic power spectrum group and the logarithmic power spectrum is subjected to discrete cosine transform to obtain an original MFCC feature vector set; an energy feature set is extracted from the original MFCC feature vector set through an analysis and processing method, and a dynamic energy feature set is obtained by first-order difference calculation based on the energy feature set and recorded as a dynamic feature set.

[0010] The energy feature set and the dynamic feature set are divided into frames, and the mapping curves of the energy features and the dynamic features on the coordinate axis are drawn through a mapping algorithm and recorded as acoustic wave feature space curves. The Hausdorff distance between the acoustic wave feature space curve and the system preset risk acoustic wave feature space curve is calculated, and the real-time operation risk of the elevator is judged according to the relationship value between the Hausdorff distance and the threshold value preset by the system.

[0011] Furthermore, the method of slicing the first enhanced sound sequence by a time slice division algorithm to obtain the first interval sound slices includes:

[0012] The first enhanced sound sequence is divided into a combination sequence of N frequency segments according to the frequency range and recorded as the second enhanced sound sequence, and the energy of each frequency segment of the second enhanced sound sequence is obtained by the frequency domain energy calculation algorithm [E 1i , E 12 ,……,E 1i ] is recorded as the first signal energy sequence, and the frequency components [F1, F2, ..., F] in the first enhanced sound sequence are calculated using the first signal energy sequence according to the frequency energy conversion formula. k ] is called frequency sequence; the frequency energy conversion formula is:

[0013]

[0014] Where E i represents the energy of the ith frequency segment, w i represents the weight of the i-th frequency band, i = 1, 2, ..., N; E total Represents the total energy.

[0015] The signal energy sequence [E1, E2, ..., E3, ..., E4, ..., E5, ..., E6, ..., E7, ..., E8, ..., E9, ..., E10, ..., E110, ..., E12, ..., E13, ..., E14, ..., E15, ..., E16, ..., E17, ..., E18, ..., E19, ..., E20, ..., E21, ..., E22, ..., k ] is recorded as the second signal energy sequence; the time slice length L is calculated by the time slice function according to the second signal energy sequence and the frequency sequence k ; The time slice function is:

[0016]

[0017] Among them, α and β are weight coefficients, max(E) is the maximum value of the energy of the first enhanced sound sequence, and max(F) is the maximum value of the frequency sequence.

[0018] The first enhanced sound sequence is sliced ​​according to the time slice length using a slice formula to obtain a first interval sound slice sequence [T 11 , T 12 , T 13 , ..., T 1m].

[0019] The dicing formula is:

[0020] T 1m [m] = T[t k +m],0≤m <L k .

[0021] Furthermore, the method for obtaining the energy of each frequency segment of the second enhanced sound sequence by a frequency domain energy calculation algorithm and obtaining the signal energy sequence of each time point of the first enhanced sound sequence by a frequency domain energy calculation algorithm includes:

[0022] The first enhanced sound sequence and the second enhanced sound sequence are subjected to fast Fourier transform to obtain a frequency domain signal X(f), and the amplitude spectrum of the first enhanced sound sequence and the second enhanced sound sequence is calculated by an amplitude spectrum calculation formula; the amplitude spectrum calculation formula is:

[0023]

[0024] Where Re(X(f)) and Im(X(f)) are the real and imaginary parts of X(f), respectively.

[0025] The signal energy sequences of the first enhanced sound sequence and the second enhanced sound sequence are calculated by using the amplitude spectrum energy formula; the amplitude spectrum energy formula is:

[0026]

[0027] Where N is the total number of sampling points of the signal.

[0028] Furthermore, the original MFCC feature vector set is used to extract an energy feature set through an analysis and processing method, and a method for obtaining a dynamic energy feature set recorded as a dynamic feature set through first-order difference calculation according to the energy feature set includes:

[0029] The original MFCC feature vector set is expanded to obtain an MFCC feature matrix, a frame vector of the MFCC feature matrix is ​​obtained, and the energy feature of the frame vector is obtained according to an energy feature calculation formula.

[0030] The energy characteristic calculation formula is:

[0031]

[0032] Where E(t) is the energy of the tth frame, x i (t) is the i-th sample of the t-th frame, t=1, 2, ..., N; L is the number of samples per frame.

[0033] The dynamic energy characteristic is obtained according to the energy characteristic through the formula ΔE(t)=E(t+1)-E(t-1).

[0034] Furthermore, the method of dividing the energy feature set and the dynamic feature set into frames and drawing a mapping curve of the energy feature and the dynamic feature on the coordinate axis through a mapping algorithm includes:

[0035] The dynamic feature set is stacked column by column to obtain a feature matrix F of m×n: According to the column vectors F1, F2, ..., F n , through the formula Get the mean vector μ of each feature; According to the mean vector μ and the column vector of the matrix F, through the formula Calculate the standard deviation vector σ of each vector; according to the standard deviation vector σ, the mean vector μ and the column vector of the matrix F, through the formula Get the standardized data matrix F scaled ; According to the standardized data matrix F scaled , through the formula Obtain the covariance matrix C; perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalue λ i and the corresponding eigenvector υ i ; The feature vector υ i According to the characteristic value λ i The size of the eigenvectors is rearranged, and the first k eigenvectors are selected to form the projection matrix W and the formula F is used. pca =W T F scaled Get the mapping matrix F after mapping to the dynamic feature principal component space pca .

[0036] The mapping matrix F mapped to the dynamic feature principal component space is obtained according to the dynamic feature set. pca The mapping matrix X of the energy feature set to the energy feature principal component space is obtained by pca .

[0037] The mapping matrix F pca and the mapping matrix X pca A distance matrix is ​​obtained by calculating the Euclidean distance, and based on the distance matrix, the principal components of the dynamic feature set and the principal components of the energy feature set are mapped to a system preset space through an MDS algorithm to draw a mapping curve.

[0038] Based on the same inventive concept, on the other hand, the present invention also provides an elevator fault detection system based on big data, the system comprising:

[0039] The acoustic monitoring module is used to obtain an elevator operation sound sequence by monitoring and reading the elevator operation sound, and obtain a first enhanced sound sequence by bandpass filtering technology from the elevator operation sound sequence; the first enhanced sound sequence is sliced ​​by a time slice division algorithm to obtain a first interval sound slice, and the first interval sound slice is separated end-to-end by a convolutional neural network endpoint detection technology to obtain a second interval sound slice.

[0040] The spectrum processing module is connected to the acoustic monitoring module, and is used to obtain a sound signal sequence by a frame windowing method for the second interval sound piece and obtain a discrete power spectrum group by a fast Fourier transform, obtain a logarithmic power spectrum group by a logarithmic operation for the discrete power spectrum group and obtain an original MFCC feature vector set by a discrete cosine transform for the logarithmic power spectrum; extract an energy feature set from the original MFCC feature vector set by an analysis and processing method, and obtain a dynamic energy feature set by a first-order difference calculation based on the energy feature set, which is recorded as a dynamic feature set.

[0041] The matching recognition module is connected to the sound spectrum processing module, and is used to frame the energy feature set and the dynamic feature set and draw the mapping curve of the energy feature and the dynamic feature on the coordinate axis through a mapping algorithm, and record it as an acoustic wave feature space curve; by calculating the Hausdorff distance between the acoustic wave feature space curve and the system preset risk acoustic wave feature space curve, the real-time operation risk of the elevator is judged according to the relationship value between the Hausdorff distance and the threshold value preset by the system.

[0042] Furthermore, the acoustic monitoring module includes:

[0043] The frequency energy conversion module divides the first enhanced sound sequence into a combination sequence of N frequency segments according to the frequency range as a second enhanced sound sequence, and obtains the energy of each frequency segment of the second enhanced sound sequence through a frequency domain energy calculation algorithm [E 1i , E 12 ,……,E 1i ] is recorded as the first signal energy sequence, and the frequency components [F1, F2, ..., F] in the first enhanced sound sequence are calculated using the first signal energy sequence according to the frequency energy conversion formula. k ] is called frequency sequence; the frequency energy conversion formula is:

[0044]

[0045] Where Ei represents the energy of the i-th frequency band, w i represents the weight of the i-th frequency band, i = 1, 2, ..., N; E total Represents the total energy.

[0046] The time slice calculation module obtains the signal energy sequence [E1, E2, ..., E3, ..., E4, ..., E5, ..., E6, ..., E7, ..., E8, ..., E9, ..., E10, ..., E11, ..., E12, ..., E13, ..., E14, ..., E15, ..., E16, ..., E17, ..., E18, ..., E19, ..., E20, ..., E21, ..., k ] is recorded as the second signal energy sequence; the time slice length L is calculated by the time slice function according to the second signal energy sequence and the frequency sequence k ; The time slice function is:

[0047]

[0048] Among them, α and β are weight coefficients, max(E) is the maximum value of the energy of the first enhanced sound sequence, and max(F) is the maximum value of the frequency sequence.

[0049] A sound slice division module divides the first enhanced sound sequence into slices according to the time slice length using a slice division formula to obtain a first interval sound slice sequence [T 11 , T 12 , T 13 , ..., T 1m ]; the dicing formula is:

[0050] T 1m [m] = T[t k +m],0≤m <L k .

[0051] Furthermore, the frequency energy conversion module includes:

[0052] The frequency domain signal X(f) is obtained by performing fast Fourier transform on the first enhanced sound sequence and the second enhanced sound sequence, and the amplitude spectrum of the first enhanced sound sequence and the second enhanced sound sequence is calculated by an amplitude spectrum calculation formula; the amplitude spectrum calculation formula is:

[0053]

[0054] Where Re(X(f)) and Im(X(f)) are the real and imaginary parts of X(f), respectively.

[0055] The signal energy sequences of the first enhanced sound sequence and the second enhanced sound sequence are calculated by using the amplitude spectrum energy formula; the amplitude spectrum energy formula is:

[0056]

[0057] Where N is the total number of sampling points of the signal.

[0058] Furthermore, the matching identification module includes:

[0059] The feature extraction module obtains an MFCC feature matrix by expanding the original MFCC feature vector set, obtains a frame vector of the MFCC feature matrix, and obtains the energy feature of the frame vector according to an energy feature calculation formula; the energy feature calculation formula is:

[0060]

[0061] Where E(t) is the energy of the tth frame, x i (t) is the i-th sample of the t-th frame, t = 1, 2, ..., N; L is the number of samples per frame;

[0062] The dynamic energy characteristic is obtained according to the energy characteristic through the formula ΔE(t)=E(t+1)-E(t-1).

[0063] Furthermore, the matching identification module also includes:

[0064] The feature mapping module is connected to the feature extraction module, and obtains a feature matrix F of m×n by stacking the dynamic feature set in columns: According to the column vectors F1, F2, ..., F n , through the formula Get the mean vector μ of each feature; According to the mean vector μ and the column vector of the matrix F, through the formula Calculate the standard deviation vector σ of each vector; according to the standard deviation vector σ, the mean vector μ and the column vector of the matrix F, through the formula Get the standardized data matrix F scaled ; According to the standardized data matrix F scaled , through the formula Obtain the covariance matrix C; perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalue λ i and the corresponding eigenvector υ i ; The feature vector υ i According to the characteristic value λ i The size of the eigenvectors is rearranged, and the first k eigenvectors are selected to form the projection matrix W and the formula F is used. pca =W T F scaled Get the mapping matrix F after mapping to the dynamic feature principal component space pca .

[0065] The mapping matrix F mapped to the dynamic feature principal component space is obtained according to the dynamic feature set. pca The mapping matrix X of the energy feature set to the energy feature principal component space is obtained by pca .

[0066] The elevator risk analysis module is connected with the feature mapping module by pca and the mapping matrix X pca A distance matrix is ​​obtained by calculating the Euclidean distance, and based on the distance matrix, the principal components of the dynamic feature set and the principal components of the energy feature set are mapped to a system preset space through an MDS algorithm to draw a mapping curve.

[0067] (3) Beneficial effects

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1. Elevator failures are often accompanied by specific sound changes, such as unusual noise or vibration. By integrating advanced elevator operation sound monitoring technology and big data analysis, real-time tracking of elevator operation status and accurate identification of potential risks can be achieved, thus becoming an extremely effective early warning mechanism.

[0070] 2. Traditional elevator maintenance often needs to be carried out regularly, which is not only time-consuming and labor-intensive, but also difficult to ensure that all potential problems can be found every time. However, sound detection technology can achieve real-time monitoring and only focus on maintenance of elevators with abnormal sounds, thereby greatly reducing maintenance costs and improving maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a flowchart of an elevator safety protection method based on big data according to Embodiment 1 of the present invention;

[0072] Figure 2 This is a schematic diagram of the module composition of an elevator safety protection system based on big data according to Example 2 of the present invention. DETAILED DESCRIPTION

[0073] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0074] Before giving examples, it is necessary to explain the application scenarios of the present invention. The method and system can be used to monitor and identify risks in real-time operation of elevators.

[0075] Example 1: Figure 1 As shown, this embodiment provides an elevator safety protection method based on big data, and the method includes:

[0076] The elevator operation sound sequence is obtained by monitoring and reading the elevator operation sound, and the elevator operation sound sequence is subjected to bandpass filtering technology to obtain a first enhanced sound sequence. The first enhanced sound sequence is sliced ​​by a time slice division algorithm to obtain a first interval sound slice, and the first interval sound slice is separated end-to-end by a convolutional neural network endpoint detection technology to obtain a second interval sound slice.

[0077] Exemplarily, the elevator real-time operation information is monitored to obtain the elevator operation sound sequence including the elevator motor sound, elevator brake sound, elevator guide rail sound, elevator cable and pulley sound, elevator car and counterweight sound, door sliding sound, door lock sound, elevator button sound and environmental noise. At this time, the elevator operation sound sequence contains other sound information except the target voice, namely additive noise. The relationship between additive noise and target voice is additive, which is manifested as the waveform superposition of noise signal and voice signal in time sequence. First, the time domain analysis method is used to remove simple periodic noise, and then the frequency domain method is applied to remove complex environmental noise. Secondly, the system preset bandpass filter parameters are set by analyzing the sound characteristics, including determining the center frequency and bandwidth of the bandpass filter, and then filtering the sound signal outside the preset range of the system. The first enhanced sound sequence is obtained here. The first interval sound piece is separated from the end to end by the convolutional neural network endpoint detection technology, which is to strip the environmental noise from the first interval sound piece to obtain a pure elevator operation sound called the second interval sound piece.

[0078] The second interval sound piece is subjected to a frame windowing method to obtain a sound signal sequence and a discrete power spectrum group is obtained by fast Fourier transform, the discrete power spectrum group is subjected to logarithmic operation to obtain a logarithmic power spectrum group and the logarithmic power spectrum is subjected to discrete cosine transform to obtain an original MFCC feature vector set. The energy feature set is extracted from the original MFCC feature vector set by an analysis and processing method, and a dynamic energy feature set is obtained by first-order difference calculation based on the energy feature set and recorded as a dynamic feature set.

[0079] Exemplarily, the MFCC feature vector set of the sound signal is obtained by processing, and the MFCC feature vector set is subdivided into original MFCC feature vector sets of a preset number of dimensions, and the energy feature set is extracted from the original MFCC feature vector set to capture the static characteristics of the sound, so as to have a deeper understanding of the intrinsic characteristics of the sound, and the dynamic feature set is obtained through analysis and calculation, thereby gaining insight into the evolution trajectory of the sound in a short period of time, thereby grasping the short-term change trend of the sound.

[0080] The energy feature set and the dynamic feature set are divided into frames, and the mapping curves of the energy features and the dynamic features on the coordinate axis are drawn through a mapping algorithm and recorded as acoustic wave feature space curves. The Hausdorff distance between the acoustic wave feature space curve and the system preset risk acoustic wave feature space curve is calculated, and the real-time operation risk of the elevator is judged according to the relationship value between the Hausdorff distance and the threshold value preset by the system.

[0081] Furthermore, the method of slicing the first enhanced sound sequence by a time slice division algorithm to obtain the first interval sound slices includes:

[0082] The first enhanced sound sequence is divided into a combination sequence of N frequency segments according to the frequency range and recorded as the second enhanced sound sequence, and the energy of each frequency segment of the second enhanced sound sequence is obtained by the frequency domain energy calculation algorithm [E 1i , E 12 ,……,E 1i ] is recorded as the first signal energy sequence, and the frequency components [F1, F2, ..., F] in the first enhanced sound sequence are calculated using the first signal energy sequence according to the frequency energy conversion formula. k ] is called a frequency sequence. The frequency energy conversion formula is:

[0083]

[0084] Where E i represents the energy of the ith frequency segment, w i represents the weight of the i-th frequency segment, i = 1, 2, ..., N. E total Represents the total energy.

[0085] The signal energy sequence [E1, E2, ..., E3, ..., E4, ..., E5, ..., E6, ..., E7, ..., E8, ..., E9, ..., E10, ..., E110, ..., E12, ..., E13, ..., E14, ..., E15, ..., E16, ..., E17, ..., E18, ..., E19, ..., E20, ..., E21, ..., E22, ..., k ] is recorded as the second signal energy sequence. The time slice length L is calculated by the time slice function according to the second signal energy sequence and the frequency sequence. k The time slice function is:

[0086]

[0087] Among them, α and β are weight coefficients, max(E) is the maximum value of the energy of the first enhanced sound sequence, and max(F) is the maximum value of the frequency sequence.

[0088] Exemplarily, α and β are parameters used to adjust the effect of energy and frequency change rate on the length of the time slice, and a genetic algorithm is used to automatically adjust the values ​​of α and β to optimize the system performance index.

[0089] The first enhanced sound sequence is sliced ​​according to the time slice length using a slice formula to obtain a first interval sound slice sequence [T 11 , T 12 , T 13 , ..., T 1m ]. The dicing formula is:

[0090] T 1m [m] = T[t k +m],0≤m <L k .

[0091] Furthermore, the method for obtaining the energy of each frequency segment of the second enhanced sound sequence by a frequency domain energy calculation algorithm and obtaining the signal energy sequence of each time point of the first enhanced sound sequence by a frequency domain energy calculation algorithm includes:

[0092] The first enhanced sound sequence and the second enhanced sound sequence are subjected to fast Fourier transform to obtain a frequency domain signal X(f), and the amplitude spectrum of the first enhanced sound sequence and the second enhanced sound sequence is calculated by an amplitude spectrum calculation formula. The amplitude spectrum calculation formula is:

[0093]

[0094] Where Re(X(f)) and Im(X(f)) are the real and imaginary parts of X(f), respectively.

[0095] The signal energy sequences of the first enhanced sound sequence and the second enhanced sound sequence are calculated by using the amplitude spectrum energy formula. The amplitude spectrum energy formula is:

[0096]

[0097] Where N is the total number of sampling points of the signal.

[0098] Furthermore, the original MFCC feature vector set is used to extract an energy feature set through an analysis and processing method, and a method for obtaining a dynamic energy feature set recorded as a dynamic feature set through first-order difference calculation according to the energy feature set includes:

[0099] The original MFCC feature vector set is expanded to obtain an MFCC feature matrix, a frame vector of the MFCC feature matrix is ​​obtained, and the energy feature of the frame vector is obtained according to an energy feature calculation formula. The energy feature calculation formula is:

[0100]

[0101] Where E(t) is the energy of the tth frame, x i(t) is the i-th sample of the t-th frame, t = 1, 2, ..., N. L is the number of samples per frame.

[0102] The dynamic energy characteristic is obtained according to the energy characteristic through the formula ΔE(t)=E(t+1)-E(t-1).

[0103] Furthermore, the method of dividing the energy feature set and the dynamic feature set into frames and drawing a mapping curve of the energy feature and the dynamic feature on the coordinate axis through a mapping algorithm includes:

[0104] The dynamic feature set is stacked column by column to obtain a feature matrix F of m×n: According to the column vectors F1, F2, ..., F n , through the formula Get the mean vector μ of each feature. According to the mean vector μ and the column vector of the matrix F, the formula The standard deviation vector σ of each vector is calculated. According to the standard deviation vector σ, the mean vector μ and the column vector of the matrix F, the formula Get the standardized data matrix F scaled According to the standardized data matrix F scaled , through the formula The covariance matrix C is obtained. The covariance matrix C is subjected to eigenvalue decomposition to obtain the eigenvalue λ i and the corresponding eigenvector υ i . The eigenvector υ i According to the characteristic value λ i The size of the eigenvectors is rearranged, and the first k eigenvectors are selected to form the projection matrix W and the formula F is used. pca =W T F scaled Get the mapping matrix F after mapping to the dynamic feature principal component space pca .

[0105] The mapping matrix F mapped to the dynamic feature principal component space is obtained according to the dynamic feature set. pca The mapping matrix X of the energy feature set to the energy feature principal component space is obtained by pca .

[0106] The mapping matrix F pca and the mapping matrix X pca A distance matrix is ​​obtained by calculating the Euclidean distance, and based on the distance matrix, the principal components of the dynamic feature set and the principal components of the energy feature set are mapped to a system preset space through an MDS algorithm to draw a mapping curve.

[0107] Embodiment 2: Based on the same inventive concept, Figure 2 As shown, this embodiment also provides an elevator safety protection system based on big data, and the system includes:

[0108] The acoustic monitoring module is used to obtain an elevator operation sound sequence by monitoring and reading the elevator operation sound, and obtain a first enhanced sound sequence by bandpass filtering technology. The first enhanced sound sequence is sliced ​​by a time slice division algorithm to obtain a first interval sound slice, and the first interval sound slice is separated end-to-end by a convolutional neural network endpoint detection technology to obtain a second interval sound slice.

[0109] The spectrum processing module is connected to the acoustic monitoring module, and is used to obtain a sound signal sequence by a frame windowing method from the second interval sound piece, obtain a discrete power spectrum group by fast Fourier transform, obtain a logarithmic power spectrum group by logarithmic operation from the discrete power spectrum group, and obtain an original MFCC feature vector set by discrete cosine transform from the logarithmic power spectrum. The energy feature set is extracted from the original MFCC feature vector set by an analysis and processing method, and a dynamic energy feature set is obtained by first-order difference calculation based on the energy feature set, which is recorded as a dynamic feature set.

[0110] The matching recognition module is connected to the sound spectrum processing module, and is used to frame the energy feature set and the dynamic feature set and draw the mapping curve of the energy feature and the dynamic feature on the coordinate axis through a mapping algorithm, and record it as an acoustic wave feature space curve; by calculating the Hausdorff distance between the acoustic wave feature space curve and the system preset risk acoustic wave feature space curve, the real-time operation risk of the elevator is judged according to the relationship value between the Hausdorff distance and the threshold value preset by the system.

[0111] Furthermore, the acoustic monitoring module includes:

[0112] The frequency energy conversion module divides the first enhanced sound sequence into a combination sequence of N frequency segments according to the frequency range as a second enhanced sound sequence, and obtains the energy of each frequency segment of the second enhanced sound sequence through a frequency domain energy calculation algorithm [E 1i , E 12 ,……,E 1i ] is recorded as the first signal energy sequence, and the frequency components [F1, F2, ..., F] in the first enhanced sound sequence are calculated using the first signal energy sequence according to the frequency energy conversion formula. k ] is called a frequency sequence. The frequency energy conversion formula is:

[0113]

[0114] Where E irepresents the energy of the ith frequency segment, w i represents the weight of the i-th frequency segment, i = 1, 2, ..., N. E total Represents the total energy.

[0115] The time slice calculation module obtains the signal energy sequence [E1, E2, ..., E3, ..., E4, ..., E5, ..., E6, ..., E7, ..., E8, ..., E9, ..., E10, ..., E11, ..., E12, ..., E13, ..., E14, ..., E15, ..., E16, ..., E17, ..., E18, ..., E19, ..., E20, ..., E21, ..., k ] is recorded as the second signal energy sequence. The time slice length L is calculated by the time slice function according to the second signal energy sequence and the frequency sequence. k The time slice function is:

[0116]

[0117] Among them, α and β are weight coefficients, max(E) is the maximum value of the energy of the first enhanced sound sequence, and max(F) is the maximum value of the frequency sequence.

[0118] A sound slice division module divides the first enhanced sound sequence into slices according to the time slice length using a slice division formula to obtain a first interval sound slice sequence [T 11 , T 12 , T 13 , ..., T 1m ]. The dicing formula is:

[0119] T 1m [m] = T[t k +m],0≤m <L k .

[0120] Furthermore, the frequency energy conversion module includes:

[0121] The frequency domain signal X(f) is obtained by performing fast Fourier transform on the first enhanced sound sequence and the second enhanced sound sequence, and the amplitude spectrum of the first enhanced sound sequence and the second enhanced sound sequence is calculated by using an amplitude spectrum calculation formula. The amplitude spectrum calculation formula is:

[0122]

[0123] Where Re(X(f)) and Im(X(f)) are the real and imaginary parts of X(f), respectively.

[0124] The signal energy sequences of the first enhanced sound sequence and the second enhanced sound sequence are calculated by using the amplitude spectrum energy formula. The amplitude spectrum energy formula is:

[0125]

[0126] Where N is the total number of sampling points of the signal.

[0127] Furthermore, the matching identification module includes:

[0128] The feature extraction module obtains an MFCC feature matrix by expanding the original MFCC feature vector set, obtains a frame vector of the MFCC feature matrix, and obtains the energy feature of the frame vector according to an energy feature calculation formula. The energy feature calculation formula is:

[0129]

[0130] Where E(t) is the energy of the tth frame, x i (t) is the i-th sample of the t-th frame, t = 1, 2, ..., N. L is the number of samples per frame.

[0131] The dynamic energy characteristic is obtained according to the energy characteristic through the formula ΔE(t)=E(t+1)-E(t-1).

[0132] Furthermore, the matching identification module also includes:

[0133] The feature mapping module is connected to the feature extraction module, and obtains a feature matrix F of m×n by stacking the dynamic feature set in columns: According to the column vectors F1, F2, ..., F n , through the formula The mean vector μ of each feature is obtained. According to the mean vector μ and the column vectors F1, F2, ..., F n , through the formula The standard deviation vector σ of each vector is calculated. According to the standard deviation vector σ, the mean vector μ and the column vector of the matrix F, the formula Get the standardized data matrix F scaled According to the standardized data matrix F scaled , through the formula The covariance matrix C is obtained. The covariance matrix C is subjected to eigenvalue decomposition to obtain the eigenvalue λ i and the corresponding eigenvector υ i . The eigenvector υ i According to the characteristic value λ i The size of the eigenvectors is rearranged, and the first k eigenvectors are selected to form the projection matrix W and the formula F is used. pca =W T F scaled Get the mapping matrix F after mapping to the dynamic feature principal component space pca .

[0134] The mapping matrix F mapped to the dynamic feature principal component space is obtained according to the dynamic feature set. pca The mapping matrix X of the energy feature set to the energy feature principal component space is obtained by pca .

[0135] The elevator risk analysis module is connected with the feature mapping module by pca and the mapping matrix X pca A distance matrix is ​​obtained by calculating the Euclidean distance, and based on the distance matrix, the principal components of the dynamic feature set and the principal components of the energy feature set are mapped to a system preset space through an MDS algorithm to draw a mapping curve.

[0136] The present invention is described by specific embodiments, and it should be understood by those skilled in the art that various changes and equivalent substitutions may be made to the present invention without departing from the scope of the present invention. In addition, various modifications may be made to the present invention for specific situations or materials without departing from the scope of the present invention. Therefore, the present invention is not limited to the disclosed specific embodiments, but should include all implementation methods falling within the scope of the claims of the present invention.

Claims

1. An elevator safety protection method based on big data, It is characterized in that The method comprises: An elevator operation sound sequence is obtained by monitoring and reading the elevator operation sound, and the elevator operation sound sequence is subjected to a bandpass filtering technology to obtain a first enhanced sound sequence; the first enhanced sound sequence is sliced ​​by a time slice partitioning algorithm to obtain a first interval sound slice, and the first interval sound slice is separated end-to-end by a convolutional neural network endpoint detection technology to obtain a second interval sound slice; The second interval sound piece is subjected to a frame windowing method to obtain a sound signal sequence and a fast Fourier transform to obtain a discrete power spectrum group, the discrete power spectrum group is subjected to a logarithmic operation to obtain a logarithmic power spectrum group and the logarithmic power spectrum is subjected to a discrete cosine transform to obtain an original MFCC feature vector set; an energy feature set is extracted from the original MFCC feature vector set by an analysis and processing method, and a dynamic energy feature set is obtained by first-order difference calculation based on the energy feature set and recorded as a dynamic feature set; The energy feature set and the dynamic feature set are divided into frames, and the mapping curves of the energy features and the dynamic features on the coordinate axis are drawn through a mapping algorithm and recorded as acoustic wave feature space curves. The Hausdorff distance between the acoustic wave feature space curve and the system preset risk acoustic wave feature space curve is calculated, and the real-time operation risk of the elevator is judged according to the relationship value between the Hausdorff distance and the threshold value preset by the system.

2. The elevator safety protection method based on big data as claimed in claim 1, It is characterized in that The method of slicing the first enhanced sound sequence by a time slice partitioning algorithm to obtain first interval sound slices includes: The first enhanced sound sequence is divided into a combination sequence of N frequency segments according to the frequency range and recorded as the second enhanced sound sequence, and the energy of each frequency segment of the second enhanced sound sequence is obtained by the frequency domain energy calculation algorithm [E 1i , E 12 ,……,E 1i ] is recorded as the first signal energy sequence, and the frequency components [F 1 , F 2 , ..., F k ] is called frequency sequence; the frequency energy conversion formula is: Where E i represents the energy of the ith frequency segment, w i represents the weight of the i-th frequency band, i = 1, 2, ..., N; E total represents the total energy; The signal energy sequence of each time point of the first enhanced sound sequence is obtained by using a frequency domain energy calculation algorithm [E 1 , E 2 ,……,E k ] is recorded as the second signal energy sequence; the time slice length L is calculated by the time slice function according to the second signal energy sequence and the frequency sequence k ; The time slice function is: Wherein, α and β are weight coefficients, max(E) is the maximum value of the energy of the first enhanced sound sequence, and max(F) is the maximum value of the frequency sequence; The first enhanced sound sequence is sliced ​​according to the time slice length using a slice formula to obtain a first interval sound slice sequence [T 11 , T 12 , T 13 , ..., T 1m ]; the dicing formula is: T 1m [m]=T[t k +m],0≤m<L k 。 3. The elevator safety protection method based on big data as claimed in claim 2, It is characterized in that The method for obtaining the energy of each frequency segment of the second enhanced sound sequence by a frequency domain energy calculation algorithm and obtaining the signal energy sequence of each time point of the first enhanced sound sequence by a frequency domain energy calculation algorithm comprises: The first enhanced sound sequence and the second enhanced sound sequence are subjected to fast Fourier transform to obtain a frequency domain signal X(f), and the amplitude spectrum of the first enhanced sound sequence and the second enhanced sound sequence is calculated by an amplitude spectrum calculation formula; the amplitude spectrum calculation formula is: Where Re(X(f)) is the real part of X(f), Im(X(f)) is the imaginary part of X(f); The signal energy sequences of the first enhanced sound sequence and the second enhanced sound sequence are calculated by using the amplitude spectrum energy formula; the amplitude spectrum energy formula is: Where N is the total number of sampling points of the signal.

4. The elevator safety protection method based on big data as claimed in claim 1, It is characterized in that The original MFCC feature vector set is extracted from an energy feature set by an analysis and processing method, and a method for obtaining a dynamic energy feature set recorded as a dynamic feature set by first-order difference calculation based on the energy feature set includes: The original MFCC feature vector set is expanded to obtain an MFCC feature matrix, a frame vector of the MFCC feature matrix is ​​obtained, and the energy feature of the frame vector is obtained according to an energy feature calculation formula; the energy feature calculation formula is: Where E(t) is the energy of the tth frame, x i (t) is the i-th sample of the t-th frame, t = 1, 2, ..., N; L is the number of samples per frame; The dynamic energy characteristic is obtained according to the energy characteristic through the formula ΔE(t)=E(t+1)-E(t-1).

5. The elevator safety protection method based on big data as claimed in claim 1, It is characterized in that The method of dividing the energy feature set and the dynamic feature set into frames and drawing a mapping curve of the energy feature and the dynamic feature on a coordinate axis through a mapping algorithm comprises: The dynamic feature set is stacked column by column to obtain a feature matrix F of m×n: According to the column vector F of the matrix F 1 , F 2 , ..., F n , through the formula Get the mean vector μ of each feature; According to the mean vector μ and the column vector F of the matrix F 1 , F 2 , ..., F n , through the formula Calculate the standard deviation vector σ of each vector; according to the standard deviation vector σ, the mean vector μ and the column vector of the matrix F, through the formula Get the standardized data matrix F scaled ; According to the standardized data matrix F scaled , through the formula Obtain the covariance matrix C; perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalue λ i and the corresponding eigenvector υ i ; The feature vector υ i According to the characteristic value λ i The size of the eigenvectors is rearranged, and the first k eigenvectors are selected to form the projection matrix W and the formula F is used. pca =W T F scaled Get the mapping matrix F after mapping to the dynamic feature principal component space pca ; The mapping matrix F mapped to the dynamic feature principal component space is obtained according to the dynamic feature set. pca The mapping matrix X of the energy feature set to the energy feature principal component space is obtained by pca ; The mapping matrix F pca and the mapping matrix X pca A distance matrix is ​​obtained by calculating the Euclidean distance, and based on the distance matrix, the principal components of the dynamic feature set and the principal components of the energy feature set are mapped to a system preset space through an MDS algorithm to draw a mapping curve.

6. An elevator safety protection system based on big data, It is characterized in that The system includes: an acoustic monitoring module, a sound spectrum processing module, and a matching recognition module; The acoustic monitoring module is used to obtain an elevator operation sound sequence by monitoring and reading the elevator operation sound, and obtain a first enhanced sound sequence by using a bandpass filtering technology on the elevator operation sound sequence; slice the first enhanced sound sequence by using a time slice partitioning algorithm to obtain a first interval sound slice, and perform end-to-end sound segment separation on the first interval sound slice by using a convolutional neural network endpoint detection technology to obtain a second interval sound slice; The spectrum processing module is connected to the acoustic monitoring module, and is used to obtain a sound signal sequence by a frame windowing method for the second interval sound piece, and obtain a discrete power spectrum group by a fast Fourier transform, obtain a logarithmic power spectrum group by a logarithmic operation for the discrete power spectrum group, and obtain an original MFCC feature vector set by a discrete cosine transform for the logarithmic power spectrum; extract an energy feature set from the original MFCC feature vector set by an analysis and processing method, and obtain a dynamic energy feature set by a first-order difference calculation based on the energy feature set, which is recorded as a dynamic feature set; The matching identification module is connected to the sound spectrum processing module, and is used to frame the energy feature set and the dynamic feature set and draw the mapping curve of the energy feature and the dynamic feature on the coordinate axis through a mapping algorithm to record it as an acoustic wave feature space curve, and calculate the Hausdorff distance between the acoustic wave feature space curve and the system preset risk acoustic wave feature space curve, and judge the real-time operation risk of the elevator according to the relationship value between the Hausdorff distance and the threshold value preset by the system.

7. An elevator safety protection system based on big data as claimed in claim 6, It is characterized in that The acoustic monitoring module comprises: The frequency energy conversion module divides the first enhanced sound sequence into a combination sequence of N frequency segments according to the frequency range as a second enhanced sound sequence, and obtains the energy of each frequency segment of the second enhanced sound sequence through a frequency domain energy calculation algorithm [E 1i , E 12 ,……,E 1i ] is recorded as the first signal energy sequence, and the frequency components [F 1 , F 2 , ..., F k ] is called frequency sequence; the frequency energy conversion formula is: Where E i represents the energy of the ith frequency segment, w i represents the weight of the i-th frequency band, i = 1, 2, ..., N; E total represents the total energy; The time slice calculation module obtains the signal energy sequence of each time point of the first enhanced sound sequence through a frequency domain energy calculation algorithm [E 1 , E 2 ,……,E k ] is recorded as the second signal energy sequence; the time slice length L is calculated by the time slice function according to the second signal energy sequence and the frequency sequence k ; The time slice function is: Wherein, α and β are weight coefficients, max(E) is the maximum value of the energy of the first enhanced sound sequence, and max(F) is the maximum value of the frequency sequence; A sound slice division module divides the first enhanced sound sequence into slices according to the time slice length using a slice division formula to obtain a first interval sound slice sequence [T 11 , T 12 , T 13 , ..., T 1m ]; the dicing formula is: T 1m [m]=T[t k +m],0≤m<L k 。 8. An elevator safety protection system based on big data as claimed in claim 7, It is characterized in that The frequency energy conversion module comprises: The frequency domain signal X(f) is obtained by performing fast Fourier transform on the first enhanced sound sequence and the second enhanced sound sequence, and the amplitude spectrum of the first enhanced sound sequence and the second enhanced sound sequence is calculated by an amplitude spectrum calculation formula; the amplitude spectrum calculation formula is: Where Re(X(f)) and Im(X(f)) are the real and imaginary parts of X(f), respectively; The signal energy sequences of the first enhanced sound sequence and the second enhanced sound sequence are calculated by using the amplitude spectrum energy formula; the amplitude spectrum energy formula is: Where N is the total number of sampling points of the signal.

9. An elevator safety protection system based on big data as claimed in claim 6, It is characterized in that The matching identification module comprises: The feature extraction module obtains an MFCC feature matrix by expanding the original MFCC feature vector set, obtains a frame vector of the MFCC feature matrix, and obtains the energy feature of the frame vector according to an energy feature calculation formula; the energy feature calculation formula is: Where E(t) is the energy of the tth frame, x i (t) is the i-th sample of the t-th frame, t = 1, 2, ..., N; L is the number of samples per frame; The dynamic energy characteristic is obtained according to the energy characteristic through the formula ΔE(t)=E(t+1)-E(t-1).

10. An elevator safety protection system based on big data as claimed in claim 6, It is characterized in that The matching identification module also includes: The feature mapping module is connected to the feature extraction module, and obtains a feature matrix F of m×n by stacking the dynamic feature set in columns: According to the column vector F of the matrix F 1 , F 2 , ..., F n , through the formula Get the mean vector μ of each feature; According to the mean vector μ and the column vector F of the matrix F 1 , F 2 , ..., F n , through the formula Calculate the standard deviation vector v of each vector; According to the standard deviation vector σ, the mean vector μ and the column vector of the matrix F, the formula Get the standardized data matrix F scaled ; According to the standardized data matrix F acaled , through the formula Obtain the covariance matrix C; perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalue λ i and the corresponding eigenvector υ i ; The feature vector υ i According to the characteristic value λ i The size of the eigenvectors is rearranged, and the first k eigenvectors are selected to form the projection matrix W and the formula F is used. pca =W T F scaled Get the mapping matrix F after mapping to the dynamic feature principal component space pca ; The mapping matrix F mapped to the dynamic feature principal component space is obtained according to the dynamic feature set. pca The mapping matrix X of the energy feature set to the energy feature principal component space is obtained by pca ; The elevator risk analysis module is connected with the feature mapping module by pca and the mapping matrix X pca A distance matrix is ​​obtained by calculating the Euclidean distance, and based on the distance matrix, the principal components of the dynamic feature set and the principal components of the energy feature set are mapped to a system preset space through an MDS algorithm to draw a mapping curve.