A method for checking the status of power grid equipment based on finite state machine

Abstract

The present invention discloses a method for checking the state of power grid equipment based on a finite state machine, which relates to the technical field of speech recognition and comprises the following steps: obtaining a speech signal to be tested, performing feature extraction to obtain speech features; performing data analysis on the speech features to construct a plurality of speech feature spaces; constructing a speech feature matrix according to the speech features, and projecting the speech feature matrix into a plurality of speech feature spaces for dimensionality reduction; evaluating each speech feature after dimensionality reduction, and screening to obtain the best speech feature after dimensionality reduction; the present application solves the problem of improving the accuracy of speech recognition while effectively reducing the dimensionality of speech feature data by evaluating the speech features after dimensionality reduction with different numbers of speech features.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and more specifically, to a method for checking the state of power grid equipment based on a finite state machine. Background Art

[0002] In the power industry, the monitoring and verification of the operating status of power grid equipment has become a key link in ensuring the safe and stable operation of the power system. Traditional methods for verifying the status of power grid equipment mainly rely on manual inspections and sensor data collection. Although these methods can reflect the operating status of equipment to a certain extent, they have problems such as low efficiency, high cost, and poor real-time performance. Especially in large-scale power grid systems, the number of equipment is huge and widely distributed, and traditional methods are difficult to meet the real-time and accurate verification needs.

[0003] With the rapid development of speech recognition technology, its application in various fields is becoming more and more extensive. Speech recognition technology is an important human-computer interaction technology. Through voice input, the system can automatically recognize and convert it into text information, which greatly improves the efficiency and accuracy of information processing. In the field of power grid equipment information verification, the introduction of speech recognition technology and the realization of voice verification have become an innovative and effective solution.

[0004] As a mathematical model that represents a finite number of states and the transitions and actions between these states, finite state machines are widely used in computer science, hardware circuit system design, software engineering and other fields. Finite state machines have the characteristics of a limited number of states and clear state transitions. They can well simulate the state sequence experienced by an object during its life cycle and how to respond to various events from the outside world.

[0005] For example, the speech recognition method, device and equipment disclosed in the invention patent with announcement number: CN108564941A belong to the field of speech recognition. The method includes: obtaining speech information; determining the starting and ending positions of the candidate speech segments in the speech information through a weighted finite state machine network; intercepting the candidate speech segments in the speech information according to the starting and ending positions of the candidate speech segments; inputting the candidate speech segments into the machine learning model, and detecting whether the candidate speech segments contain preset keywords through the machine learning model. The present application verifies the candidate speech segments roughly located by the weighted finite state machine network through a machine learning model to determine whether the candidate speech segments contain preset keywords, thereby solving the problem in the related art that speech information without semantics may be identified as speech information with semantics, thereby causing false wake-up, and improving the accuracy of speech recognition.

[0006] For example, the invention patent with announcement number: CN105895081A discloses a method and device for speech recognition decoding, which belongs to the field of speech processing. The method includes: receiving speech information and extracting acoustic features; calculating the information of the acoustic features according to the connection time series classification model; if the frame in the acoustic feature information is a non-empty model frame, a weighted finite state machine adapted to the acoustic modeling information is used to search for linguistic information and store the history, otherwise the frame is discarded. The present invention makes acoustic modeling more accurate by establishing a continuous time series classification model; uses an improved weighted finite state machine to make the model representation more efficient, reducing the consumption of computing and memory resources by nearly 50%; uses the phoneme synchronization method in decoding to effectively reduce the amount of calculation and the number of model searches.

[0007] The above disclosed technical solutions have at least the following technical problems:

[0008] Before traditional speech verification, the speech signal is feature extracted and then dimensionally reduced, ignoring the fact that different numbers of speech features construct different speech feature spaces, and the degree of retention of important information in the speech after dimensionality reduction is also different.

[0009] In view of the above problems, the present invention proposes a solution. Summary of the invention

[0010] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a power grid equipment status verification method based on a finite state machine, which solves the problem of improving the accuracy of speech recognition while effectively reducing the dimension of speech feature data by evaluating speech features after dimensionality reduction of different speech feature quantities.

[0011] To achieve the above object, the present invention provides the following technical solutions:

[0012] A method for checking the state of power grid equipment based on a finite state machine comprises the following steps: obtaining a voice signal to be tested, performing feature extraction to obtain voice features; performing data analysis on the voice features, and constructing a plurality of voice feature spaces; constructing a voice feature matrix according to the voice features, and projecting the voice feature matrix into a plurality of voice feature spaces for dimensionality reduction; evaluating each voice feature after dimensionality reduction, and selecting the best voice feature after dimensionality reduction.

[0013] In a preferred embodiment, a speech signal to be tested is obtained, and feature extraction is performed to obtain speech features, specifically as follows: a speech signal to be tested is obtained, and preprocessed; speech features are extracted based on the preprocessed speech signal, and the speech features are standardized; a speech feature matrix is ​​constructed based on the speech features, and a linear analysis is performed on the speech feature data to construct a covariance matrix of the speech feature matrix; eigenvalue decomposition is performed on the covariance matrix of the speech feature matrix to obtain speech eigenvalues ​​and speech feature vectors.

[0014] In a preferred embodiment, data analysis is performed on speech features to construct several speech feature spaces, specifically as follows: corresponding speech feature vectors greater than preset speech feature values ​​are screened out according to speech feature values; the dimension of the speech feature space is determined according to the number of screened speech feature vectors, and the dimension of the speech feature space should match the number of screened speech feature vectors; the speech feature vectors corresponding to the first k largest speech feature values ​​are extracted from the screened speech feature vectors as input data, and are input into a machine learning model to construct multiple groups of speech feature spaces of different dimensions.

[0015] In a preferred embodiment, a speech feature matrix is ​​constructed according to speech features, and the speech feature matrix is ​​projected into several speech feature spaces for dimensionality reduction, specifically as follows: the extracted speech features are arranged in chronological order to construct a two-dimensional speech feature matrix; the speech feature vectors of the speech feature matrix are projected into the feature space; the projected speech feature vectors are reorganized according to the order of the speech feature matrix to obtain a speech feature matrix after dimensionality reduction.

[0016] In a preferred embodiment, each speech feature after dimensionality reduction is evaluated, and the best speech feature after dimensionality reduction is screened, as follows: a reconstructed feature data matrix is ​​constructed according to the speech feature matrix, the matrix composed of the first k speech feature vectors and the speech feature matrix after dimensionality reduction; the reconstructed feature data matrix and the initial speech feature matrix are compared for information differentiation to obtain a reconstruction error; the speech features after dimensionality reduction are input into a pre-trained speech recognition model and converted into text data; the converted text data is input into a finite state machine for matching and classification with the database text, and the classification accuracy of the speech features after dimensionality reduction is evaluated based on cross-validation; the sum of the eigenvalues ​​of the first k speech feature vectors is compared with the sum of all eigenvalues ​​in the initial speech feature vector to obtain the variance ratio of the speech feature values ​​before and after dimensionality reduction; according to the variance ratio of the speech feature values ​​before and after dimensionality reduction, the variance retention rate of the speech feature values ​​after dimensionality reduction is obtained; the reconstruction error, classification accuracy and variance retention rate are input into the speech feature evaluation model to evaluate each speech feature after dimensionality reduction.

[0017] In a preferred embodiment, the best speech features after dimensionality reduction are screened out as follows: each speech feature after dimensionality reduction is evaluated to construct an evaluation data set; data comparison is performed based on the evaluation data set to screen out the best speech features after dimensionality reduction.

[0018] The technical effects and advantages of the power grid equipment status verification method based on a finite state machine of the present invention are as follows:

[0019] 1. The present invention effectively reduces the impact of noise and redundant data on the verification process through fine speech feature extraction and dimensionality reduction processing. In the feature extraction stage, the speech signal is preprocessed and standardized so that the noise in the original signal is filtered, so that the obtained speech features are more stable and accurate. Next, the speech features are further refined through linear analysis and the construction of the covariance matrix, and the effective speech feature vector is obtained by eigenvalue decomposition. This process ensures that the extracted feature information reflects the key information of the speech signal to be tested to the maximum extent. In the data analysis stage, speech feature value screening, feature space construction, and voice space dimension adjustment based on machine learning provide accurate speech matching capabilities for the verification system. In particular, when screening feature vectors, the combination of preset thresholds and feature optimization after dimensionality reduction makes the speech features more in line with the needs of practical applications during the processing process, thereby improving the matching accuracy during the verification process.

[0020] 2. The present invention optimizes speech features through dimensionality reduction technology, making the speech data after dimensionality reduction more compact and representative. This dimensionality reduction method not only retains the main speech information, but also effectively reduces the burden of the computing system. Especially in the speech feature matrix evaluation stage after dimensionality reduction, the reconstruction error, classification accuracy and variance retention rate are combined to accurately evaluate the effect of dimensionality reduction. Through the comprehensive analysis of these three, it can be ensured that important information will not be lost during the dimensionality reduction process, and a high classification accuracy can be maintained, further optimizing the efficiency and performance of the computing process.

[0021] 3. The present invention can effectively realize efficient and accurate speech classification by inputting the reduced-dimensional speech features into a finite state machine and matching them with the text data in the database. In the present invention, the use of a finite state machine not only enhances the robustness of the system, but also improves the classification accuracy through cross-validation evaluation. At the same time, dimensionality reduction processing and feature selection techniques are used, which not only reduces the amount of calculation, but also reduces the demand for data storage. This advantage is particularly important for the large amount of speech data that needs to be processed in power grid equipment. By optimizing and compressing the speech feature data, the system can still operate efficiently under limited resources, saving computing and storage resources and reducing operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The figure is a flow chart of a method for checking the state of power grid equipment based on a finite state machine according to the present invention. DETAILED DESCRIPTION

[0023] 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.

[0024] Embodiment 1, Figure 1 The present invention provides a method for checking the state of power grid equipment based on a finite state machine, and the specific steps are as follows:

[0025] S1, obtaining a speech signal to be tested, and performing feature extraction to obtain speech features.

[0026] In this embodiment, the acquisition of the voice signal to be tested and the extraction of features to obtain voice features are specifically as follows:

[0027] Acquire the speech signal to be tested and perform preprocessing to improve the quality of the speech signal;

[0028] Extracting speech features according to the preprocessed speech signal and standardizing the speech features to eliminate the dimension differences and value range differences between different features, wherein the speech features include but are not limited to energy features, Mel-frequency cepstral coefficients, formant frequencies, and fundamental frequencies;

[0029] Constructing a speech feature matrix according to the speech features, and performing linear analysis on the speech feature data to construct a covariance matrix of the speech feature matrix;

[0030] Perform eigenvalue decomposition on the covariance matrix of the speech feature matrix to obtain speech eigenvalues ​​and speech eigenvectors.

[0031] The speech feature matrix formula is as follows:

[0032]

[0033] The covariance matrix calculation formula of the speech feature matrix is ​​as follows:

[0034]

[0035] In the formula, is the speech feature matrix, is the feature vector of the Nth frame, is the number of frames of the speech signal, is the covariance matrix of the speech feature matrix, is the transposed matrix of the speech feature matrix.

[0036] It should be noted that during the input process, the voice signal is affected by the power grid equipment, which increases the noise signal. The noise signal has a negative impact on the accuracy of subsequent voice feature extraction. Therefore, before extracting the voice features, the noise signal needs to be suppressed and de-noised.

[0037] S2, performs data analysis on speech features and constructs several speech feature spaces.

[0038] Speech feature space refers to a series of vector spaces constructed by extracting features and analyzing data of speech signals in speech recognition technology. These spaces are composed of speech feature vectors, each of which represents the feature information of a speech signal in a certain dimension. The process of constructing speech feature space involves preprocessing, feature extraction, and standardization of the original speech signal, as well as linear analysis and covariance matrix construction of speech feature data, and finally obtaining speech eigenvalues ​​and speech feature vectors through eigenvalue decomposition.

[0039] In this embodiment, the speech features are analyzed for data to construct several speech feature spaces, as follows:

[0040] According to the speech feature value, corresponding speech feature vectors greater than the preset speech feature value are selected, and the speech feature vectors are sorted in descending order according to the speech feature value;

[0041] According to the number of the screened speech feature vectors, the dimension of the speech feature space is determined, and the dimension of the speech feature space should match the number of the screened speech feature vectors;

[0042] The speech feature vectors corresponding to the first k largest speech feature values ​​after the screening are extracted as input data, which are input into the machine learning model to construct multiple groups of speech feature spaces of different dimensions.

[0043] It should be noted that the association between the speech feature space and the power grid equipment status is mainly reflected in the following aspects:

[0044] Information extraction and verification: In the power service industry, users’ voice information is often used to verify basic information, such as name, contact information, etc. The accuracy of this information is directly related to the smooth implementation of services such as electricity bill collection and power outage notification. By constructing a voice feature space, key information in the voice signal can be extracted more finely, thereby achieving accurate verification of user voice information;

[0045] Condition monitoring and diagnosis: Although it may not be direct to monitor the physical status of power grid equipment (such as voltage, current, etc.) directly through the voice feature space, the voice feature space can indirectly reflect some abnormal conditions of power grid equipment during use. For example, the failure of power grid equipment may cause the quality of voice communication to deteriorate, which is reflected in the changes in voice features. By monitoring these changes, possible problems with power grid equipment can be discovered in a timely manner, and appropriate measures can be taken to diagnose and repair them;

[0046] Intelligent interaction and response: With the development of smart grids, the interaction between grid equipment and users is becoming more and more intelligent. By constructing a speech feature space, the accuracy and robustness of the speech recognition system can be improved, allowing users to interact with grid equipment more conveniently and accurately. For example, users can use voice commands to query electricity bills, report faults, etc., and grid equipment can respond accordingly based on the user's voice commands.

[0047] Let's further explain the differences in speech feature spaces constructed with different numbers of speech features:

[0048] Granularity is an important parameter for describing the refinement of speech feature space. The smaller the granularity, the fewer the markers, the lower the refinement of speech feature space and the weaker the descriptive ability. On the contrary, the larger the granularity, the more the markers, the stronger the descriptive ability of speech feature space and the more conducive to calculating accurate feature distribution.

[0049] The more feature sample points used to construct the speech feature space, the more complete the speech feature space will be, and the higher the coverage of any speech feature will be. This means that when the number of speech features increases, the constructed speech feature space can more comprehensively reflect the diversity and complexity of speech;

[0050] To a certain extent, increasing the amount of data used to construct the speech feature space (i.e., the number of speech features) can greatly improve recognition performance. More speech features can provide richer information, thereby improving the accuracy and robustness of speech recognition.

[0051] It should be noted that the more speech features there are, the better. When the number of speech features increases to a certain level, the improvement in speech recognition performance may no longer be significant, and may even lead to performance degradation due to problems such as overfitting. Therefore, in practical applications, it is necessary to find a suitable balance point in the number of speech features.

[0052] S3, constructing a speech feature matrix according to the speech features, and projecting the speech feature matrix into several speech feature spaces for dimensionality reduction.

[0053] In this embodiment, the speech feature matrix is ​​constructed according to the speech features, and the speech feature matrix is ​​projected into several speech feature spaces for dimensionality reduction, as follows:

[0054] Arrange the extracted speech features in chronological order to construct a two-dimensional speech feature matrix;

[0055] Each row of the speech feature matrix represents a feature vector of a speech frame, and each column represents a specific speech feature;

[0056] Projecting the speech feature vector of the speech feature matrix into the feature space;

[0057] The projected speech feature vectors are reorganized according to the order of the speech feature matrix to obtain a speech feature matrix after dimensionality reduction.

[0058] The formula of the speech feature matrix after dimension reduction is as follows:

[0059]

[0060] In the formula, is the speech feature matrix after dimensionality reduction, is the matrix composed of the first k speech feature vectors, is the speech feature matrix.

[0061] It should be noted that because different feature spaces can capture key information in speech signals from different angles and levels, by projecting these feature vectors into multiple feature spaces, we can obtain a more comprehensive and in-depth representation of speech features, thereby more effectively reducing the dimension of the data and retaining key information.

[0062] Furthermore, the impact of speech feature space on power grid equipment status verification is mainly reflected in the following aspects:

[0063] First, multi-space projection dimensionality reduction can remove noise and redundant data in speech signals and improve the accuracy of speech recognition; second, by constructing speech feature spaces of different dimensions, we can more flexibly adapt to different power grid equipment status verification requirements and improve the efficiency and accuracy of verification; finally, these feature spaces also provide a more stable and reliable speech feature representation for subsequent text matching and classification, further improving the performance of power grid equipment status verification.

[0064] S4, evaluate each speech feature after dimensionality reduction, and select the best speech feature after dimensionality reduction.

[0065] In this embodiment, the speech features after dimension reduction are evaluated to select the best speech features after dimension reduction, as follows:

[0066] The specific steps of evaluating each speech feature after dimensionality reduction are as follows:

[0067] According to the speech feature matrix, the matrix composed of the first k speech feature vectors and the speech feature matrix after dimensionality reduction, the speech features after dimensionality reduction are restored to the approximate value of the original data to obtain a reconstructed feature data matrix;

[0068] Perform information differentiation comparison between the reconstructed feature data matrix and the initial speech feature matrix to obtain the reconstruction error;

[0069] Input the reduced-dimensional speech features into the pre-trained speech recognition model and convert them into text data;

[0070] The converted text data is input into the finite state machine for matching and classification with the database text, and the classification accuracy of the speech features after dimensionality reduction is evaluated based on cross-validation;

[0071] Compare the sum of the eigenvalues ​​of the first k speech feature vectors with the sum of all eigenvalues ​​in the initial speech feature vector to obtain the variance ratio of the speech feature values ​​before and after dimensionality reduction;

[0072] The variance ratio of the speech feature value before and after dimensionality reduction is divided to obtain the variance retention ratio of the speech feature value after dimensionality reduction;

[0073] The reconstruction error, classification accuracy and variance retention rate are input into the speech feature evaluation model, each speech feature after dimensionality reduction is evaluated, and an evaluation data set is constructed;

[0074] Data comparison is performed based on the evaluation data set to screen out the best speech features after dimensionality reduction.

[0075] The calculation formula for reconstructing the characteristic data matrix is ​​as follows:

[0076]

[0077] The calculation formula of the reconstruction error is as follows:

[0078]

[0079] The calculation formula for evaluating each speech feature after dimensionality reduction is as follows:

[0080]

[0081] In the formula, is the transposed matrix of the matrix composed of the first k speech feature vectors, is the speech feature matrix after dimensionality reduction, is the speech feature matrix, is the reconstructed feature data matrix, represents the evaluation vector of the i-th speech feature after dimensionality reduction, is the reconstruction error, is the classification accuracy, is the variance preservation ratio.

[0082] It should be noted that the reconstruction error, classification accuracy and variance retention rate evaluation indicators are used as features, the reduced-dimensional speech feature space is bound to the evaluation indicators, and each reduced-dimensional speech feature space will correspond to a speech feature vector composed of the evaluation indicators.

[0083] Furthermore, a lower reconstruction error means that the speech features after dimensionality reduction retain most of the original information, which helps to improve performance when applying the model. If the reconstruction error is high, it may mean that a lot of key information has been lost in the dimensionality reduction process, which may lead to poor performance of the model; if the classification accuracy is high, it means that the speech features after dimensionality reduction can help the model perform better prediction or classification tasks. Lower classification accuracy means that the speech features after dimensionality reduction are not suitable for classification tasks and need further adjustment or selection of other dimensionality reduction methods; higher variance retention rate means that the speech feature space after dimensionality reduction retains more variability of the original data. A lower variance retention value means that too much information has been lost in the dimensionality reduction process, which may lead to poor performance of the model in practical applications.

[0084] Finally, the value of the evaluation vector of the speech feature after dimensionality reduction is between [0,1]. The closer it is to 1, the better the dimensionality reduction method performs in retaining the original information, improving classification accuracy, and preserving variance; the closer it is to 0, the worse the dimensionality reduction method performs in these aspects.

[0085] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0086] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0087] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0088] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0089] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0090] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for checking the state of power grid equipment based on a finite state machine, characterized in that: The steps include: Acquire the speech signal to be tested, perform feature extraction to obtain speech features, construct a speech feature matrix based on the speech features, and perform linear analysis to construct a covariance matrix; Perform eigenvalue decomposition on the covariance matrix to obtain speech eigenvalues ​​and speech eigenvectors; Perform data analysis on speech features and construct several speech feature spaces; Constructing a speech feature matrix according to the speech features, and projecting the speech feature matrix into a plurality of speech feature spaces for dimensionality reduction, thereby obtaining a speech feature matrix after dimensionality reduction; According to the speech feature matrix, the matrix composed of speech feature vectors corresponding to the first k largest speech feature values ​​and the speech feature matrix after dimensionality reduction, a reconstruction feature data matrix is ​​constructed, and the reconstruction error is obtained by differential comparison with the speech feature matrix information; The reduced-dimensional speech features are input into the pre-trained speech recognition model, converted into text data, and input into the finite state machine to match and classify with the database text. The classification accuracy of the reduced-dimensional speech features is evaluated based on cross-validation. Compare the sum of the eigenvalues ​​of the first k speech feature vectors with the sum of all eigenvalues ​​in the speech feature vector to obtain the variance ratio of the speech feature values ​​before and after dimensionality reduction, and calculate the variance retention rate of the speech feature values ​​after dimensionality reduction; The reconstruction error, classification accuracy and variance retention rate are input into the speech feature evaluation model, each speech feature after dimensionality reduction is evaluated, and the best speech feature after dimensionality reduction is screened out; The calculation formula for reconstructing the feature data matrix is ​​as follows: The calculation formula of reconstruction error is as follows: In the formula, is the transposed matrix of the matrix composed of the first k speech feature vectors, is the speech feature matrix after dimensionality reduction, is the speech feature matrix, is the reconstructed feature data matrix, is the evaluation vector of the i-th speech feature after dimensionality reduction, is the reconstruction error.

2. The method for checking the state of power grid equipment based on finite state machine according to claim 1, characterized in that: The speech features are analyzed for data and several speech feature spaces are constructed, as follows: According to the speech feature value, a corresponding speech feature vector greater than a preset speech feature value is selected; Determine the dimension of the speech feature space according to the number of the screened speech feature vectors; The speech feature vectors corresponding to the first k largest speech feature values ​​after the screening are extracted as input data, which are input into the machine learning model to construct multiple groups of speech feature spaces of different dimensions.

3. The method for checking the state of power grid equipment based on finite state machine according to claim 2, characterized in that: The speech feature matrix is ​​constructed according to the speech features, and the speech feature matrix is ​​projected into several speech feature spaces for dimensionality reduction, as follows: Arrange the extracted speech features in chronological order to construct a two-dimensional speech feature matrix; Projecting the speech feature vector of the speech feature matrix into the speech feature space according to each row; The projected speech feature vectors are reorganized according to the order of the speech feature matrix to obtain a speech feature matrix after dimensionality reduction.

4. The method for checking the state of power grid equipment based on finite state machine according to claim 3 is characterized in that: The screening obtains the best speech features after dimensionality reduction, which are as follows: Evaluate each speech feature after dimensionality reduction and construct an evaluation data set; Data comparison is performed based on the evaluation data set to screen out the best speech features after dimensionality reduction.

5. The method for checking the state of power grid equipment based on finite state machine according to claim 4, characterized in that: The speech feature matrix formula is as follows: The covariance matrix calculation formula of the speech feature matrix is ​​as follows: In the formula, is the speech feature matrix, is the feature vector of the Nth frame, is the number of frames of the speech signal, is the covariance matrix of the speech feature matrix, is the transposed matrix of the speech feature matrix.

6. The method for checking the state of power grid equipment based on finite state machine according to claim 5, characterized in that: The formula of the speech feature matrix after dimension reduction is as follows: In the formula, is the speech feature matrix after dimensionality reduction, is the matrix composed of the first k speech feature vectors, is the speech feature matrix.

Images