Substation equipment operation state judgment method and system based on voiceprint data

Abstract

The invention discloses a transformer substation equipment operation state judgment method and system based on voiceprint data, and relates to the technical field of transformer substation equipment fault early warning, and the method comprises the steps: constructing a first voiceprint database and a second voiceprint database, carrying out the spectrum feature analysis of voiceprint real-time collection data, and obtaining a feature vector; performing similarity analysis on the feature vector and a first voiceprint database and a second voiceprint database, and respectively generating a first similarity value time sequence arrangement and a second similarity value time sequence arrangement according to the first similarity value and the second similarity value, and judging the operation state of the target substation equipment in combination with the first similarity value time sequence arrangement and the second similarity value time sequence arrangement. According to the judgment method and system, the voiceprint databases of the normal operation state and the abnormal operation state are constructed, the collected target sound signals are matched with the two voiceprint databases respectively, comprehensive judgment of the operation state of the equipment is carried out according to the matching result of the two voiceprint databases, and the purpose of early warning of the operation state of the substation equipment can be achieved more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of substation equipment fault early warning, and more specifically, to a method and system for judging the operation state of substation equipment based on voiceprint data. Background Art

[0002] The online monitoring and early warning of substation equipment operation is generally carried out through two methods: image recognition and voiceprint recognition. Compared with the image recognition method, voiceprint recognition can detect abnormal equipment operation earlier. Judging the operation state of substation equipment based on voiceprint data mainly relies on collecting and analyzing the sound signals generated during equipment operation, and using the differences in the sound characteristics emitted by the equipment under different states to monitor and diagnose the operation health status of the equipment through voiceprint recognition technology.

[0003] Currently, machine learning algorithms (such as support vector machines, neural networks, etc.) are generally used to model and train voiceprint data in known states, and a model that can distinguish different operation states is established to identify the operation state of the collected sound signals. In this process, not only a large amount of labeled data is relied on for training, but also the model needs to have good anti-noise performance. Especially when the background noise is complex or multiple fault types occur, the recognition accuracy is relatively low. In addition, there are also some methods that establish a target voiceprint database containing key voiceprint features. If the key voiceprint features appear in the collected sound signals, the operation state of the equipment can be further judged. This mode has stronger pertinence and lower initial training requirements, but it depends on the completeness and real-time update of the voiceprint database.

[0004] In the prior art, in the mode of using the target voiceprint database for feature matching, when matching the collected sound signals with the voiceprint features included in the established voiceprint database, there are factors such as insufficient accuracy in the matching target and matching method, resulting in the problem of low reliability of the judgment result.

[0005] In view of this, the present application is specifically proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for judging the operation state of substation equipment based on voiceprint data. The judgment method and system construct voiceprint databases for normal and abnormal operation states, respectively match the collected target sound signals with the two voiceprint databases, and comprehensively judge the operation state of the equipment according to the matching results of the two, so as to more accurately achieve the purpose of early warning of the operation state of substation equipment.

[0007] The embodiments of the present invention are implemented as follows: In a first aspect, a method for judging the operation state of substation equipment based on voiceprint data includes the following steps: constructing a first voiceprint database and a second voiceprint database, where the first voiceprint database refers to a feature database obtained by collecting audio signals when substation equipment is in a normal operation state and performing spectrum analysis, and the second voiceprint database refers to a feature database obtained by collecting audio signals when substation equipment is in a fault operation state and performing spectrum analysis; obtaining real-time voiceprint acquisition data of target substation equipment, dividing the real-time voiceprint acquisition data into time periods to obtain multiple segments of real-time voiceprint data, and performing spectrum feature analysis on each segment of the real-time voiceprint data to obtain feature vectors; comparing the feature vectors with the features in the first voiceprint database to obtain a first similarity value, and comparing the feature vectors with the features in the second voiceprint database to obtain a second similarity value; wherein, each of the first similarity values and each of the second similarity values are given time sequence marks, a first similarity value time sequence arrangement and a second similarity value time sequence arrangement are respectively generated based on all the first similarity values and all the second similarity values given time sequence marks, and the operation state of the target substation equipment is judged by combining the first similarity value time sequence arrangement and the second similarity value time sequence arrangement.

[0008] In some optional embodiments, the judging the operation state of the target substation equipment by combining the first similarity value time sequence arrangement and the second similarity value time sequence arrangement includes the following steps: using the first similarity value time sequence arrangement and the second similarity value time sequence arrangement to determine a difference sequence, identifying large value points and small value points in the difference sequence, determining the basic operation state of the target substation equipment according to the interval distribution form of all large value points, and determining the risk trend of the basic operation state according to the interval distribution form of the small value points; wherein, the large value point refers to a point where the absolute value of the corresponding value is not less than a first preset value, and the small value point refers to a point where the absolute value of the corresponding value is lower than a second preset value.

[0009] In some optional embodiments, after using the first similarity value time sequence arrangement and the second similarity value time sequence arrangement to determine the difference sequence, it further includes a step of adjusting the difference sequence: identifying the confidence level of each point in the difference sequence, screening out the points with a confidence level lower than a first preset confidence value to obtain an adjusted difference sequence.

[0010] In some optional embodiments, after screening out the points with a confidence level lower than the preset confidence value, it further includes the following steps: obtaining a confidence level compensation value, and comparing the confidence level compensation value with the preset confidence value after assigning the confidence level compensation value to the confidence level of each point, wherein the confidence level compensation value is obtained by using the absolute value of the corresponding value of the large value point and / or the small value point as the calculation basis.

[0011] In some optional embodiments, the confidence compensation value is obtained by using the absolute value of the corresponding value at the large value point and / or the small value point as the calculation basis, including the following situations: when the confidence compensation value is obtained by using the absolute value of the corresponding value at the large value point as the calculation basis: obtain the first absolute value of the corresponding value at each large value point, calculate the first deviation value of each first absolute value from 1, and calculate the confidence compensation value based on the mean value of all the first deviation values; or, when the confidence compensation value is obtained by using the absolute value of the corresponding value at the small value point as the calculation basis: obtain the second absolute value of the corresponding value at each small value point, calculate the second deviation value of each second absolute value from 0, and calculate the confidence compensation value based on the mean value of all the second deviation values; or, when the confidence compensation value is obtained by using the absolute values of the corresponding values at the large value point and the small value point as the calculation basis: obtain the first absolute value of the corresponding value at each large value point, calculate the first deviation value of each first absolute value from 1; obtain the second absolute value of the corresponding value at each small value point, calculate the second deviation value of each second absolute value from 0; calculate the confidence compensation value based on the mean value of all the first deviation values and all the second deviation values.

[0012] In some optional embodiments, when dividing the time periods of the real-time voiceprint acquisition data, the real-time voiceprint acquisition data is divided according to different time intervals respectively to obtain a difference sequence corresponding to each group of real-time voiceprint data; all the difference sequences are screened to determine a reference difference sequence, and the basic operating state of the target substation equipment is determined according to the interval distribution form of the large value points in the reference difference sequence, and the risk trend of the basic operating state is determined according to the interval distribution form of the small value points in the reference difference sequence.

[0013] In some optional embodiments, the step of screening all the difference sequences to determine a reference difference sequence includes the following steps: perform pairwise similarity comparison on all the difference sequences to determine a difference sequence with the highest similarity to all the other difference sequences as the reference difference sequence, wherein when performing pairwise similarity comparison on all the difference sequences, the interval distribution forms of the large value points and the small value points are used as the comparison basis.

[0014] In some optional embodiments, it further includes a review step of judging the operating state of the target substation equipment according to the reference difference sequence: judging the accuracy of the judgment result of the operating state of the target substation equipment by the reference difference sequence according to the actual measurement result of the operating state of the target substation equipment. If the accuracy meets the preset requirements, update the corresponding feature vector of the reference difference sequence to the corresponding first voiceprint database and / or the second voiceprint database; otherwise, perform a feature vector screening step; wherein, the corresponding feature vector refers to the feature vector corresponding to the similarity values in the first similarity value time series arrangement and the second similarity value time series arrangement that form the reference difference sequence.

[0015] In some optional embodiments, performing the feature vector screening step includes: performing point position reliability identification on all the difference sequences including the reference difference sequence, extracting the feature vectors of the points not lower than the second preset reliability value, and updating the extracted feature vectors to the corresponding first voiceprint database and / or the second voiceprint database, wherein the second preset reliability value is greater than the first preset reliability value.

[0016] In a second aspect, a substation equipment operating state judgment system based on voiceprint data includes: A first construction unit, which is used to construct a first voiceprint database and a second voiceprint database. The first voiceprint database refers to a feature database obtained by collecting audio signals of substation equipment in a normal operating state and performing spectrum analysis. The second voiceprint database refers to a feature database obtained by collecting audio signals of substation equipment in a faulty operating state and performing spectrum analysis. A first acquisition unit, which is used to acquire the real-time voiceprint acquisition data of the target substation equipment, divide the real-time voiceprint acquisition data into time periods to obtain multiple segments of real-time voiceprint data, and perform spectrum feature analysis on each segment of the real-time voiceprint data to obtain feature vectors. A first comparison unit, which is used to compare the similarity of the feature vectors with the features in the first voiceprint database to obtain a first similarity value, and compare the similarity of the feature vectors with the features in the second voiceprint database to obtain a second similarity value. A first judgment unit, which is used to assign a time sequence mark to each of the first similarity values and each of the second similarity values, generate a first similarity value time series arrangement and a second similarity value time series arrangement based on all the first similarity values and all the second similarity values assigned with time sequence marks, and combine the first similarity value time series arrangement and the second similarity value time series arrangement to perform the operating state judgment of the target substation equipment.

[0017] The beneficial effects of the embodiments of the present invention are: The method and system for judging the operation state of substation equipment based on voiceprint data provided by the embodiments of the present invention construct a first voiceprint database and a second voiceprint database. The first voiceprint database and the second voiceprint database respectively represent the voiceprint feature databases collected when the substation equipment is in normal operation state and abnormal operation state. Then, the spectral feature analysis is performed on the collected target equipment voice signal, and the obtained feature vectors are respectively subjected to similarity matching with the first voiceprint database and the second voiceprint database. According to the matching results of the two, comprehensively consider which voiceprint database each segment of the feature vector of the target voice signal is more similar to, and finally make an overall judgment on all the similarity results, so as to analyze the operation state of the target substation equipment, making the judgment result more reliable; Generally speaking, compared with the method of using a machine learning model to judge the fault state of substation equipment, the method and system for judging the operation state of substation equipment based on voiceprint data provided by the embodiments of the present invention reduce the requirements for the amount of target annotation and training, and are more targeted. Similarly, compared with the method of only matching the voiceprint features in the fault state or normal state, it can conduct comprehensive analysis and comparison from both the normal and fault aspects, making the analysis and judgment results more reliable, making up for the defect that although the voiceprint data matching is more targeted, the comprehensiveness is relatively insufficient, and improving the accuracy of the judgment result. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a flowchart of the main steps of the judgment method provided by the embodiments of the present invention; Figure 2 For Figure 1 It is a flowchart of one of the steps S400 of the main steps shown; Figure 3 For Figure 1 It is a flowchart of one of the steps S200 of the main steps shown; Figure 4 For Figure 3 It is a flowchart of the sub-steps of the step S220 shown; Figure 5 It is a modular schematic diagram of the judgment system provided by the embodiments of the present invention.

[0020] Icon: 500 - Judgment system; 510 - First construction unit; 520 - First acquisition unit; 530 - First comparison unit; 540 - First judgment unit. Detailed implementation manners

[0021] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0022] Therefore, the detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] It should be understood that the "system", "device" and / or "module" used in the present invention is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the said words can be replaced by other expressions.

[0024] As shown in the present invention and the claims, unless the context clearly indicates an exception, words such as "a", "an", "one" and / or "the" are not specifically singular, but may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of the clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. A method or device may also include other steps or elements.

[0025] Flowcharts are used in the present invention to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the operations before or after do not necessarily need to be executed precisely in sequence. On the contrary, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0026] Embodiment: In terms of the intelligent monitoring of the operation of substation equipment, voiceprint recognition can detect abnormal operating conditions faster than image recognition. Especially when the equipment emits abnormal sounds, it often indicates the early stage of a fault. Therefore, it is more suitable for the on-line monitoring of the operation status of substation equipment. For the voiceprint recognition mode, the target voice signal can be analyzed and judged through a deep learning model, but it depends on a large amount of previous training data and test data. If the training set and test set are insufficient, the model accuracy is relatively low. To address this problem, the operating status can also be judged by matching the voiceprint features of the target voice signal. This method is more targeted and does not require a large amount of data for annotation training. Only the completeness of the matching feature library needs to be considered. However, in this mode, on the one hand, the update and completeness of the feature library need to be considered, and on the other hand, the key is how to perform feature matching. For example, the method of determining normal or faulty when key voiceprint features appear is one-sided. Therefore, the rationality and comprehensiveness of the feature matching method need to be considered. To address the above problems, this embodiment provides a method for judging the operation status of substation equipment based on voiceprint data, which can continuously and similarly match with normal voiceprint data and abnormal voiceprint data, and finally comprehensively judge the operation status of the equipment according to the matching data sets of the two, making the judgment result more comprehensive, reasonable and reliable.

[0027] Specifically, please refer to Figure 1 , a method for judging the operation status of substation equipment based on voiceprint data provided by this embodiment includes the following steps: S100: Construct a first voiceprint database and a second voiceprint database. The first voiceprint database refers to a feature database obtained by collecting audio signals of substation equipment in a normal operating state and performing spectrum analysis. The second voiceprint database refers to a feature database obtained by collecting audio signals of substation equipment in a faulty operating state and performing spectrum analysis. This step means establishing the first voiceprint database and the second voiceprint database in advance. The two voiceprint databases can be data features obtained by spectrum analysis of the collected sound data when all in-use substation equipment is in a normal operating state and an abnormal operating state respectively.

[0028] Specifically, first, a high-sensitivity and wide-band sound sensor or microphone array can be selected to ensure that the subtle sound changes of substation equipment in different operating states can be captured. When the substation equipment is in a normal operating state (which can be judged manually), sound collection is carried out at set time intervals or triggered by specific events (such as startup, shutdown, etc.), and these data are stored to form the basic data of the first voiceprint database. When a device failure is detected, the sound information at this time is similarly recorded as the data source of the second voiceprint database. Next, in the spectrum analysis step, the original audio signal collected can be first subjected to necessary preprocessing (such as filtering to remove background noise, normalization, etc.), then the audio signal in the time domain is converted into a frequency-domain representation by applying the fast Fourier transform or other appropriate frequency-domain conversion methods, and finally, key feature parameters that can characterize the device state, such as peak frequency, frequency band energy distribution, harmonic components, etc. (as the feature data for storage in the database), are extracted from the spectrogram. It should be noted that the feature data obtained after the above spectrum analysis (annotated or labeled) also need to be stored in the database to form the first voiceprint database and the second voiceprint database respectively.

[0029] S200: Obtain the real-time voiceprint acquisition data of the target substation equipment, divide the real-time voiceprint acquisition data into time periods to obtain multiple segments of real-time voiceprint data, and perform spectrum feature analysis on each segment of the real-time voiceprint data to obtain feature vectors; this step indicates that after the voiceprint database is established, the matching operation of the sound signal begins. First, the sound signal of the target substation equipment collected is preprocessed, that is, the real-time voiceprint acquisition data of the target substation equipment is first divided into time periods (the segment of real-time voiceprint acquisition data is segmented according to different time intervals) to obtain multiple segments of continuously combined real-time voiceprint data, and then spectrum analysis is similarly performed on each small segment of real-time voiceprint data (the same spectrum analysis method as above) to obtain the feature vector corresponding to this small segment of real-time voiceprint data (that is, feature data such as peak frequency, frequency band energy distribution, harmonic components), and then the subsequent matching and comparison operations are carried out.

[0030] S300: Compare the feature vector with the features in the first voiceprint database to obtain a first similarity value, and compare the feature vector with the features in the second voiceprint database to obtain a second similarity value; this step means that the feature vectors corresponding to each small segment of real-time voiceprint data are respectively subjected to similarity matching with the first voiceprint database and the second voiceprint database. Specifically, all feature vectors are compared with the features in the first voiceprint database one by one to obtain different first similarity values, and then all feature vectors are also compared with the features in the second voiceprint database one by one to obtain different second similarity values. It should be noted that when performing feature similarity comparison, the similarity value results can be calculated by using the cosine similarity, Manhattan distance or correlation coefficient of two feature vectors.

[0031] S400: Assign a time sequence mark to each of the first similarity values and each of the second similarity values, generate a first similarity value time sequence arrangement and a second similarity value time sequence arrangement based on all the first similarity values and all the second similarity values assigned with time sequence marks, and combine the first similarity value time sequence arrangement and the second similarity value time sequence arrangement to judge the operating state of the target substation equipment; this step means that time information marks are assigned to each of the obtained first similarity values and each of the second similarity values, that is, all the first similarity values and the second similarity values are arranged respectively according to the time sequence (since the real-time voiceprint data is obtained continuously in segments along time, at this time, the corresponding time sequence marks during each similarity comparison can be synchronously assigned to represent the time order of each similarity value), and a first similarity value time sequence arrangement and a second similarity value time sequence arrangement are respectively generated. At this time, the operating state of the target substation equipment is judged by combining the similarity value distribution in the two sequences of the first similarity value time sequence arrangement and the second similarity value time sequence arrangement. For example, according to the positive and negative difference situation, the sum value situation, the specific numerical situation, etc. of the first similarity value and the second similarity value within the same time sequence mark, to further judge whether the target substation equipment is more inclined to the normal operation mode or the abnormal operation mode. If it is judged to be in the normal operation mode, what is its confidence level? Compared with the method of only unidirectionally judging whether it is normal or abnormal, the judgment perspective is more comprehensive and the result credibility is higher.

[0032] Through the above technical solution, key feature parameters are extracted after spectrum analysis of the historically collected sounds, and two comparison databases (normal and abnormal) are respectively formed. Then, for the soundprint data of the target substation equipment collected in real time, after preprocessing, time period division, and spectrum feature analysis, feature vectors are obtained, and are respectively compared with the features in the two soundprint databases to calculate the corresponding first similarity value and second similarity value. By assigning time sequence marks to these similarity values and generating a time sequence arrangement, the combined time sequence arrangements with two different performances can more comprehensively evaluate the operating state of the target equipment, determine whether it is closer to the normal or abnormal mode, and have a corresponding confidence evaluation basis, improving the accuracy of the operating state judgment.

[0033] When judging the operating state of the target substation equipment by using the time sequence arrangement of the first similarity value and the time sequence arrangement of the second similarity value, it can be accurately judged according to their combination situation, that is, it means further analyzing after merging the two sequences (that is, the way of adding and subtracting the values at the same point). For details, please refer to Figure 2 , the judging of the operating state of the target substation equipment by combining the time sequence arrangement of the first similarity value and the time sequence arrangement of the second similarity value includes the following steps: S410: Determine a difference sequence by using the time sequence arrangement of the first similarity value and the time sequence arrangement of the second similarity value; this step means directly merging the time sequence arrangement of the first similarity value and the time sequence arrangement of the second similarity value, and generating a difference sequence by using the difference situation of the similarity values of the two. It should be noted that each numerical point in this difference sequence is the difference between the first similarity value and the second similarity value under the same time sequence mark. Each numerical point can be a positive value (representing a result more biased towards the first similarity value, that is, a sound signal data more similar to the normal operating state), or a negative value (representing a result more biased towards the second similarity value, that is, a sound signal data more similar to the abnormal operating state). According to the magnitude of the absolute value of this numerical point, the bias law of the real-time soundprint data of all segments can be further judged to assist in further judging the operating state of the equipment.

[0034] S450: Identify the large-value points and small-value points in the difference sequence, determine the basic operating state of the target substation equipment according to the interval distribution form of all large-value points, and determine the risk trend of this basic operating state according to the interval distribution form of the small-value points; wherein, the large-value point refers to the point where the absolute value of the corresponding value is not less than the first preset value, and the small-value point refers to the point where the absolute value of the corresponding value is lower than the second preset value; this step means analyzing the operating state of the equipment by analyzing the magnitude and distribution of the absolute values of the values in the difference sequence. The large-value point in it refers to the point where the absolute value is greater than or equal to the first preset value (empirical value, such as designed to be 0.7, 0.8 or 0.9) at the corresponding value point in the difference sequence. Similarly, the small-value point refers to the point where the absolute value is less than or equal to the second preset value (empirical value, such as designed to be 0.1, 0.2 or 0.3) at the corresponding value point. The large-value point reflects the eigenvector with obvious tendency (normal or abnormal), and the small-value point reflects the eigenvector with fuzzy tendency. In particular, the operating state of the equipment is further judged according to the specific distribution forms of the large-value points and small-value points.

[0035] Furthermore, determining the basic operating state of the target substation equipment according to the interval distribution form of all large-value points means that whether the large-value points are continuously distributed or what kind of interval distribution is adopted can be used to preliminarily judge the basic operating state of the target substation equipment. For example, if large-value points with positive values appear continuously, it means that the equipment is probably operating normally. On the contrary, if large-value points with negative values appear continuously, it means that the equipment is probably operating abnormally; if large-value points with positive values appear at intervals and the intervals are small, it also means that the equipment is probably operating normally. If the intervals are large, it means that the equipment may be in a pre-abnormal state, etc., so as to carry out the preliminary judgment mode of the basic operating state of the target substation equipment. On this basis, the risk trend of this basic operating state can also be determined according to the interval distribution form of the small-value points, that is, whether the small-value points are continuously distributed means that the confidence of the above preliminary judgment is insufficient. On the contrary, the interval distribution of the small-value points means that the confidence of the above preliminary judgment is guaranteed, and the greater the interval, the higher the confidence.

[0036] Through the above technical solution, this method uses the positive and negative difference situations of the first similarity value and the second similarity value, the specific values of the differences and their distribution situations to further judge the operating state and reliability index of the target substation equipment, and can more comprehensively and accurately evaluate the health status and potential risks of the target substation equipment.

[0037] Please refer to again Figure 2, when considering using the difference sequence to judge the operating state of the target substation equipment, the numerical magnitudes and their distribution forms of each numerical point can both be used as one of the judgment reference bases. However, during the formation of the difference sequence, there may occasionally be a situation where a certain eigenvector has insufficient similarity with the features of the first voiceprint database and also has insufficient similarity with the features of the second voiceprint database. Especially when the completeness of the previous data in the first voiceprint database and the second voiceprint database is insufficient, this will result in the situation that the calculated first similarity value and second similarity value corresponding to this eigenvector are both not high (for example, both are less than 0.3). Although the corresponding difference value can be compared with the first preset value and the second preset value, the numerical point of this difference value itself has a problem of insufficient confidence, and participating in the subsequent operating state judgment process will reduce the accuracy of the judgment result. Therefore, to address this problem, after determining the difference sequence using the chronological arrangement of the first similarity value and the chronological arrangement of the second similarity value, it further includes the step of adjusting the difference series: S420: Identify the confidence level of each point in the difference sequence; this step means first making a confidence level judgment on each numerical point in the difference sequence, that is, making a judgment through the specific numerical values of the first similarity value and the second similarity value corresponding to this numerical point. For example, respectively judging the numerical magnitudes of the first similarity value and the second similarity value and the magnitude of their sum. If both the first similarity value and the second similarity value are less than, for example, 0.7, it means that the similarity with both of them is doubtful. If at this time the sum of the first similarity value and the second similarity value is less than, for example, 0.9, it means that the similarity judgment method is also doubtful. Therefore, it is necessary to first make a confidence level judgment on each numerical point in the difference sequence in order to obtain the confidence level value of each point.

[0038] S440: Screen out the points with a confidence level lower than the first preset confidence value to obtain an adjusted difference sequence; this step means making a comparison and judgment on the confidence level of each point in the difference sequence, screening out the points with a confidence level lower than the first preset confidence value from the difference sequence, and finally obtaining a screened and adjusted difference sequence. Using this adjusted difference sequence to judge the operating state of the target substation equipment can make the judgment result more reliable. It should be noted that the first preset confidence value is an empirical value, which can be a percentage value such as 90% or 95%, or a decimal value such as 0.9 or 0.95. It only needs to pre-calculate the confidence level of each numerical point in the difference sequence. This calculation method can be based on the numerical magnitudes of the first similarity value and the second similarity value and the magnitude of their sum. For example, the ratio of the larger of the two to 0.7 determines the basic value of the confidence level, and then the ratio of their sum to 1 determines the adjustment value of the confidence level. Finally, the final value of the confidence level is obtained based on the basic value and the adjustment value of the confidence level.

[0039] Through the above technical solution, a confidence evaluation and screening mechanism is introduced to optimize and adjust the difference sequence calculated from the first similarity value and the second similarity value. That is, by analyzing the first similarity value and the second similarity value of each point in the difference sequence, its confidence is judged, and the data points with confidence lower than the preset threshold are screened out, so as to obtain a more reliable and accurate adjusted difference sequence. This can effectively solve the problem that the reliability of the difference sequence is low due to the low similarity of feature vectors in the case of insufficient data completeness in the (first and second) voiceprint databases, and improve the accuracy and reliability of the judgment of the operating state of the target substation equipment based on the difference sequence.

[0040] On the basis of the above technical solution, especially when calculating the confidence according to the numerical sizes of the first similarity value and the second similarity value and the size of their sum, considering that there may be errors in the similarity calculation model or the similarity calculation logic process, resulting in a unified deviation in the similarity value calculation results. This deviation may occur either in the comparison process with the features of the first voiceprint database or in the comparison process with the features of the second voiceprint database. Therefore, it is necessary to correct this deviation. Please refer to Figure 2 , after screening out the points with confidence lower than the preset confidence value, the following steps are further included: S430: Obtain the confidence compensation value, and compare the confidence compensation value with the preset confidence value after assigning the confidence compensation value to the confidence of each point. Among them, the confidence compensation value is obtained by using the absolute value of the corresponding value of the large-value point and / or the small-value point as the calculation basis. This step means that by configuring the confidence compensation value as the basis for the above deviation correction, it is necessary to assign the confidence compensation value to the confidence result calculated for each point in the difference sequence, and then compare and screen the confidence with the preset confidence value after assigning the confidence compensation value. This method can try to correct the above deviation for a more authentic comparison, so as to retain the points that are mis-screened due to the similarity value comparison error, and make the obtained adjusted difference sequence more real and reliable for the judgment result of the operating state of the target substation equipment.

[0041] Among them, the confidence compensation value is obtained by taking the absolute value of the corresponding numerical value of the large value point and / or the small value point as the calculation basis. On the one hand, considering that the large value point is calculated by the situation that one of the first similarity value and the second similarity value is larger and the other is smaller, this situation shows that the similarity judgment result of the feature vector with the normal voiceprint feature or the abnormal voiceprint feature is relatively reliable, that is, the large value point is more referenceable. On this basis, the calculation method of the confidence compensation value can be determined by the numerical result of the large value point; of course, on the other hand, the small value point is calculated by the situation that the first similarity value and the second similarity value are close to each other. At this time, it means that it is impossible to judge more accurately whether the feature vector is more inclined to the normal voiceprint feature or the abnormal voiceprint feature. In this case, the small value point is a point with fuzzy confidence, which is more closely related to the above-mentioned deviation. On this basis, the calculation method of the confidence compensation value can be determined by the numerical result of the small value point; of course, in some other implementation methods, the calculation method of the confidence compensation value can also be determined by the numerical results of the large value point and the small value point at the same time.

[0042] Based on the above analysis of the confidence compensation value calculation method, the following three specific confidence compensation value calculation methods are provided in this embodiment, that is, the confidence compensation value is obtained by taking the absolute value of the corresponding value of the large value point and / or the small value point as the calculation basis, including the following specific calculation situations: First, when the confidence compensation value is obtained by taking the absolute value of the numerical value corresponding to the large value point as the calculation basis: obtaining the first absolute value of the numerical value corresponding to each large value point, calculating the first deviation value of each first absolute value from 1, and calculating the confidence compensation value based on the average of all first deviation values; Second, when the confidence compensation value is obtained by taking the absolute value of the numerical value corresponding to the small value point as the calculation basis: obtaining the second absolute value of the numerical value corresponding to each small value point, calculating the second deviation value of each second absolute value from 0, and calculating the confidence compensation value based on the average of all the second deviation values; Third, when the confidence compensation value is obtained by taking the absolute values ​​of the corresponding values ​​of the large value point and the small value point as the calculation basis: obtain the first absolute value of the corresponding value of each large value point, and calculate the first deviation value of each first absolute value and 1; obtain the second absolute value of the corresponding value of each small value point, and calculate the second deviation value of each second absolute value and 0; calculate the confidence compensation value based on the average of all first deviation values ​​and all second deviation values.

[0043] Through the above technical solution, the calculation and application of confidence compensation values ​​are introduced to correct the systematic deviations that may occur in the similarity calculation process, so as to more accurately evaluate the confidence of each data point and ensure that even when there are errors in the original similarity calculation, those data points that are actually of high value can still be retained. And for large-value points (the similarity judgment results of the feature vector and normal or abnormal voiceprint features are more reliable) and small-value points (the similarity judgment of the feature vector is fuzzy), the confidence compensation values ​​are calculated separately or jointly based on the absolute values ​​of their corresponding values, so as to correct the systematic deviations that may be caused by the similarity calculation model.

[0044] Considering that when dividing the time periods of the real-time voiceprint data, it can be divided by a single time interval, that is, the acoustic signal time periods are divided by the same time interval when constructing the first voiceprint database and the second voiceprint database, and the real-time voiceprint data is also divided by the same time interval; it can also be divided by different time intervals, that is, in addition to dividing by the first time interval (such as 100ms), it can also be divided again by the second time interval (such as 200ms), thereby increasing the richness of the features in the voiceprint database, but the time interval of the real-time voiceprint data needs to be consistent with the time interval of the feature division in the voiceprint database, so as to have a reference comparison basis. The single time interval operation is relatively simple and more targeted, while multiple time intervals make the feature data richer and have a more reliable basis for result calculation.

[0045] Therefore, in some embodiments, the first voiceprint database and the second voiceprint database can be divided into time periods by multiple time intervals, so as to obtain richer voiceprint feature data. In these embodiments, the real-time voiceprint data can also be divided into time periods by multiple time intervals, and then similarity comparisons are performed respectively, and finally the operating status of the target substation equipment is comprehensively predicted through multiple groups of comparison results, so that the judgment result is more reliable. For details, please refer to Figure 3 , when dividing the real-time voiceprint data into time periods, the following steps are included: S210: Divide the voiceprint real-time collection data according to different time intervals to obtain a difference sequence corresponding to each group of voiceprint real-time data; this step means that the voiceprint real-time collection data is divided into time periods in turn by configuring multiple time intervals (such as 100ms, 200ms, 300ms, etc.), and then a group of difference sequences is obtained each time the division is performed (also calculated and determined by the first similarity value time series arrangement and the first similarity value time series arrangement obtained in the above steps), and the difference sequences of all groups can be subsequently compared.

[0046] S220: Screen all difference sequences to determine a reference difference sequence, determine the basic operating status of the target substation equipment in the form of an interval distribution of large-value points in the reference difference sequence, and determine the risk trend of the basic operating status in the form of an interval distribution of small-value points in the reference difference sequence; this step represents representative screening of all difference sequences to determine a group of difference sequences as a reference difference sequence, which has a higher judgment accuracy when predicting the operating status of the target substation equipment. It should be noted that the reference difference sequence is also obtained by determining the basic operating status of the target substation equipment in the form of an interval distribution of large-value points in the reference difference sequence, and determining the risk trend of the basic operating status in the form of an interval distribution of small-value points in the reference difference sequence (refer to the relevant description of the above steps S410-S450, which will not be repeated here).

[0047] Through the above technical scheme, the data in the voiceprint database are divided into time periods using multiple time intervals, and voiceprint features of different granularities can be obtained. This not only increases the diversity of the feature database, but also the real-time voiceprint collection data is divided into different time intervals, and the difference sequence is calculated for the voiceprint data after each time period. Finally, a representative reference difference sequence is screened out, which can accurately determine the basic operating status of the target substation equipment and its risk trend.

[0048] In some embodiments, the reference difference sequence can be screened by selecting the most reliable one among all difference sequences, taking into account that the difference sequences obtained at different time intervals all have similarities in fault judgment as the screening basis. For details, please refer to Figure 4 The screening of all difference sequences to determine the reference difference sequence comprises the following steps: S221: perform a pairwise similarity comparison on all difference sequences, and use the interval distribution form of large-value points and small-value points as the comparison basis; this step means performing a similarity comparison on every two groups of all difference sequences, so as to facilitate the subsequent positioning of the group with the highest similarity to the remaining groups. When performing the similarity comparison, since the length intervals of each group of difference sequences are inconsistent, the interval distribution form of the respective large-value points and small-value points can be used as the comparison basis, that is, judging where the points of the large-value points and the small-value points are distributed, how long the interval in between is, etc. can be used as the basis for similarity comparison.

[0049] S222: Determine one of the difference sequences that has the highest similarity with all other difference sequences as a reference difference sequence; this step means that after the above-mentioned pairwise similarity comparison, a group of difference sequences that has the highest similarity with the remaining groups of difference sequences can be found as a reference difference sequence, so that it is more reliable in the subsequent basic operating status judgment of the target substation equipment.

[0050] Through the above technical solution, on the one hand, the completeness of the voiceprint database is increased by using a variety of time interval division methods, so that when the subsequent voiceprint real-time data is collected and then recognized and analyzed, there are more reference comparison objects, ensuring the accuracy and reliability of the final analysis results; on the other hand, the completeness of the voiceprint database can be increased by continuously updating accurate voiceprint feature vectors to the voiceprint database, and the accuracy of the judgment and prediction results can be improved. For example, in some implementation methods, the voiceprint features that have been determined to be accurate can be updated. Please refer to Figure 3 , further comprising a step of reviewing the operating status of the target substation equipment according to the reference difference sequence: S230: The accuracy of the operating status judgment result of the target substation equipment using the reference difference sequence is judged based on the actual measured operating status result of the target substation equipment, which means that it is possible to judge whether the actual measured operating status result of the target substation equipment is consistent with the predicted judgment result through the reference difference sequence through other methods such as on-site manual or machine inspection, and whether its accuracy reaches 100% or is less than 100%. If the accuracy meets the preset requirements (for example, the preset accuracy is 95%), the corresponding feature vector of the reference difference sequence is updated to the corresponding first voiceprint database and / or the second voiceprint database, otherwise the feature vector screening step is performed; that is, if the result of the operation status prediction through the reference difference sequence each time is accurate and reliable, the corresponding feature vector of the reference difference sequence (the feature vector corresponding to the analysis of the similarity values ​​in the first similarity value time series arrangement and the second similarity value time series arrangement that form the reference difference sequence) is updated to the first voiceprint database and / or the second voiceprint database accordingly, for example, the feature vector that is more similar to the first voiceprint database is updated to the first voiceprint database, and the feature vector that is more similar to the second voiceprint database is updated to the second voiceprint database, so as to ensure the accuracy, reliability and completeness of the voiceprint features of the voiceprint database.

[0051] Through the above technical solution, combined with the means of multiple time interval division and real-time update of accurate voiceprint feature vectors, the completeness and accuracy of the voiceprint database are greatly enhanced, especially through the composite step to judge the most reliable reference difference sequence, and use the reference difference sequence that meets the preset requirements as one of the sources of voiceprint database update. On this basis, if the preset requirements are not met, the source of voiceprint database update can also be guaranteed, that is, the requirement of voiceprint database completeness is met through the feature vector screening step. Specifically, the feature vector screening step includes: Perform point - position reliability identification on all the difference sequences including the reference difference sequence, that is, it means further analyzing all the determined difference sequences, analyzing the confidence level of each (numerical) point position in each group of difference sequences (the confidence - level judgment method refers to the step description of S410 - S450), so as to use the feature vector corresponding to the point position with higher confidence as the update source, that is, extracting the feature vector of the point position not lower than the second preset confidence value, and updating the extracted feature vector to the corresponding first voiceprint database and / or the second voiceprint database, where the second preset confidence value is greater than the first preset confidence value (its confidence - level requirement is higher). Through the foregoing technical solution, in order to ensure the completeness requirement of the voiceprint database, further processing is carried out on the update source, that is, in the case of not meeting the preset accuracy requirement, a feature - vector screening step is introduced. By deeply analyzing the point - position reliability in all difference sequences, preferentially selecting the feature vector corresponding to the point position with higher confidence for updating, especially when the confidence level of these point positions is not lower than the higher second preset confidence value, so as to ensure that even when the prediction result does not reach the expected accuracy, the voiceprint database can still obtain high - quality data supplementation.

[0052] In this embodiment, a substation equipment operation - state judgment system 500 based on voiceprint data is also provided. Please refer to Figure 5 the modular schematic diagram of the substation equipment operation - state judgment system 500 based on voiceprint data in [reference], which is mainly used to divide the function modules of the substation equipment operation - state judgment system 500 based on voiceprint data according to the embodiments of the above - mentioned method. For example, each function module can be divided, or two or more functions can be integrated into one processing module. The above - integrated module can be implemented in the form of hardware or in the form of a software function module. It should be noted that the division of modules in the present invention is schematic, only a logical function division, and there can be other division methods in actual implementation. For example, in the case of dividing each function module corresponding to each function, Figure 5 only a system / device schematic diagram is shown. Among them, the substation equipment operation - state judgment system 500 based on voiceprint data can include a first construction unit 510, a first acquisition unit 520, a first comparison unit 530, and a first judgment unit 540. The functions of each unit module will be elaborated below.

[0053] The first construction unit 510 is used to construct a first voiceprint database and a second voiceprint database. The first voiceprint database refers to a feature database obtained by collecting audio signals when the substation equipment is in a normal operation state and performing spectrum analysis. The second voiceprint database refers to a feature database obtained by collecting audio signals when the substation equipment is in a fault operation state and performing spectrum analysis; A first acquisition unit 520, which is configured to acquire real-time voiceprint acquisition data of target substation equipment, divide the real-time voiceprint acquisition data into time periods to obtain multiple segments of real-time voiceprint data, perform spectral feature analysis on each segment of the real-time voiceprint data, and obtain feature vectors; in some embodiments, the first acquisition unit 520 is further configured to divide the real-time voiceprint acquisition data according to different time intervals respectively to obtain a difference sequence corresponding to each group of real-time voiceprint data; screen all the difference sequences to determine a reference difference sequence, determine the basic operating state of the target substation equipment in the form of the interval distribution of the large value points in the reference difference sequence, and determine the risk trend of the basic operating state in the form of the interval distribution of the small value points in the reference difference sequence; is also configured to perform pairwise similarity comparison on all the difference sequences to determine a difference sequence with the highest similarity to the rest of all the difference sequences as the reference difference sequence, wherein when performing pairwise similarity comparison on all the difference sequences, the interval distribution forms of the large value points and the small value points are used as the comparison basis; and is configured to judge the accuracy of the operation state judgment result of the reference difference sequence for the target substation equipment according to the actual measurement result of the operation state of the target substation equipment. If the accuracy meets the preset requirements, update the corresponding feature vector of the reference difference sequence to the corresponding first voiceprint database and / or the second voiceprint database, otherwise perform the feature vector screening step.

[0054] A first comparison unit 530, which is configured to perform similarity comparison between the feature vector and the features in the first voiceprint database to obtain a first similarity value, and perform similarity comparison between the feature vector and the features in the second voiceprint database to obtain a second similarity value; The first judgment unit 540 is configured to assign a timing tag to each of the first similarity values and each of the second similarity values, generate a first similarity value timing arrangement and a second similarity value timing arrangement respectively based on all the first similarity values and all the second similarity values assigned with timing tags, and perform an operation state judgment on the target substation equipment by combining the first similarity value timing arrangement and the second similarity value timing arrangement. In some embodiments, the first judgment unit 540 is further configured to determine a difference sequence by using the first similarity value timing arrangement and the second similarity value timing arrangement, identify large value points and small value points in the difference sequence, determine the basic operation state of the target substation equipment according to the interval distribution form of all the large value points, and determine the risk trend of the basic operation state according to the interval distribution form of the small value points; and is further configured to identify the confidence level of each point in the difference sequence, screen out the points with a confidence level lower than a first preset confidence value to obtain an adjusted difference sequence; and is further configured to obtain a confidence level compensation value, compare the confidence level obtained by assigning the confidence level compensation value to each point with the preset confidence value respectively, where the confidence level compensation value is obtained by using the absolute value of the corresponding value of the large value point and / or the small value point as a calculation basis.

[0055] In the above embodiments, the more specific working processes of the functional units can refer to the corresponding content disclosed in the foregoing method embodiments. In addition, the functional units can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless manner. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)).

[0056] Embodiments of the present application are described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a means for implementing the functions specified in multiple blocks.

[0057] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction means that implements the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a means for implementing the functions specified in multiple blocks.

[0058] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or a means for implementing the functions specified in multiple blocks.

[0059] Obviously, those skilled in the art can make various modifications and variations to the embodiments of the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A method for judging the operating status of substation equipment based on voiceprint data, characterized in that: The steps include: Constructing a first voiceprint database and a second voiceprint database, wherein the first voiceprint database refers to a feature database obtained by performing spectrum analysis on audio signals collected when the substation equipment is in a normal operating state, and the second voiceprint database refers to a feature database obtained by performing spectrum analysis on audio signals collected when the substation equipment is in a faulty operating state; Acquire the real-time voiceprint data of the target substation equipment, divide the real-time voiceprint data into time periods, obtain multiple segments of real-time voiceprint data, perform spectrum feature analysis on each segment of the real-time voiceprint data and obtain a feature vector; Performing a similarity comparison between the feature vector and the features in the first voiceprint database and obtaining a first similarity value, and performing a similarity comparison between the feature vector and the features in the second voiceprint database and obtaining a second similarity value; Among them, each of the first similarity values ​​and each of the second similarity values ​​are assigned a timing mark, and a first similarity value timing arrangement and a second similarity value timing arrangement are respectively generated based on all the first similarity values ​​and all the second similarity values ​​assigned with the timing mark, and the operating status of the target substation equipment is judged in combination with the first similarity value timing arrangement and the second similarity value timing arrangement.

2. The method for determining the operating status of substation equipment based on voiceprint data according to claim 1 is characterized in that: The step of combining the first similarity value time series arrangement and the second similarity value time series arrangement to judge the operating status of the target substation equipment comprises the following steps: The first similarity value time series arrangement and the second similarity value time series arrangement are used to determine a difference sequence, and the large value points and small value points in the difference sequence are identified. The basic operating status of the target substation equipment is determined according to the interval distribution form of all large value points, and the risk trend of the basic operating status is determined according to the interval distribution form of the small value points; wherein the large value point refers to a point whose absolute value of the corresponding numerical value is not less than a first preset value, and the small value point refers to a point whose absolute value of the corresponding numerical value is lower than a second preset value.

3. The method for determining the operating status of substation equipment based on voiceprint data according to claim 2 is characterized in that: After determining the difference sequence using the first similarity value time series arrangement and the second similarity value time series arrangement, the method also includes the step of adjusting the difference series: identifying the confidence of each point in the difference sequence, screening out the points whose confidence is lower than the first preset confidence value, and obtaining the adjusted difference sequence.

4. The method for determining the operating status of substation equipment based on voiceprint data according to claim 3 is characterized in that: After screening out the points whose confidence is lower than the preset confidence value, the following steps are also included: obtaining a confidence compensation value, assigning the confidence compensation value to the confidence of each point respectively, and then comparing it with the preset confidence value, wherein the confidence compensation value is obtained by using the absolute value of the corresponding numerical value of the large-value point and / or the small-value point as the calculation basis.

5. The method for determining the operating status of substation equipment based on voiceprint data according to claim 4 is characterized in that: The confidence compensation value is obtained by taking the absolute value of the numerical value corresponding to the large value point and / or the small value point as the calculation basis, including the following situations: When the confidence compensation value is obtained by taking the absolute value of the numerical value corresponding to the large value point as the calculation basis: obtaining the first absolute value of the numerical value corresponding to each large value point, calculating the first deviation value of each first absolute value from 1, and calculating the confidence compensation value based on the average of all the first deviation values; Alternatively, when the confidence compensation value is obtained by taking the absolute value of the numerical value corresponding to the small value point as the calculation basis: obtaining the second absolute value of the numerical value corresponding to each small value point, calculating the second deviation value of each second absolute value from 0, and calculating the confidence compensation value based on the average of all the second deviation values; Alternatively, when the confidence compensation value is obtained by taking the absolute values ​​of the corresponding values ​​of the large value point and the small value point as the calculation basis: obtain the first absolute value of the corresponding value of each large value point, and calculate the first deviation value of each first absolute value and 1; obtain the second absolute value of the corresponding value of each small value point, and calculate the second deviation value of each second absolute value and 0; calculate the confidence compensation value based on the average of all first deviation values ​​and all second deviation values.

6. The method for determining the operating status of substation equipment based on voiceprint data according to claim 3 or 5, characterized in that: When dividing the voiceprint real-time collection data into time periods, the voiceprint real-time collection data are divided according to different time period intervals to obtain a difference sequence corresponding to each group of voiceprint real-time data; all difference sequences are screened to determine a reference difference sequence, and the basic operating status of the target substation equipment is determined in the form of an interval distribution of large-value points in the reference difference sequence, and the risk trend of the basic operating status is determined in the form of an interval distribution of small-value points in the reference difference sequence.

7. The method for determining the operating status of substation equipment based on voiceprint data according to claim 6 is characterized in that: The screening of all difference sequences to determine the reference difference sequence includes the following steps: performing a pairwise similarity comparison on all difference sequences, and determining a group of difference sequences with the highest similarity to all other difference sequences as a reference difference sequence, wherein, when performing a pairwise similarity comparison on all difference sequences, the interval distribution form of large-value points and small-value points is used as the comparison basis.

8. The method for determining the operating status of substation equipment based on voiceprint data according to claim 6 is characterized in that: The method further includes a step of reviewing the operating status of the target substation equipment according to the reference difference sequence: The accuracy of the operating status judgment result of the target substation equipment by the reference difference sequence is judged according to the actual measured results of the operating status of the target substation equipment. If the accuracy meets the preset requirements, the corresponding feature vector of the reference difference sequence is updated to the corresponding first voiceprint database and / or second voiceprint database, otherwise a feature vector screening step is performed; wherein the corresponding feature vector refers to the feature vector corresponding to the similarity value in the first similarity value time series arrangement and the second similarity value time series arrangement that form the reference difference sequence.

9. The method for determining the operating status of substation equipment based on voiceprint data according to claim 8 is characterized in that: The steps of feature vector screening include: Perform point position reliability identification on all the difference sequences including the reference difference sequence, extract feature vectors from points that are not lower than a second preset reliability value, and update the extracted feature vectors to the corresponding first voiceprint database and / or the second voiceprint database, wherein the second preset reliability value is greater than the first preset reliability value.

10. A substation equipment operation status judgment system based on voiceprint data, characterized in that: include: A first construction unit, which is used to construct a first voiceprint database and a second voiceprint database, wherein the first voiceprint database refers to a feature database obtained by performing spectrum analysis on audio signals collected when the substation equipment is in a normal operating state, and the second voiceprint database refers to a feature database obtained by performing spectrum analysis on audio signals collected when the substation equipment is in a faulty operating state; A first acquisition unit is used to acquire the real-time voiceprint data of the target substation equipment, divide the real-time voiceprint data into time periods, obtain multiple segments of real-time voiceprint data, perform spectrum feature analysis on each segment of the real-time voiceprint data and obtain a feature vector; a first comparison unit, configured to perform a similarity comparison between the feature vector and features in the first voiceprint database and obtain a first similarity value, and perform a similarity comparison between the feature vector and features in the second voiceprint database and obtain a second similarity value; A first judgment unit is used to assign a timing mark to each of the first similarity values ​​and each of the second similarity values, generate a first similarity value timing arrangement and a second similarity value timing arrangement based on all the first similarity values ​​and all the second similarity values ​​assigned with the timing marks, and judge the operating status of the target substation equipment in combination with the first similarity value timing arrangement and the second similarity value timing arrangement.

Images