CN111325485B - Light-weight gradient elevator power quality disturbance identification method considering internet-of-things bandwidth constraint - Google Patents
Light-weight gradient elevator power quality disturbance identification method considering internet-of-things bandwidth constraint Download PDFInfo
- Publication number
- CN111325485B CN111325485B CN202010204617.0A CN202010204617A CN111325485B CN 111325485 B CN111325485 B CN 111325485B CN 202010204617 A CN202010204617 A CN 202010204617A CN 111325485 B CN111325485 B CN 111325485B
- Authority
- CN
- China
- Prior art keywords
- disturbance
- power quality
- internet
- things
- time domain
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention relates to a method for identifying the power quality disturbance of a lightweight gradient elevator considering the bandwidth constraint of an internet of things, which is characterized by comprising the following steps of: the method comprises the steps of constructing a power quality analysis system architecture based on a typical Internet of things communication mode, taking power quality time domain feature efficient edge extraction considering Internet of things bandwidth constraint, constructing an optimal feature set and an optimal classifier, analyzing a power quality disturbance efficient recognition effect and the like, carrying out signal acquisition on 17 types of power quality disturbance signals, training a LightGBM disturbance recognition classifier through acquired data, and analyzing the power quality disturbance signals after training is completed to judge the type of the power quality disturbance. The LightGBM has high classification precision, can be subjected to parallelization processing, has strong portability and can effectively prevent overfitting. The method has the advantages of being scientific and reasonable, strong in adaptability, high in practical value, rapid in disturbance identification, high in accuracy and the like.
Description
Technical Field
The invention relates to a power quality disturbance identification method, in particular to a power quality disturbance identification method for a lightweight gradient elevator considering tie bandwidth constraint, which is applied to the online identification of power quality disturbance.
Background
In the traditional electric energy quality disturbance identification, a disturbance signal is collected by electric energy quality acquisition equipment, and then the original disturbance signal is transmitted to an upper system for disturbance characteristic extraction and identification. After distributed power sources and power electronic equipment in a power distribution network are applied in a large quantity, the communication pressure of the network is increased by the transmission of the electric energy quality data with the high sampling rate of the mass monitoring points.
The development of the internet of things technology provides a new idea for the construction of a power quality monitoring system containing a large number of monitoring points. Through low-cost edge equipment, the electric energy quality of a large amount of nodes in a power distribution network system can be effectively sensed. But is limited by the bandwidth of a typical Internet of things communication mode, and original data is difficult to be directly transmitted to an upper system by adopting Internet of things communication modes such as Lora, NB-IoT and the like; and the bottom layer equipment is limited by cost, so that complex disturbance identification is difficult to directly develop. Therefore, edge (acquisition) side disturbance feature extraction is carried out based on low-cost edge equipment; and then, only transmitting the effective disturbance classification characteristics to an upper system for analysis. Therefore, the system communication data volume is reduced, and the application of the Internet of things in the field of power quality monitoring and analysis is effectively promoted.
From the aspect of feature extraction, common signal processing methods include S transformation, empirical mode decomposition, ensemble empirical mode decomposition, variation mode decomposition, and the like. The S-transform provides a large number of time-frequency domain features. However, the calculation amount is large and is easily affected by parameters such as the window width factor. Empirical mode decomposition, collective empirical mode decomposition and variational mode decomposition methods decompose the interference signal into multiple layers and extract features from each layer of signal, but are susceptible to the number of decomposition layers. Although the method can achieve a good effect, the time complexity and the space complexity are high, and when the feature extraction calculation work is transferred to the edge side equipment, the low-complexity calculation requirement of the low-cost edge calculation equipment cannot be met. The method can be used for identifying numerous characteristics of power quality disturbance, and increases the edge calculation amount and the complexity of an upper system. Therefore, the feature dimension of the original feature set needs to be reduced by a feature selection method. Common feature selection methods are roughly classified into three types: embedded and wrapped, filtering. The embedded method integrates the characteristic selection process and the classifier training process, and automatically selects the characteristics in the classifier training process; the wrapped method directly uses the performance of the final used learner as an evaluation standard of the feature subset, but the optimization efficiency is low. The filtering method utilizes a statistical means, is irrelevant to a subsequent classifier, but needs a reliable classifier to support a later optimal feature subset determination link. From the design perspective of the classifier, a support vector machine, a decision tree, a self-adaptive enhancement, a sparse automatic encoder and the like are widely applied to the field of electric energy quality disturbance identification, and a good effect is achieved. However, the sparse automatic encoder is an algorithm under a deep learning framework, can automatically extract characteristics of power quality disturbance for classification, is an analysis method under a traditional power quality analysis framework, needs original data to mine relevant characteristics, has high requirements on communication bandwidth, and is not beneficial to application under a system framework in a mode of internet of things communication. The decision tree accuracy is superior to that of a support vector machine under the condition that the same training and testing sample set is adopted. But the classification threshold setting depends on training samples, and the generalization capability is poor. The self-adaptive enhancement noise immunity is good, the influence of the over-fitting problem is small, and the generalization capability is better compared with that of a decision tree. But the training time is long and the memory consumption is large. The method is not suitable for the analysis requirement of the mass power quality disturbance big data at present.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides the method for identifying the power quality disturbance of the light-weight gradient elevator, which is scientific and reasonable, strong in adaptability, high in practical value, quick in disturbance identification and high in accuracy and takes the tie bandwidth constraint into consideration.
The purpose of the invention is realized by the following technical means: a method for recognizing the power quality disturbance of a lightweight gradient elevator considering the bandwidth constraint of an Internet of things is characterized by comprising the following steps of:
1) construction of electric energy quality analysis system architecture based on typical Internet of things communication mode
The electric energy quality acquisition device is usually installed at an electric vehicle charging station, a photovoltaic power station, a wind power plant, a transformer substation, an electrified railway, an industrial load, a residential load side and a massive distributed power grid-connected point position, acquires relevant characteristics by using low-cost electric energy quality edge perception and characteristic extraction equipment, and realizes communication on the basis of analyzing an electric energy quality disturbance identification framework based on internet of things communication in a LoRa and NB-IoT typical internet of things communication mode so as to meet the bandwidth constraint of a narrow-band internet of things;
2) power quality time domain feature efficient edge extraction considering Internet of things bandwidth constraints
The massive power quality data of the power distribution network are difficult to transmit through a limited bandwidth Internet of things channel, so that the characteristic is used for replacing an original signal to upload, in the characteristic extraction process, after time domain segmentation can be carried out on the original signal, time domain characteristics are directly extracted, and the rule is as follows: selecting 20 time domain characteristics: harmonizing an average number, a variance, a standard deviation, an average difference, skewness, kurtosis, entropy, Shannon entropy, Renyi entropy, Tsallis entropy, a mean value, a root mean square, a maximum value, a minimum value, a difference between the maximum value and the minimum value, a sum of the maximum value and the minimum value, energy, a normalized amplitude factor, 1/4 period energy falling amplitude and 1/4 period energy rising amplitude, and respectively calculating various time domain characteristic values for each divided interval; then, respectively selecting maximum and minimum parameters in the same class of characteristic values in each time domain interval, wherein 40-dimensional characteristics are used for constructing an original characteristic set, and in addition, 20-dimensional characteristics of various time domains are integrally calculated for an original signal, and finally, 60-dimensional time domain characteristics are extracted; the experiment is completed on a microcomputer configured with Intel Core i5-7500CPU and DDR 42666 Hz12GB memory, 17 classes of PQ disturbance signals are generated through simulation, and the PQ disturbance signals comprise a voltage standard signal C0, a voltage temporary drop C1, a voltage temporary rise C2, a voltage interruption C3, a flicker C4, a transient oscillation C5, a harmonic C6, a voltage shear mark C7, a voltage spike C8, a harmonic containing temporary drop C9, a harmonic containing temporary rise C10, an oscillation containing temporary drop C11, an oscillation containing temporary rise C12, a flicker containing temporary drop C13, a flicker containing temporary rise C14, a flicker containing harmonic C15 and a break containing harmonic C16;
3) construction of optimal feature set and optimal classifier
Establishing a lightweight Gradient Boosting Machine (LightGBM) classifier in Spyder software under an Anaconda development environment, and performing LightGBM parameter optimization by using 10-fold cross validation and Bayesian optimization to obtain a primary LightGBM classifier by aiming at the minimum classification error rate obtained in the classification process of the LightGBM classifier; obtaining importance values of all time domain features through a gradient lifting algorithm of a primary LightGBM classifier, sequencing the importance values from high to low, sequentially adding the features into a feature subset according to an order of descending the importance, calculating the identification accuracy of the classifier after parameter optimization under the feature subset, repeating the process until all the features are added into the feature set, finally determining an optimal feature subset according to the highest identification accuracy, and inputting 600 groups of electric energy quality signals into the primary LightGBM classifier for further training after performing feature calculation on the features belonging to the optimal feature vector set through a time domain feature formula in the step 2) from 17 types of electric energy quality disturbance signals to obtain the trained LightGBM disturbance identification classifier;
4) electric energy quality disturbance efficient recognition effect analysis
And (3) passing the collected power quality disturbance signals through the time domain feature formula in the step 2), extracting the time domain features of the optimal feature vector set obtained in the step 3), and inputting the extracted features into the LightGBM disturbance recognition classifier trained in the step 3) to obtain the reflected power quality problem.
Further, the high bandwidth of the LoRa and NB-IoT typical internet-of-things communication method in step 1) is 100 Kbps.
Further, the fundamental frequency of the voltage standard signal in the step 2) is 50Hz, and the sampling rate is 6400 Hz.
Through the design scheme, the invention has the following beneficial effects:
according to the method for recognizing the power quality disturbance of the lightweight gradient elevator considering the tie bandwidth constraint, 17 types of power quality disturbance signals are subjected to signal acquisition, the LightGBM disturbance recognition classifier is trained through the acquired data, and the power quality disturbance signals are analyzed after the training is finished and are used for judging the type of the power quality disturbance. The LightGBM has high classification precision, can be subjected to parallelization processing, has strong portability and can effectively prevent overfitting. The method has the advantages of being scientific and reasonable, strong in adaptability, high in practical value, rapid in disturbance identification, high in accuracy and the like.
Drawings
FIG. 1 is a diagram of a narrowband Internet of things communication-based power quality analysis system architecture for a lightweight gradient elevator power quality disturbance identification method considering Internet of things bandwidth constraints according to the present invention;
FIG. 2 is a diagram showing the time domain division effect of 8 kinds of composite disturbance in the method for identifying the power quality disturbance of the lightweight gradient elevator considering the bandwidth constraint of the Internet of things;
FIG. 3 is a graph of the importance of the Split characteristics of the method for identifying the power quality disturbance of the lightweight gradient elevator considering the bandwidth constraint of the Internet of things of the invention;
FIG. 4 is a diagram of a forward feature selection process of the method for identifying power quality disturbance of a lightweight gradient elevator in consideration of tie bandwidth constraints according to the present invention;
fig. 5 is a diagram of accuracy of power quality disturbance identification under mixed noise in the method for identifying power quality disturbance of a lightweight gradient elevator considering tie bandwidth constraints.
Detailed Description
The invention is further described with reference to the following figures and detailed description:
the invention relates to a method for identifying the power quality disturbance of a lightweight gradient elevator considering the bandwidth constraint of an internet of things, which comprises the following steps:
1) establishment of electric energy quality analysis system architecture based on typical Internet of things communication mode
The electric energy quality acquisition device is generally installed at the positions of an electric vehicle charging station, a photovoltaic power station, a wind power plant, a transformer substation, an electrified railway, an industrial load, a resident load side, a massive distributed power supply grid-connected point and the like. The method has the advantages that the low-cost electric energy quality edge perception and feature extraction equipment collects relevant features, a traditional original signal uploading mode is replaced, and the communication cost of the system can be effectively reduced.
If communication is realized by adopting a typical reliable internet-of-things communication mode such as LoRa and NB-IoT, the internet-of-things bandwidth limitation constraint must be considered. As shown in fig. 1, the invention considers the application in the electric energy quality disturbance identification architecture based on internet of things communication, so as to satisfy the narrowband internet of things bandwidth constraint. Taking the common internet of things communication methods LoRa and NB-IoT as an example, the highest bandwidth is 100 Kbps.
In the design of the related system structure, in order to ensure the effective application of the internet of things technology, the bandwidth limitation of the communication mode needs to be considered, and the related system needs to be designed. Edge acquisition and feature calculation equipment is arranged on an acquisition side to acquire power quality disturbance signals and extract features; uploading relevant disturbance characteristics to an upper system through a narrow-band Internet of things; and the upper system completes the analysis of the power quality disturbance signal. It should be noted that the present invention mainly researches an edge feature extraction and high-efficiency disturbance identification technology of an upper system, which meets the above requirement of the architecture communication bandwidth limitation. The system architecture is presented merely as an analytical background to the invention.
2) Power quality time domain feature efficient edge extraction considering Internet of things bandwidth constraints
Massive power quality data of the power distribution network are difficult to transmit through a limited bandwidth Internet of things channel. Thus, the original signal is uploaded in place of the features. When the feature extraction calculation work is transferred to the edge side device, the low-complexity calculation requirement of the low-cost edge calculation device cannot be met. Therefore, in the feature extraction process, the time domain feature can be directly extracted after the time domain segmentation is carried out on the original signal. Therefore, the calculation complexity is reduced, and the requirement of low-cost terminal edge calculation is met. The rule is as follows: calculating various time domain characteristic values of each divided interval respectively; and then, respectively selecting maximum and minimum parameters in the same class of characteristic values in each time domain interval, wherein the total 40-dimensional characteristics are used for constructing an original characteristic set. In addition, various types of time domain features are calculated for the whole original signal in 20 dimensions. Finally, 60-dimensional time domain features are extracted. F1-F60 represent the indices of 60 time domains, as shown in Table 1:
TABLE 1 feature classes and designations
As shown in FIG. 5, the experiments were all performed on a microcomputer configured as Intel Core i5-7500CPU, DDR 42666 Hz12GB memory. Simulating and generating a 17-class PQ disturbance signal, comprising: the voltage standard signal C0, the voltage sag C1, the voltage sag C2, the voltage interruption C3, the flicker C4, the transient oscillation C5, the harmonic C6, the voltage shear mark C7, the voltage spike C8, the harmonic containing sag (C9, the harmonic containing sag C10, the oscillation containing sag C11, the oscillation containing sag C12, the flicker containing sag C13, the flicker containing sag C14, the flicker containing harmonic C15 and the interruption harmonic containing C16. the fundamental frequency of the voltage standard signal is 50Hz, and the sampling rate is 6400 Hz.
Taking the 1-cycle signal time domain division unit as an example, the extraction of the disturbance feature is performed. The time domain division result of the 8 kinds of composite power quality disturbance signals is shown in fig. 2, and it can be known from the figure that in different time domain division ranges, when the disturbance occurs, ends, lasts in various special periods, the waveforms of different disturbance types are obviously different from those of other disturbances. Therefore, the difference can be characterized through the time domain characteristics, and the disturbance classification is realized.
3) Construction of optimal feature set and optimal classifier
And (3) taking 600 groups of the 17-class power quality disturbance signals as training samples of the classifier, performing feature counting on the training samples, inputting the feature vectors belonging to the optimal feature subset into the LightGBM for training, and obtaining the LightGBM disturbance recognition classifier which is finally trained. In order to improve the classification accuracy and classification effect of the LightGBM, parameters in the LightGBM model need to be optimized. The results after optimization are shown in table 2.
Table 2 LightGBM hyper-parameter set settings
The LightGBM classifier uses a gradient boosting algorithm, and the more an attribute is used to construct a decision tree in a model, the higher its importance is. Therefore, after the lifting tree is created, the importance value of each attribute can be directly obtained, which measures the value of the feature in the construction of the lifting decision tree, and the larger the importance value is, the greater the contribution is made in the state recognition. With the feature importance value as a main judgment basis, as shown in fig. 3, 60 time domain features are subjected to importance value sorting, the features are sequentially added into a feature subset according to the descending order of importance for the feature importance indexes of each classifier, each feature is added, the identification accuracy of the classifier after parameter optimization under the feature subset is calculated, and the process is repeated until all the features are added into the feature set. And finally, determining an optimal feature subset according to the highest recognition accuracy, wherein the forward feature selection process is shown in fig. 4, and the optimal feature subset dimension and the highest accuracy corresponding to the feature importance index are shown in table 3.
TABLE 3 feature selection results
4) Electric energy quality disturbance recognition effect analysis
Amount of edge calculation
In order to analyze the efficiency of feature extraction compared with the conventional method, table 4 shows the feature extraction time required by the edge side to process 1 group of disturbance signals under different feature extraction methods under the conditions that the sampling rate is 6400Hz and the sampling waveform is 50 cycles.
TABLE 4 feature extraction time
As can be seen from the analysis of Table 4, compared with the S transformation, the empirical mode decomposition, the wavelet transformation and the like, the feature extraction method adopted by the invention has no signal processing time, and the overall feature extraction time is far lower than that of the comparison method. In addition, the signal processing time is positively correlated to the signal sampling rate. With the increase of the sampling rate of the signal acquisition equipment, the existing feature extraction method can not meet the requirement of real-time monitoring of the power quality disturbance at a high sampling rate. Therefore, the method simplifies the feature extraction process while ensuring high classification accuracy, thereby improving the feature extraction efficiency and having important practical significance.
Amount of data to be transmitted
For the applicability of the invention under the constraint of the bandwidth of the Internet of things communication, the transmission edge of the invention is compared with the bandwidth required by the traditional original signal transmission mode after the optimal characteristic subset and the original characteristic set are extracted and transmitted. Assuming that the sampling rate is 6400Hz, a single disturbance samples a waveform for 50 cycles, and in an extreme case, a power quality event occurs 1 time per second inside the system, 1 group of disturbance optimal feature subsets, an original feature set, a bandwidth required by an original signal, and the like, which need to be uploaded are shown in table 5.
Table 5 comparison of data volumes for different transmission types
As can be seen from table 5, assuming that the data transmission delay is controlled within 1s, when the original signal is uploaded on the edge side, the initial measurement bandwidth requirement is about 933.9 Kbps. When uploading the set of raw features, it was 5.8 Kbps. When uploading the optimal feature set, it is 4 Kbps. Compared with uploading the original signal, the invention reduces the bandwidth requirement by 99.6%; compared with uploading the original feature set, the method reduces the bandwidth requirement by 31.0%. The invention meets the bandwidth constraints of the IoT communication modes such as LoRa, NB-LOT and the like. And the uploading of the optimal feature set can further reduce the edge calculation complexity and the data internet-of-things transmission pressure in the distribution network.
Lightweight PQ disturbance classification model recognition effect analysis
In order to comprehensively verify the effectiveness of the invention under different noise environments, Matlab R2016b is used to generate 600 groups of disturbance signals in a simulation mode under 50dB to 20dB mixed noise, and the disturbance signals are used for training LightGBM under the optimal feature subset. Under the environment of mixed noise of 50dB to 20dB and specific signal-to-noise ratios of 50dB, 40dB, 30dB and 20dB, 200 groups of all the disturbing signals are generated in a simulation mode and used for constructing a test set. The disturbing signal identification result under the mixed noise environment of 50dB to 20dB is shown in figure 5, the identification accuracy of the invention is over 99.53% under the environment with the signal-to-noise ratio of over 30dB, and the identification accuracy can still reach 94.85% under the environment with the low signal-to-noise ratio of 20 dB. The method has the advantages of obviously improving the characteristic extraction efficiency and having good identification accuracy and anti-noise capability.
Time domain segmentation scale analysis
Through statistical experiments, the corresponding feature extraction time, the optimal feature subset dimension and the accuracy rate under different time domain segmentation intervals (1/4 periods, 1/2 periods and 1 period) are analyzed. As shown in table 6.
TABLE 6 impact of temporal partition size
As can be seen from table 6, for the disturbing signals with different noise levels, the accuracy is the highest when the time domain division interval is 1 cycle, which is 99.59%. When the signal-to-noise ratio is 20dB, the classification accuracy rate is better than that of 1/2 periods and 1/4 periods by taking 1 period as a time domain segmentation scale method, and the noise robustness is better. Therefore, factors such as classification accuracy, feature extraction time, noise robustness and the like are comprehensively considered, and the optimal time domain division interval is determined to be 1 cycle. It should be noted that the statistical experiment in this section is performed based on the disturbance data with the sampling rate of 6400Hz, and when signals with other sampling rates are adopted, the statistical experiment can be constructed according to the same idea to determine the targeted optimal time domain division unit.
The transmission of massive high-sampling-rate power quality data increases the communication pressure of a network, and the complex feature extraction executed in an upper system increases the communication and calculation cost and the system response time, so that the method is difficult to be effectively applied to a power internet of things monitoring system. The invention provides an efficient electric energy quality edge feature extraction and disturbance identification method for a lightweight gradient elevator, which takes account of tie bandwidth constraints. The main work comprises the following steps:
(1) after the time domain segmentation is carried out on the original signal at the edge side, the time domain characteristics are directly extracted from the original signal. The method realizes the characteristic extraction of the power quality disturbance signal with low time complexity and space complexity, and can meet the low calculation requirement of low-cost edge data acquisition equipment.
(2) The optimal feature subset replaces the original signal to be uploaded, and the calculation amount of the edge equipment and the complexity of a classifier are further reduced; meanwhile, the communication bandwidth constraint of a typical Internet of things communication mode can be met.
(3) The data are preprocessed by adopting a unilateral gradient sampling and mutual exclusion sparse feature binding method, and a Leaf-wise growth strategy optimization classifier with depth limitation is adopted for construction, so that the classification efficiency and the generalization capability of an upper system are effectively improved.
Experiments prove that the method still has a good complex disturbance classification effect and good applicability on the basis of meeting the typical tie bandwidth constraint, and can effectively promote the application of the electric energy quality disturbance identification technology in the electric power tie scene.
The computing conditions, diagrams, etc. in the embodiments of the present invention are used for further description, are not exhaustive, and do not limit the scope of the claims, and those skilled in the art can conceive of other substantially equivalent alternatives without inventive step in light of the teachings of the embodiments of the present invention, which are within the scope of the present invention.
Claims (3)
1. A method for recognizing the power quality disturbance of a lightweight gradient elevator considering the bandwidth constraint of an Internet of things is characterized by comprising the following steps of:
1) construction of electric energy quality analysis system architecture based on typical Internet of things communication mode
The electric energy quality acquisition device is usually installed at an electric vehicle charging station, a photovoltaic power station, a wind power plant, a transformer substation, an electrified railway, an industrial load, a residential load side and a massive distributed power grid-connected point position, acquires relevant characteristics by using low-cost electric energy quality edge perception and characteristic extraction equipment, and realizes communication on the basis of analyzing an electric energy quality disturbance identification framework based on internet of things communication in a LoRa and NB-IoT typical internet of things communication mode so as to meet the bandwidth constraint of a narrow-band internet of things;
2) power quality time domain feature efficient edge extraction considering Internet of things bandwidth constraints
The massive power quality data of the power distribution network are difficult to transmit through a limited bandwidth Internet of things channel, so that the characteristic is used for replacing an original signal to upload, in the characteristic extraction process, after time domain segmentation can be carried out on the original signal, time domain characteristics are directly extracted, and the rule is as follows: selecting 20 time domain characteristics: harmonizing an average number, a variance, a standard deviation, an average difference, skewness, kurtosis, entropy, Shannon entropy, Renyi entropy, Tsallis entropy, a mean value, a root mean square, a maximum value, a minimum value, a difference between the maximum value and the minimum value, a sum of the maximum value and the minimum value, energy, a normalized amplitude factor, 1/4 period energy falling amplitude and 1/4 period energy rising amplitude, and respectively calculating various time domain characteristic values for each divided interval; then, respectively selecting maximum and minimum parameters in the same class of characteristic values in each time domain interval, wherein 40-dimensional characteristics are used for constructing an original characteristic set, and in addition, 20-dimensional characteristics of various time domains are integrally calculated for an original signal, and finally, 60-dimensional time domain characteristics are extracted; the experiment is completed on a microcomputer configured as an Intel Core i5-7500CPU and a DDR 42666 Hz12GB memory, 17 classes of PQ disturbance signals are generated in a simulation mode, and the PQ disturbance signals comprise a voltage standard signal (C0), a voltage sag (C1), a voltage temporary rise (C2), a voltage interruption (C3), a flicker (C4), a transient oscillation (C5), a harmonic (C6), a voltage shear mark (C7), a voltage spike (C8), a harmonic-contained temporary fall (C9), a harmonic-contained temporary rise (C10), an oscillation-contained temporary fall (C11), an oscillation-contained temporary rise (C12), a flicker-contained temporary fall (C13), a flicker-contained temporary rise (C14), a flicker-contained harmonic (C15) and a break-contained harmonic (C16);
3) construction of optimal feature set and optimal classifier
Establishing a lightweight Gradient Boosting Machine (LightGBM) classifier in Spyder software under an Anaconda development environment, and performing LightGBM parameter optimization by using 10-fold cross validation and Bayesian optimization to obtain a primary LightGBM classifier by aiming at the minimum classification error rate obtained in the classification process of the LightGBM classifier; obtaining importance values of all time domain features through a gradient lifting algorithm of a primary LightGBM classifier, sequencing the importance values from high to low, sequentially adding the features into a feature subset according to an order of descending the importance, calculating the identification accuracy of the classifier after parameter optimization under the feature subset, repeating the process until all the features are added into the feature set, finally determining an optimal feature subset according to the highest identification accuracy, and inputting 600 groups of electric energy quality signals into the primary LightGBM classifier for further training after performing feature calculation on the features belonging to the optimal feature vector set through a time domain feature formula in the step 2) from 17 types of electric energy quality disturbance signals to obtain the trained LightGBM disturbance identification classifier;
4) electric energy quality disturbance efficient recognition effect analysis
And (3) passing the collected power quality disturbance signals through the time domain feature formula in the step 2), extracting the time domain features of the optimal feature vector set obtained in the step 3), and inputting the extracted features into the LightGBM disturbance recognition classifier trained in the step 3) to obtain the reflected power quality problem.
2. The method for identifying the power quality disturbance of the lightweight gradient elevator considering the bandwidth constraint of the internet of things according to claim 1, which is characterized in that: the high bandwidth of the LoRa and NB-IoT typical Internet of things communication mode in the step 1) is 100 Kbps.
3. The method for identifying the power quality disturbance of the lightweight gradient elevator considering the bandwidth constraint of the internet of things according to claim 1, which is characterized in that: the fundamental frequency of the voltage standard signal in the step 2) is 50Hz, and the sampling rate is 6400 Hz.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010204617.0A CN111325485B (en) | 2020-03-22 | 2020-03-22 | Light-weight gradient elevator power quality disturbance identification method considering internet-of-things bandwidth constraint |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010204617.0A CN111325485B (en) | 2020-03-22 | 2020-03-22 | Light-weight gradient elevator power quality disturbance identification method considering internet-of-things bandwidth constraint |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111325485A CN111325485A (en) | 2020-06-23 |
| CN111325485B true CN111325485B (en) | 2022-03-18 |
Family
ID=71164035
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010204617.0A Active CN111325485B (en) | 2020-03-22 | 2020-03-22 | Light-weight gradient elevator power quality disturbance identification method considering internet-of-things bandwidth constraint |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111325485B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111970374B (en) * | 2020-08-27 | 2023-02-03 | 山东有人物联网股份有限公司 | Data node grouping method, system and medium based on machine learning |
| CN112100497B (en) * | 2020-09-14 | 2021-10-19 | 北京嘀嘀无限科技发展有限公司 | Data processing method and device, electronic equipment and readable storage medium |
| CN113238110B (en) * | 2021-05-10 | 2022-10-14 | 合肥工业大学 | Power quality disturbance diagnosis method |
| CN116541686B (en) * | 2022-11-01 | 2024-03-15 | 河海大学 | Electric energy quality disturbance classification method based on multi-domain feature fusion extreme learning machine |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104459397B (en) * | 2014-12-08 | 2017-05-17 | 东北电力大学 | Power quality disturbance recognizing method with self-adaptation multi-resolution generalized S conversion adopted |
| CN107832777B (en) * | 2017-10-12 | 2021-01-26 | 吉林化工学院 | Electric energy quality disturbance identification method adopting time domain compression multi-resolution rapid S-transform feature extraction |
| CN110428346A (en) * | 2019-09-02 | 2019-11-08 | 云南电网有限责任公司电力科学研究院 | A kind of integrated energy system containing hydrogen energy storage |
| CN110570114A (en) * | 2019-09-03 | 2019-12-13 | 湖南大学 | Power quality disturbance identification method, system and medium based on variable cycle neural network |
| CN110705762B (en) * | 2019-09-20 | 2023-04-18 | 天津大学 | Ubiquitous power Internet of things perception data missing repairing method based on matrix filling |
| CN110718910B (en) * | 2019-10-29 | 2020-09-15 | 国网四川省电力公司经济技术研究院 | Transient stability evaluation method for Bayesian optimization LightGBM |
-
2020
- 2020-03-22 CN CN202010204617.0A patent/CN111325485B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111325485A (en) | 2020-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111325485B (en) | Light-weight gradient elevator power quality disturbance identification method considering internet-of-things bandwidth constraint | |
| CN105354636B (en) | A kind of wind power swing probability density modeling method based on nonparametric probability | |
| CN109685314A (en) | A kind of non-intruding load decomposition method and system based on shot and long term memory network | |
| CN107038167A (en) | Big data excavating analysis system and its analysis method based on model evaluation | |
| CN104459397B (en) | Power quality disturbance recognizing method with self-adaptation multi-resolution generalized S conversion adopted | |
| CN104408562A (en) | Photovoltaic system generating efficiency comprehensive evaluation method based on BP (back propagation) neural network | |
| CN103995237A (en) | Satellite power supply system online fault diagnosis method | |
| CN110119545B (en) | Non-invasive power load identification method based on stack type self-encoder | |
| CN113344346B (en) | Power utilization abnormity detection method and system based on non-intrusive load decomposition | |
| CN113132399A (en) | Industrial control system intrusion detection method based on time convolution network and transfer learning | |
| CN111327118B (en) | Non-invasive power load identification method | |
| CN115099296A (en) | Sea wave height prediction method based on deep learning algorithm | |
| CN114169445A (en) | Day-ahead photovoltaic power prediction method, device and system based on CAE and GAN hybrid network | |
| CN110399796A (en) | A kind of electrical energy power quality disturbance recognition methods based on improvement deep learning algorithm | |
| CN115238759A (en) | Electric power data compression and electric energy quality disturbance identification method based on compressed sensing | |
| CN114298080A (en) | Hydro-turbo generator set monitoring method based on throw data mining | |
| CN117349657A (en) | Distributed data acquisition module and monitoring system for hydraulic engineering environment monitoring | |
| CN116561569A (en) | Industrial power load identification method based on EO feature selection and AdaBoost algorithm | |
| CN115828091A (en) | Non-invasive load identification method and system based on end-cloud cooperation | |
| CN114372640A (en) | Wind power prediction method based on fluctuation sequence classification correction | |
| CN114065651A (en) | Fault time prediction method for rotary equipment | |
| CN112949938A (en) | Wind power climbing event direct forecasting method for improving training sample class imbalance | |
| CN105701591A (en) | Power grid service classification method based on neural network | |
| CN116633021B (en) | Switch cabinet energy consumption monitoring method and system based on Internet of things | |
| CN109345115B (en) | Evaluation method for evolution driving metric value of power information physical system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |