Iot-based power detection equipment management and control system

4.1 SVM algorithm based on Internet of Things

A SVM is a binary classification model that maps the feature vector of an instance to some points in space. The purpose of the SVM is to draw a line to “best” distinguish the two types of points. SVM is one of the most widely used classifiers in the field of computer vision. It looks for the segmentation hyperplane located in the “right middle” of the two types of training samples to make the positive and negative samples have a considerable separation distance from the hyperplane. The segmented hyperplane obtained in this way has the best tolerance to local noise [16].

For linear SVM classifiers, the steps in the prediction phase mainly involve the calculation of the inner product, as shown in formula equation (1):

An algorithm that uses the weighted sum of multiple binarized vectors f(a) to approximate a vector. After such binarized approximation, the inner product ≺ w , a ≻ of the vector can be quickly calculated using simple operations. The binary image occupies a very important position. The binarization of the image greatly reduces the amount of data in the image, so that the outline of the target can be highlighted.

After the binarization approximation algorithm of w in formula (1), an approximate expression can be obtained, as in formula (2):

After obtaining the binarized vector, it can be expressed as the sum of a binary vector and its complement a _j , as in formula (3):

In the target detection task, two tasks need to be completed, one is to judge whether the candidate frame is the target or the background, and the other is to return the accurate position of the target according to the candidate frame and features [17]. The loss function is used to evaluate the degree to which the predicted value of the model is different from the true value. The better the loss function, the better the performance of the model. When these two tasks are completed at the same time, the loss functions of the two tasks can be combined. Such as formula (4):

Among them, i represents the subscript of the candidate frame in the batch gradient descent process, P _i represents the probability of the ith candidate frame being identified as the target, and P i ∗ is the category label.

L _cls represents the classification loss function, which is a log loss function for two categories of target and non-target. Such as formula (5):

The classification loss function used to classify the target and background and the classification loss function used to regress the target position are both fully connected layers. The number of parameters of the fully connected layer is very large [18], as shown in Figure 5:

Figure 5

SVD decomposition diagram of the fully connected layer.

As shown in Figure 5, SVD decomposition will be carried out, and the first t largest eigenvalues will be taken to approximate the original w matrix. Such as formula (6):

At this time, the forward propagation of the network becomes UΣV ^T , then the computational complexity is changed from u × v to u × t + t × v . If t is much smaller than min(u, v), then t(u + v) is much smaller than u × v [19]. This can significantly accelerate the forward propagation speed of the fully connected layer [20].

Activation function f(a) is a very important part of the deep learning system. It adds non-linear changes to the whole system, which makes the deep learning model 1 1 + e − a have more powerful expressive ability [21]. Such as formula (7):

4.2 Clustering analysis algorithm based on the Internet of Things

The degree of informatization in more and more fields has continued to increase. How to effectively mine valuable information from these data, extract common characteristics, and provide an accurate theoretical basis for power system analysis, decision-making, and control. In this way, ensuring the safe, reliable, and economical operation of the power grid has become a key issue in today’s power monitoring.

Cluster analysis aims to divide data sets, patterns, or objects into meaningful or useful clusters. Considering that the short-distance measurement is Euclidean distance data, this article uses the sum of squared errors as the objective function to measure the quality of clustering. It defines the error of each data point as the Euclidean distance to the nearest centroid and calculates it. Euclidean distance generally refers to the Euclidean metric. In mathematics, the Euclidean distance or Euclidean metric is the “ordinary” (i.e., straight line) distance between two points in Euclidean space.

The smaller the error sum of squares, the more ideal the clustering result. Such as formula (8):

c _i is the ith cluster, and c _i is the centroid of the cluster. The centroid of the ith cluster is as in formula (9):

where m _i is the number of objects in the ith cluster, m is the number of objects in the data set, and a is the number of clusters. The objective function is the desired form of the objective expressed by the design variables, so the objective function is the function of the design variables, which is a scalar.

The sum of squared distances between each data point a _i and its nearest cluster center c _r is minimized, the centroid of each cluster recalculated, and the deviation calculated as formula (10):

In the clustering process, an objective function is usually selected as the evaluation function of the clustering effect, and the value of this function determines the similarity between objects in the data set.

Fuzzy clustering analysis methods can be roughly divided into two types: one is the fuzzy clustering method based on fuzzy relationships. Also known as systematic cluster analysis. The other is called non-systematic clustering method. It first roughly divides the samples and then classifies them according to their optimal principles. After many iterations, until the classification is reasonable, this method is also called the step-by-step clustering method. Fuzzy clustering algorithms are widely used in the field of image segmentation, mainly because the membership of image pixels requires the use of fuzzy theory. Such as formula (11):

Among them, δ is the fuzzification index, and u m n δ is the membership degree of the nth sample to the mth class. The determination of the number of clusters needs to traverse the possible range of values.

The basic idea of the hierarchical clustering method is to calculate the similarity between nodes by a certain similarity measure, sort them from high to low by similarity, and gradually reconnect each node. The advantage of this method is that the division can be stopped at any time, The aggregation rules in the hierarchical clustering algorithm are divided into: single connection aggregation rules-that is, the distance between classes is equal to the minimum distance between two types of objects; fully connected aggregation rule-that is, the distance between classes is equal to the maximum distance between two types of objects; average inter-class connection aggregation rule-that is, the inter-class distance is the average distance between two types of objects. Arbitrarily select five sets of two-dimensional coordinates as the data set, as shown in Table 2.

As shown in Table 2, according to the proximity matrix, the distance between P3 and P5 is the smallest. To calculate the distance between two combined data points and one combined data point, the three aggregation rules mentioned earlier can be used: full connection aggregation rules, single connection aggregation rules, and this article uses single connection aggregation rules. In single-join aggregation clustering, the distance between two clusters is defined as the shortest distance between two objects in each cluster, and this calculation rule makes clustering easy to implement.

Grid is an effective method, especially to organize data sets in low-dimensional space. The purpose of the grid-based clustering algorithm is to divide the possible values of each attribute into multiple adjacent intervals and create a group of grid cells.

As shown in Figure 6, in the grid-based clustering algorithm, there are many ways to divide the possible values of each attribute into multiple adjacent intervals, so there are many ways to define the grid unit. Clustering is a widely used exploratory data analysis technique. People’s first intuition about data is often by meaningfully grouping data, and by grouping objects, similar objects are grouped into one group.

Figure 6

Grid-based clustering algorithm.

Model-based clustering algorithms are mainly methods based on probability models and methods based on neural network models, especially methods based on probability models.

The Gaussian distribution model is as formula (12):

Gaussian mixture model is to use Gaussian probability density function (normal distribution curve) to accurately quantify things. It is a model that decomposes things into several models based on the Gaussian probability density function (normal distribution curve). Where a is the column vector with dimension d, u is the model expectation, and Σ is the model variance.

The Gaussian mixture model is shown in equation (13):

In the model, K is the number of clusters, which needs to be determined in advance, and the number of clusters is determined to be consistent in K-Means. π _k is the weight factor.

Assuming that the sample size is N, the number of samples belonging to K categories is N ₁, N ₂, …, N _K respectively, and the sample set belonging to the Kth category is L(k). The calculation expression is as formula (14):

A likelihood function is a function of parameters in a statistical model that represents the likelihood in the parameters of the model. Likelihood functions play an important role in statistical inference, such as applications between maximum likelihood estimation and Fisher information. In the case of unclassified samples, assuming that there are N data points, find a set of parameters θ; usually, this expression is called the likelihood function. The calculation formula is as formula (15):

Since the probability of a single data point is generally smaller, the result of the likelihood function will be smaller. For the convenience of calculation, the logarithm is generally calculated as formula (16):

In order to maximize the value of the result of formula (16), the EM method can be used to solve it, so this method is called the expectation-maximization algorithm, where b is an implicit variable. The EM solution expression is equation (17):

Initialize the distribution parameter Pr(A = a _j , B = b,θ) so that the likelihood function of the parameter is equal to its lower bound.

Cluster effectiveness can be divided into internal effectiveness indicators, external effectiveness indicators, and relative effectiveness indicators. For internal indicators, they are usually divided into three types: indicators based on the fuzzy division of data sets, indicators based on the geometric structure of the data set samples, and indicators based on the statistical information of the data set.

The index based on the fuzzy division of the data set describes the closeness between the clusters through the within-class dispersion matrix. The index is defined as formula (18):

trB(h) is the deviation matrix between classes, and trW(h) is the trace of the deviation matrix within the class.

The index definition based on the geometric structure of the data set sample is as shown in formula (19):

w _i represents the distance from all samples in class c _ij to their cluster centers. j represents the average distance of all sample points in class c _ij to the center of class c _ij , and c _ij is the average distance of class. It can be seen that the smaller the index, the better the clustering result.

Under the action of wind pressure and hot pressure, air tightness is an important control index to ensure the stability of thermal insulation performance of building exterior windows, the smaller the heat loss. The tightness is measured by the average distance between samples within a class, and the minimum distance between samples between classes is used to measure the separation. The indicator is defined as formula (20):

where a(n) is the average distance from sample n to other samples in the class, and b(n) is the minimum value of the average distance from sample n to samples in other classes.

5.1 Experiment and analysis of traditional power detection improved system

Traditional detection systems are generally wired. Today’s intelligent detection systems mostly use radio frequency technology, which means that they can be plug and play. This makes up for the defects of the traditional intelligent management system that are complex, inconvenient to use, and incompletely intelligent. It has the characteristics of real-time status feedback, real-time remote control, or interactive intelligent control.

In this article, four electric power experts scored the deficiencies and defects of the traditional detection system in various aspects, as shown in Figure 7.

Figure 7

The shortcomings and shortcomings of traditional inspection systems in various aspects.

As shown in Figure 7, the special defects of the traditional inspection system include complex process, inconvenient use, and incomplete intelligence. The adverse effects caused by this are high cost, low comfort, weak safety, and so on. This article selects the load curve monitoring points of 5 electricity consumption units in the monitoring system as the test data set to test the actual clustering effect of the K-Means clustering algorithm when processing the data set. The scatter diagram of sample points is shown in Figure 8.

Figure 8

The effect comparison chart before and after using the clustering algorithm.

As shown in Figure 8, it can be seen from the figure that as the data becomes larger, the distance gradually becomes smaller, and the slope of the curve also gradually decreases. It can be seen that with the increase in the number of clusters, the decreasing trend of the data set distance tends to be slow. As the number of clusters increases, the scale of each data center will decrease. At the same time, each cluster center is relatively more compact, the distance will be reduced, and the clustering effect will be better. This article compares the traditional power detection system algorithm and the power detection system algorithm based on the clustering algorithm.

It can be seen from Tables 3 and 4 that the algorithm detection rate of the power detection system based on the clustering algorithm is between 77 and 86%. The lowest detection rate is 77%, and the highest detection rate is 86%; The false detection rate of the power detection system algorithm based on the clustering algorithm is between 10 and 13%. The lowest false detection rate is 10%, and the highest false detection rate is 13%. It can be seen that the power detection system algorithm based on the clustering algorithm makes the error of the detection power equipment system smaller, and this method is effective to detect the power system.

5.2 Experiment and analysis of the improved power testing improved system

The main equipment involved in this article includes sensors, thermometers, access control, cameras, energy controllers, and fire alarm controllers.

5.2.2 Main plan

1. Laboratory environmental monitoring: The temperature and humidity sensors and smoke sensors are installed in the laboratory to monitor the environmental information in the laboratory at all times and transmit the environmental data to the PC software in real time and display the current temperature and humidity on the display screen in the laboratory to achieve environmental maintenance. The temperature and humidity of the laboratory are kept within a reasonable range through remote control of the start and stop of the air conditioner, humidifier, and other equipment.

2. Monitoring of platform testing equipment: the appearance of the platform is unified, and the interface of the platform with the conditions is modified. In order to collect various data of the platform and monitor the operating status of the platform, the following modifications are made:

① The terminal temperature detection device is installed. The terminal temperature detection device is a high-precision and high-performance device designed based on the latest digital technology. The hardware is mainly composed of high-precision signal source, main control, 0.05-level digital standard electric energy meter, SPWM power source, and meter unit. The software is a PC control management software with a configurable test plan, which can be adjusted according to the different functional requirements of customers to provide a complete test plan for the verification of the electric energy meter. The device has the characteristics of high detection efficiency, high accuracy, wide measuring range, stability and reliability, and high-cost performance. The platform model can be built on the PC software, and the data at each terminal of each epitope can be displayed in real time.

② The smart circuit breaker is installed, and the smart circuit breaker is installed on the table body to realize protection functions such as overcurrent and leakage. It also continuously monitors the working status and duration of the platform, and has the function of remote control start and stop. It also adds additional protective functions to the table body to ensure the safety of equipment and personnel.

③ The operation monitoring unit is installed to monitor the voltage and current data of different meter positions in real time and compare with the data set in advance to prevent operation failure. Through the platform model on the PC-side software, various data are displayed intuitively. When the data exceeds the predetermined value, the color changes and an alarm is issued on the platform.

(3) Inspection personnel: increase the laboratory safety identification module, install the access card in the laboratory, and install the “operation monitoring unit” on the platform. After swiping the card, the table body and the corresponding detection function of the table body are used. The station body identity verification module, the operating mode of the console body, is increased, and each staff member holds a card. Before the test work, send the test plan process and so on to the card through the PC. The staff needs to swipe the card on the platform for authentication, and the running content of the platform will not exceed the set plan. If you need to change it, you need to re-issue and re-authenticate the card.

As shown in Figure 9, the second part is the monitoring of the various operating data of the platform, which is collected by the collection unit and uploaded to the PC.

Figure 9

Modification diagram of platform testing equipment.

(4) Monitoring of samples submitted for inspection: The samples submitted for inspection are mainly integrated into unified management through warehousing coding, file creation, intelligent storage, automatic circulation, detection matching, cycle control, and early warning.

Traditional detection systems are designed to detect and query the operating data of power equipment, and lack the effective analysis and mining of historical operating data, and cluster analysis-based load classification research, medium, and long-term load forecasting, and power consumption pattern analysis have emerged. According to the development direction and plan of the smart grid.

(5) Data acquisition and transmission plan

1. Taiwan body identity verification module: For routine testing of electric energy meters, the testing process and plan can be determined, and the control unit can strictly control the running time and current of the platform. If the plan exceeds the predetermined plan, an alarm will be issued, and the test can be re-tested after re-certification.

2. HPLC/wireless converter: monitor the temperature, voltage and current in real time through MUS/485, and store the data in the collection unit. The collection unit transmits the collected data to the energy controller through the HPLC dual-mode in real time, and the energy controller uploads the data to the master station through the 5 G module or Ethernet.

3. Main station: The main station side establishes the space model diagram of the platform and displays various data at the corresponding position on the way.

As shown in Figure 10: Through the experiments, it can be seen that the improved detection system of this project has not only improved safety but also greatly improved its accuracy.

Figure 10

Data transfer diagram.

In view of the goal of the development of smart grids, smart grid equipment management requires a fast interactive real-time network platform. It uses the online monitoring system of the equipment to obtain the real-time status of the equipment and sends out a warranty signal in time when the equipment fails, so that the maintenance personnel can perform maintenance in time, so as to ensure that the equipment is running in the best state at all times. The equipment management of the smart grid is a very important part of the normal and efficient operation of the smart grid. Standardizing the itinerant overhaul, daily maintenance and maintenance of the equipment in the system can greatly improve work efficiency.

The specific advantages of the improved power detection equipment are shown in Figure 11.

Figure 11

Comparison diagram of two power detection improved systems: (a) traditional power detection equipment improved system; (b) improved power detection equipment improved system.

As shown in Figure 11, this article mainly compares the safety, accuracy, and economy of the power detection equipment improved system and the traditional power detection improved system of the Internet of Things.

It can be seen from Figure 11 that the safety, accuracy, and economy of the traditional power detection improved system of the Internet of Things are not as good as the power detection equipment improved system and the traditional power detection improved system of the Internet of Things. Therefore, it can be seen that the power testing equipment improved system is more conducive to the testing development of today’s electrical industry.