Outdoor metering box remote operation control system based on Internet of Things

Abstract

The invention relates to the technical field of power monitoring, in particular to an outdoor metering box remote operation control system based on the Internet of Things, which comprises a data acquisition module used for acquiring equipment parameters of a metering box and environmental parameters of a metering box installation position, and combining the equipment parameters and the environmental parameters to construct a basic data set X (t); and the data processing module is used for performing data cleaning, noise reduction processing and format standardization processing on the basic data set, so that the basic data set is converted into a standard data set S (t). According to the method, quantitative calculation is carried out on external interference with dynamics and randomness, historical data are combined, the abnormal detection threshold value of parameters is reasonably set, and abnormal points in data collected by the metering box are effectively recognized; and the abnormal source is corrected in a targeted manner to ensure that the data received by the user side is closer to the real value.

Description

Technical Field

[0001] The present invention relates to the technical field of power monitoring, and particularly relates to a remote operation control system for outdoor metering boxes based on the Internet of Things. Background Art

[0002] With the rapid development of urbanization and smart grids, electricity, as an important form of energy, its metering and management have become important links in the operation and maintenance of power systems. During the process of power transmission and distribution, power metering boxes are key devices for measuring, recording, and controlling power consumption, and are widely used in industrial, commercial, and residential electricity consumption scenarios. Traditional outdoor power metering boxes usually have basic power metering functions, but due to the lack of support for intelligence and remote control, there are many limitations in their operation management and maintenance.

[0003] Especially in the long-term unmanned outdoor environment, affected by bad weather, the performance of the equipment may decline, which may further cause the electronic components in the metering box to drift, age, or even fail; in addition, when the metering box in the wild environment is in use, the activities of animals and plants will also cause chain problems for the detection of the metering box; including animals building nests at the metering box, resulting in blocked ventilation ports, reduced heat dissipation performance, and even overheating of the equipment, or the roots of plants may extend into the metering box through underground cable ducts, causing damage to the cable sheath, resulting in exposed conductors contacting or short-circuiting with external metal components; due to the above problems, when the metering box detects data and conducts abnormal analysis, traditional fixed thresholds are difficult to adapt to the changing fluctuations of parameters, resulting in frequent false alarms or missed alarms of parameter abnormalities; and after being uploaded to the user side, the data received by the user side has serious deviations.

[0004] Therefore, a remote operation control system for outdoor metering boxes based on the Internet of Things is proposed to solve the above-mentioned problems. Summary of the Invention

[0005] Technical Problems to be Solved

[0006] In view of the above-mentioned disadvantages existing in the prior art, the present invention provides a remote operation control system for outdoor metering boxes based on the Internet of Things, which can effectively solve the problem that the measurement results of metering boxes in the wild environment are deviated due to the influence of bad weather and plant activities in the prior art.

[0007] Technical Solutions

[0008] To achieve the above object, the present invention is realized through the following technical solutions:

[0009] The present invention provides a remote operation control system for an outdoor metering box based on the Internet of Things. The technical solution adopted by the present invention is as follows: It includes a data acquisition module, which is used to collect the device parameters of the metering box and the environmental parameters at the installation location of the metering box, and combine the device parameters and environmental parameters to construct a basic data set X(t); it also includes:

[0010] A data processing module, which is used to preprocess the basic data set and convert it into a standard data set S(t);

[0011] A data analysis module, which is used to screen abnormal data from the standard data set S(t) and combine them to construct an abnormal data set A(t); it is also used to extract a comprehensive interference index and a health assessment index from the standard data set S(t); construct a device health assessment model based on the standard data set S(t) to obtain the health assessment index of electrical components; construct a dynamic abnormal analysis model based on the abnormal data set A(t) to obtain the comprehensive interference index;

[0012] An abnormal diagnosis module, which calculates a dynamic anti-interference threshold based on the extracted comprehensive interference index and health assessment index, and then extracts the confirmed data in the abnormal data set A(t) based on the dynamic anti-interference threshold, and constructs a confirmed data set D(t);

[0013] A data repair module, which is used to correct the confirmed data, obtain the repaired parameter data and transmit it to the user side.

[0014] Among them, the environmental parameters include temperature, humidity, wind speed, box tilt angle, rainfall, lightning activity, and whether there is biological intrusion; the device parameters include current, voltage, as well as the maintenance records and cumulative operation time of the corresponding electrical components.

[0015] Among them, the method for screening abnormal data is:

[0016] Extract the parameter data in the standard data set S(t), compare the parameter data with the preset static threshold range, and record the parameter data within the static threshold range as normal data; compare the parameter data outside the static threshold range with the preset dynamic threshold range, and record the parameter data outside the dynamic threshold range as risk data;

[0017] Construct a collaborative analysis model based on the risk data to obtain abnormal data and the source of abnormality; combine the sources of abnormality to construct an abnormal source collection AEC(t).

[0018] Among them, the setting method of the dynamic threshold range is:

[0019] Collect the parameter data within a fixed past time, and construct it into a historical data set {x 1 ,x 2 ,...,xN}; where x is the name index of the parameter data, and N is the number of samples of the parameter data;

[0020] Calculate the mean value of the parameter data. The calculation formula is: In the formula, represents the mean value of the parameter data; x i is the i-th parameter data;

[0021] Calculate the standard deviation of the parameter data. The calculation formula is: In the formula, σ x represents the standard deviation of the parameter data;

[0022] Set the dynamic threshold range of the abnormal data based on the mean value and standard deviation of the parameter data Among them, is the lower limit of the dynamic threshold, is the upper limit of the dynamic threshold; ω is the threshold coefficient.

[0023] Among them, the method for constructing the collaborative analysis model is:

[0024] Extract the parameter data with an association relationship between the basic data sets X(t), and construct it into the current state vector XV = [D x1 , D x2 ,..., D xM ; where D x1 , D x2 ,..., D xM represent different parameter data with an association relationship, and M is the number of acquisitions of the parameter data with an association relationship;

[0025] Then extract the parameter data within a fixed past time, and construct it into the historical state vector

[0026] Calculate the deviation degree DEV between the current state vector and the historical state vector. The calculation formula is: In the formula, is the Euclidean distance between the current state vector and the historical state vector; D xi is the current value of the i-th parameter data, is the historical value of the i-th parameter data;

[0027] Compare the deviation degree DEV with the preset deviation threshold, and mark the current state vector with a deviation degree DEV greater than the deviation threshold as an abnormal state;

[0028] Extract all the parameter data in the abnormal state, and calculate the deviation amount ΔD of each parameter data xi , and the calculation formula is:

[0029] Calculate the total deviation value again

[0030] For each parameter data, calculate its contribution ratio in the total deviation value. The calculation formula is: Calculate the contribution ratio threshold TH of each data parameter again xi , and the calculation formula is: In the formula, is the historical average contribution ratio of the i-th parameter data; μ xi is the historical contribution ratio standard deviation of the i-th parameter data; τ is the discrete range factor;

[0031] Compare the contribution ratio Cont of all parameter data in the abnormal state xi with the corresponding contribution ratio threshold TH xi , and mark the parameter data with the contribution ratio Cont xi greater than the contribution ratio threshold TH xi as abnormal data

[0032] Among them, the determination method of the abnormal source is:

[0033] Extract the contribution ratio Cont of all abnormal data xi , and calculate the main abnormal threshold and the associated abnormal threshold. The calculation formula is:

[0034]

[0035]

[0036] In the formula, TH main is the main abnormal threshold, and the abnormal data with the contribution ratio greater than the main abnormal threshold is the main abnormal source; TH rel is the associated abnormal threshold, and the abnormal data with the contribution ratio greater than the associated abnormal threshold and less than the main abnormal threshold is the secondary abnormal source; is the average contribution ratio of all abnormal data; is the contribution ratio of the abnormal data arranged in descending order; E is the total number of abnormal data; u is the dynamic adjustment coefficient; v is the fixed offset part

[0037] Among them, the method of constructing the equipment health assessment model is:

[0038] Extract interference parameters from the standard data set S(t), including the temperature change range ΔT env , the humidity change range ΔH env , the wind speed fluctuation max(W speed ), the cumulative rainfall ΔP rain , the lightning strike frequency F lgt , the plant and animal activity frequency Fap , the area A of plant and animal occlusion ap and the offset angle θ of the metering box tilt ; Combine all interference parameters to construct an interference set Z;

[0039] Calculate the comprehensive interference index based on the interference set. The calculation formula is:

[0040]

[0041] In the formula, CII is the comprehensive interference index; z is the name index of the interference parameter; S z is the standardized value of the interference parameter; λ is the non-linear adjustment index; is the non-linear adjustment factor; γ z is the importance coefficient of the interference parameter, and the value range is: 0 < γ z ≤1; η z is the normalized weight of the interference parameter, satisfying Σ z∈Z η z = 1;

[0042] The method for constructing the dynamic anomaly analysis model is as follows:

[0043] Extract the health scoring parameters from the standard data set S(t), including the cumulative operation duration historical maintenance times

[0044] Based on the cumulative operation duration Calculate the aging score of the electrical component. The calculation formula is: In the formula, is the aging score of the electrical component for collecting the corresponding parameter data; is the characteristic life for collecting the corresponding parameter data; b is the shape parameter;

[0045] Based on the historical maintenance times N maint Calculate the maintenance score of the electrical component. The calculation formula is: In the formula, is the maintenance score of the electrical component for collecting the corresponding parameter data; is the maximum maintenance times of the electrical component; is the time interval since the last maintenance of the electrical component; is the allowable longest maintenance time interval of the electrical component; α 1 is the weight coefficient of the maintenance times, α 2 is the weight coefficient of the maintenance time interval;

[0046] The formula for calculating the equipment health assessment index is:

[0047] In the formula, HAIx is the health assessment index of the electrical component for collecting corresponding parameter data; β 1 is the weight coefficient of the aging score, β 2 is the weight coefficient of the maintenance score;

[0048] The formula for calculating the dynamic anti-interference threshold is:

[0049] In the formula, (δ 1 ·CII) is the interference correction term, δ 1 is the weight coefficient of the comprehensive interference index; (1 + δ 2 (1 - HAI x )) is the health correction term, δ 2 is the weight coefficient of the health assessment index.

[0050] Among them, the fitting and correction method of the shape parameter b is:

[0051] Under uncertain conditions, predict the shape parameter b through the prior distribution and observed data. The prediction formula is: P(b) ∝ P((b prior )P(S(t)|b prior )); In the formula, P(b) is the prediction distribution; P(b prior ) is the prior distribution; b prior is the assumed shape parameter; P(St)|b prior ) is the likelihood function;

[0052] Combining the likelihood value and prior distribution obtained by prediction, calculate the posterior distribution of the shape parameter b. The calculation formula is: P(b post ) ∝ P(S(t)|b)·P(b); In the formula, P(b post ) is the posterior distribution;

[0053] Use the updated posterior distribution P(b post ) as the prior distribution for the next round, and repeat the above steps; Among them, the calculation expression of the prior distribution is:

[0054]

[0055] The calculation expression of the likelihood function is:

[0056]

[0057] In the formula, μ b is the prior mean; is the prior variance; exp(...) is the exponential function; is the product symbol, J is the total number of data points in the standard data set S(t); S(t) jis the j-th data point in the standard data set S(t); σ ε represents the uncertainty of the observed data; H(t j | prior ) represents the health value corresponding to the j-th time point t prior predicted according to the assumed shape parameter b j .

[0058] Among them, the data analysis module generates extreme condition data in combination with the GAN model, and the generation method is as follows:

[0059] Collect the health index set of the standard data set S(t) at time t, and mark it as the real data set S real ; Define the upper threshold TH real and the lower threshold TH UB of the real data set S NT ;

[0060] Define the structure of the GAN model including a generator G and a discriminator D;

[0061] The generator G is used to receive a random noise vector Ψ ∼ N(0, 1) and a prior health state distribution P(S real ) as inputs, and output a generated data set S gen of extreme health states;

[0062] The discriminator D is used to receive the standard data set S(t) and the generated distribution from the real data set as inputs, and discriminate the probability that the standard data set S(t) comes from the real distribution;

[0063] Define the loss function of the GNN model:

[0064]

[0065] In the formula, S real ∼ P(S real ) represents the real data set sampled from the prior health state distribution; logD(S real ) represents the confidence of the real data set; Ψ ∼ P Ψ represents a random vector sampled from the random noise distribution P Ψ ; G(Ψ) is the generated sample generated by the generator simulating the real data distribution; D(G(Ψ)) is the output of the discriminator for the generated sample;

[0066] Add a boundary loss term L bnd to the loss function, and the calculation formula of the boundary loss term is: where hyp is a hyperparameter that controls the strength of the boundary constraint;

[0067] Fuse the generated data set and the real data set to construct an extended data set Sext and return the extended data set S ext to the Bayesian dynamic update and correction model.

[0068] Among them, the method for correcting the confirmed data is as follows:

[0069] Extract the confirmed data from the confirmed data set D(t), and then extract the abnormal source of the confirmed data from the abnormal source collection AEC(t); the correction formula is:

[0070]

[0071] In the formula, is the repaired parameter data; is the confirmed data; is the abnormal source correction term, where: k Q is the correction coefficient corresponding to the abnormal source, Q is the name index of the abnormal source, is the current value of the abnormal source, is the best value of the abnormal source, is the limit upper value of the abnormal source, is the limit upper value of the abnormal source; is the equipment health correction term, where: ΔHAI x is the equipment health correction value, and the calculation formula is: is the reference evaluation index in the healthy state, is the weight coefficient of the equipment health.

[0072] To sum up, due to the adoption of the above technical solutions, the beneficial effects of the present invention are as follows:

[0073] 1. In the present invention, by setting the dynamic threshold range, the personalized threshold of the historical data set of different parameter data can be satisfied, and the volatility of different parameter data can be accurately adapted, and the abnormal points in the data collected by the metering box can be effectively identified; by statistically analyzing the historical data of long-term operation, the normal data distribution range is dynamically quantified, so that the determination of abnormal data has solid data support, rather than relying on manual experience or a single rule, making the applicability of this solution to the determination of abnormal data stronger; and in the process of identifying abnormal data, not only can it identify whether the current state is abnormal, but also can accurately judge the main problem source, analyze the main abnormal data parameters and secondary abnormal parameters, reducing the ambiguity of abnormal analysis.

[0074] 2. In the present invention, by quantifying and calculating dynamic and random external interferences, abnormal data can be effectively extracted from the data detected by the metering box. Compared with the traditional method of identifying with static thresholds, abnormal data can be identified more accurately, avoiding misjudging normal fluctuations as abnormalities easily when the external environment changes. At the same time, the dynamic threshold adjustment mechanism can achieve personalized optimization by constructing an analysis model for electrical components for different detected data, making the overall system more compatible, responding to environmental changes in real time, adapting to various complex scenarios, and realizing unified dynamic abnormal detection.

[0075] 3. In the present invention, by specifically correcting the diagnosed data, removing the abnormal sources causing the abnormality of the parameter data and the deviations caused by problems of electrical component devices, the accuracy of the parameter data collected by the metering box can be effectively improved, ensuring that the data received by the user side is closer to the true value. And when correcting the diagnosed data, the specific reasons for the data abnormality are fully considered, providing a targeted basis for data repair and effectively improving the credibility of the repaired parameter data. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a schematic diagram of the control system structure of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0077] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0078] Embodiment

[0079] Referring to Figure 1 , this case proposes a remote operation control system for outdoor metering boxes based on the Internet of Things, including a data acquisition module, a data processing module, a data analysis module, an abnormal diagnosis module, and a data repair module; among them:

[0080] The data acquisition module includes a temperature and humidity sensor, an anemometer, a rain sensor, a lightning sensor, an infrared sensor, an inclination sensor, and a power sensor; the data acquisition module is used to collect environmental parameters and equipment parameters at the metering box; the environmental parameters include temperature, humidity, wind speed, the inclination angle of the box body, rainfall, lightning activity, and whether there is biological intrusion; the equipment parameters include current, voltage, as well as the maintenance records and cumulative operation time of the corresponding electrical components; the data acquisition module collects the environmental parameters and equipment parameters at a preset acquisition frequency (i.e., setting the acquisition interval time t), and combines them to construct a basic data set X(t);

[0081] A data processing module, which is used to perform data cleaning, noise reduction processing, and format standardization processing on the basic data set, so as to convert it into a standard data set S(t), facilitating subsequent analysis and monitoring;

[0082] A data analysis module, which is used to screen out abnormal data in the standard data set S(t) and merge them to construct an abnormal data set A(t); it is also used to extract a comprehensive interference index and a health assessment index from the standard data set S(t);

[0083] An abnormal diagnosis module, which calculates a dynamic anti-interference threshold based on the extracted comprehensive interference index and health assessment index, then extracts the confirmed data in the abnormal data set A(t) based on the dynamic anti-interference threshold, and merges all the confirmed data to construct a confirmed data set D(t);

[0084] A data repair module, which is used to correct the confirmed data, obtain the repaired parameter data and transmit it to the user side.

[0085] Specifically, in this case, the method for the data analysis module to screen abnormal data is as follows:

[0086] Extract the parameter data (i.e., the collected environmental parameters and device parameters) in the standard data set S(t), compare the parameter data with the pre-set static threshold range, and record the parameter data within the static threshold range as normal data; compare the parameter data outside the static threshold range with the pre-set dynamic threshold range, and record the parameter data outside the dynamic threshold range as risk data; the setting method of the dynamic threshold range is as follows:

[0087] Collect the parameter data within a fixed past time and construct a historical data set {x 1 , x 2 ,..., x N}; where x is the name index of the parameter data and N is the sample number of the parameter data; first calculate the mean of the parameter data, and the calculation formula is: In the formula, represents the mean of a set of parameter data, which is used to reflect the central tendency of the parameter data; x i is the i-th parameter data; then calculate the standard deviation of the parameter data, and the calculation formula is: In the formula, σ x represents the standard deviation of a set of parameter data, which reflects the degree of dispersion of the data distribution; set the dynamic threshold range of abnormal data based on the mean and standard deviation of the parameter data Among them, is the lower limit of the dynamic threshold, is the upper limit of the dynamic threshold; ω is the threshold coefficient, which determines the width of the dynamic threshold range. In statistics, the fluctuation characteristics of data distribution usually follow a normal distribution. Therefore, the threshold coefficient ω usually takes a value of 2.0 or 3.0;

[0088] By setting the dynamic threshold range, the personalized thresholds of historical data sets of different parameter data can be satisfied, accurately adapting to the volatility of different parameter data, and effectively identifying the abnormal points in the data collected by the metering box.

[0089] More specifically, in this case, the collaborative fluctuation abnormality of the metering box in a complex abnormal scenario cannot be determined by the abnormality of a single parameter data (i.e., risk data); the data analysis module constructs a collaborative analysis model based on the basic data set X(t), and then identifies the abnormal association mode between multiple parameter data under specific conditions; the method of constructing the collaborative analysis model is as follows:

[0090] Extract the parameter data with an associated relationship between the basic data sets X(t), and construct it into the current state vector XV = [D x1 , D x2 ,..., D xM ; where D x1 , D x2 ,..., D xM represent different parameter data with an associated relationship, and M is the number of acquisitions of parameter data with an associated relationship; for example, when the temperature in the metering box rises, it may cause an increase in current and voltage fluctuations; then extract the parameter data within a fixed past time and construct it into the historical state vector Then calculate the deviation degree DEV between the current state vector and the historical state vector. The calculation formula is: In the formula, is the Euclidean distance between the current state vector and the historical state vector. The larger the distance, the more serious the current state deviates from the historical benchmark; D xi is the current value of the i-th parameter data, is the historical value of the i-th parameter data; by comparing the deviation degree DEV with a preset deviation threshold, the current state vector with a deviation degree DEV greater than the deviation threshold is marked as an abnormal state;

[0091] Extract all the parameter data in the abnormal state and calculate the deviation amount ΔD xi of each parameter data. The calculation formula is: Then calculate the total deviation value Among them, squaring the deviation amount ΔD xi of each parameter data is to eliminate the influence of positive and negative signs and amplify the influence of larger deviations; for each parameter data, calculate its contribution ratio in the total deviation value. The calculation formula is: Calculate the contribution ratio threshold TH for each data parameter again xi , and the calculation formula is: In the formula, is the historical contribution ratio mean of the i-th parameter data, which is used to reflect the typical level of deviation of this parameter data from the overall anomaly in the normal state. The acquisition formula is: Where is the historical contribution ratio of the i-th parameter data, T is the total number of sampling time points; μ xi is the historical contribution ratio standard deviation of the i-th parameter data, which measures the volatility of the contribution ratio of the i-th parameter data in the historical normal data; τ is the discrete range factor, which is used to define the range width of parameter anomalies, and is generally set to 2.0 or 3.0 based on experience;

[0092] Compare the contribution ratio Cont xi of all parameter data in the abnormal state with the corresponding contribution ratio threshold TH xi , and mark the parameter data with the contribution ratio Cont xi greater than the contribution ratio threshold TH xi as abnormal data. At the same time, sort all abnormal data in descending order according to the difference between the contribution ratio Cont xi and the contribution ratio threshold TH xi to determine the main abnormal source and the secondary abnormal source; and merge all abnormal data to construct an abnormal data set A(t), and merge all abnormal sources to construct an abnormal source collection AEC(t).

[0093] By statistically analyzing the historical data of long-term operation, dynamically quantify the distribution range of normal data, so that the determination of abnormal data has solid data support, rather than relying on manual experience or a single rule, making the applicability of this solution to the determination of abnormal data stronger; and in the process of identifying abnormal data, not only can it identify whether the current state is abnormal, but also can accurately judge the main problem source, analyze the main abnormal source and the secondary abnormal source, reducing the ambiguity of abnormal analysis.

[0094] It should be noted that in this case, the method for confirming the main abnormal source and the secondary abnormal source is as follows:

[0095] Extract the contribution ratio Cont xi of all abnormal data, and calculate the main abnormal threshold and the associated abnormal threshold. The calculation formula is:

[0096]

[0097] In the formula, TH main is the main abnormal threshold, and the abnormal data with a contribution ratio greater than the main abnormal threshold is the main abnormal source; TH relis the correlation anomaly threshold. Anomaly data with a contribution ratio greater than the correlation anomaly threshold and less than the main anomaly threshold is the secondary anomaly source; is the average contribution ratio of all anomaly data; is the contribution ratio of anomaly data arranged in descending order (from largest to smallest), representing several anomaly data with the greatest impact on the overall anomaly; E is the total number of anomaly data; u is the dynamic adjustment coefficient. If u is high, it is more sensitive to the collaborative anomalies of multi-parameter data and is applicable to critical equipment systems. Conversely, if u is low, it is more inclined to ignore minor collaborative anomalies and is more applicable to stable working scenarios; v is the fixed offset part, representing the reference value for collaborative anomaly determination.

[0098] Furthermore, in this case, the setting of the contribution ratio threshold based on historical data may be interfered by the external environment. For example, bad weather and animal and plant activities will have a significant impact on the monitoring data of electronic components in the metering box, resulting in false alarms or missed alarms in anomaly detection. The data analysis module constructs a dynamic anomaly analysis model based on the standard data set S(t) and the anomaly data set A(t) to obtain a comprehensive interference index, which is used to correct the contribution ratio threshold, thereby improving the accuracy of anomaly data diagnosis. The construction method of the dynamic anomaly analysis model is as follows:

[0099] Extract interference parameters from the standard data set S(t), including the temperature change range ΔT env (affecting the accuracy of electrical components), the humidity change range ΔH env (causing detection deviation of electrical components and even electrical insulation problems), the wind speed fluctuation max(W speed )(physically interfering with electrical components), the cumulative rainfall ΔP rain (affecting the errors related to humidity and temperature), the lightning strike frequency F lgt (electromagnetic interference causing anomalies in electrical signals such as current and voltage), the animal and plant activity frequency F ap (directly contacting, biting, vibrating and impacting electrical components), the animal and plant occlusion area A ap (affecting the measurement of optical sensors in electrical components) and the metering box offset angle θ tilt (affecting the measurement accuracy of sensors and possibly causing abnormal equipment operation); all interference parameters are combined to construct an interference set Z;

[0100] Calculate the comprehensive interference index based on the interference set. The calculation formula is:

[0101]

[0102] In the formula, CII is the comprehensive interference index, which quantifies the influence degree of all interference parameters on the operation of the metering box; z is the name index of the interference parameter; S zis the standardized value of the interference parameter, and the calculation formula is: where C z is the current value of the interference parameter, is the historical mean of the interference parameter, and σ z is the historical standard deviation of the interference parameter; by adjusting different interference parameters to a dimensionless standardized range, it is convenient to compare the influence degrees of different interference factors; λ is the non-linear adjustment index, which is generally set based on experience; is the non-linear adjustment factor, and |S z | is the absolute value of the standardized value of the interference parameter, which can represent the degree of deviation of the standardized value of the interference parameter from the mean; through the non-linear adjustment factor, the influence of moderate changes can be smoothly strengthened while suppressing the influence of extreme changes, so as to avoid the unreasonable amplification of outliers on the overall index; specifically, for small changes (i.e., when |S z | is small), the interference factor grows approximately linearly; for extreme changes (i.e., when |S z | is large), the increase amplitude of the interference index is limited to avoid the excessive amplification of extreme values on the result; γ z is the importance coefficient of the interference parameter, indicating the influence weight of this interference parameter on the comprehensive interference index, and the value range is: 0 < γ z ≤ 1; η z is the normalized weight of the interference parameter, satisfying ∑ z∈Z η z = 1;

[0103] Furthermore, in this case, during the long-term operation of the metering box, the internal electrical components will age during long-term operation, and problems may occur and maintenance may be carried out during operation; the data analysis module constructs a device health assessment model based on the standard data set S(t) to obtain the health assessment index of the electrical components, which is used to correct the contribution ratio threshold, thereby improving the accuracy of abnormal data diagnosis; the construction method of the device health assessment model is:

[0104] Extract health scoring parameters from the standard data set S(t), including the cumulative operation duration historical maintenance times Based on the cumulative operation duration calculate the aging score of the electrical component, and the calculation formula is: In the formula, is the aging score of the electrical component for collecting the corresponding parameter data; The characteristic life of the corresponding parameter data is collected to measure the typical operating time of the electrical component during its entire life cycle, which is usually obtained based on historical data fitting; b is a shape parameter used to characterize the aging behavior (i.e., health curve) or failure mode of the electrical component. Specifically: when b<1, it means that the electrical component is in early decline (i.e., the probability of aging failure is low), and the aging rate of the electrical component gradually accelerates with the increase of the cumulative operating time; when b=1, it means that the electrical component has a constant failure rate (i.e., the aging change is an exponential function relationship); when b>1, it means that the electrical component is in an accelerated aging stage (i.e., the initial aging rate of the electrical component is slow, but as the cumulative operating time increases, the aging rate accelerates); based on the historical maintenance times N maint Calculate the maintenance score of electrical components using the following formula: In the formula, Provide maintenance scores for electrical components that collect corresponding parameter data; The theoretical maximum number of maintenance times for the electrical component; The time interval between the last maintenance of the electrical component; The theoretically allowed longest maintenance interval for the electrical component; α 1 is the weight coefficient of the maintenance times, α 2 is the weight coefficient of the maintenance time interval; finally, the equipment health assessment index is calculated In the formula, β 1 is the weight coefficient of aging score, β 2 The weight coefficient for the maintenance score.

[0105] It is worth mentioning that in this case, the shape parameter b is a parameter that describes the decay rate of the health status of the electrical component over time, which is usually based on historical data fitting activities; however, in the process of historical data fitting, the shape parameter b will be disturbed by abnormal emergencies, which will lead to errors or deviations, and thus affect the accuracy of the health curve; among them, abnormal emergencies include: voltage anomalies and overload shocks of electrical components during operation, which will cause the health of electrical components to drop sharply in a short period of time; also include electrical components being subjected to abnormal vibration peaks, causing the electrical components to show random upper limit fluctuations in a certain period of time; and isolated abnormal pairs appear in the data collected by electrical components, causing the overall trend to deviate significantly from the true value;

[0106] The data analysis module performs fitting correction on the shape parameter b based on the Bayesian dynamic update correction model. The fitting correction method is:

[0107] Under uncertain conditions, the shape parameter b is predicted by prior distribution and observed data, and the prediction formula is:

[0108] P(b)∝P(b prior )P(S(t)|bprior ); where P(b) is the predictive distribution, representing the comprehensive inference result of the shape parameter b based on prior knowledge and current observed data; P(b prior ) is the prior distribution, representing the preliminary assumption of the shape parameter b based on historical experience before observing data; b prior is the assumed shape parameter, used to compare with the actual observed value (shape parameter b); P(S(t)|b prior ) is the likelihood function, representing the probability of the observed data (i.e., the standard data set S(t) collected in the current state) occurring under the condition of the assumed shape parameter b prior ;

[0109] Combining the predicted likelihood value and the prior distribution, calculate the posterior distribution of the shape parameter b. The calculation formula is: P(b post ) ∝ P(S(t)|b)·P(b); where P(b post ) is the posterior distribution, representing the corrected probability distribution of the shape parameter b under the observed data, which is the updated predicted result of the shape parameter b after combining the prior distribution and the observed data;

[0110] Use the updated posterior distribution P(b post ) as the prior distribution for the next round, repeat the above steps, implement recursive correction of the shape parameter b, so as to dynamically adjust the shape parameter b and return it to the equipment health assessment model to update the equipment health assessment index;

[0111] Among them, the calculation expression of the prior distribution is: where μ b is the prior mean, representing the preliminary assumed value of the shape parameter b, that is, in the absence of any observed data, the shape parameter b is most likely to approach the prior mean μ b ; is the prior variance, representing the uncertainty range of the preliminary assumption of the shape parameter b. The larger its value, the more it affects the fluctuation range of the shape parameter b; exp(...) is the exponential function;

[0112] The calculation expression of the likelihood function is:

[0113]

[0114] where is the product symbol, J is the total number of data points in the standard data set S(t); is the normalization coefficient of the normal distribution, used to ensure that the total probability density of the entire distribution is 1, where σ ε represents the uncertainty or noise magnitude of the observed data. The larger the value, the greater the data error is considered; H(t j |bprior indicating the health value corresponding to the assumed shape parameter b prior and the predicted j-th time point t j ; (S(t) j -H(t j |b prior )) represents the deviation between the observed value and the predicted value, reflecting the fitting effect on the actual data; S(t) j is the j-th data point in the standard data set S(t);

[0115] By combining the observed data and historical experience, and providing recursive iteration to continuously optimize the predicted value of the shape parameter b, the device health assessment model can more accurately reflect the true operating conditions of electrical components, solving the limitations brought by the fixed assumption of the shape parameter b; at the same time, based on the probability distribution to describe the uncertainty of the parameters, rather than only giving a single definite value, allowing real-time dynamic update of the predicted value, and being able to adjust the results with the change of the environment, thus increasing the adaptability of the device health assessment model in a non-linear and uncertain environment, especially in the application of health monitoring; in particular, the predicted value of the shape parameter b depends on the real-time data iteration update, can quickly capture the abnormal state caused by abnormal emergencies, and more timely reflect the device deterioration problem, improving the accuracy of the device health assessment index results;

[0116] When fitting and correcting the shape parameter b, it is based on the standard data set S(t). Most of the data distribution in the standard data set S(t) is in the normal operating area, and the distribution under extreme conditions is incomplete and sparse; resulting in a distribution deviation when the shape parameter b is fitted and corrected based on the normal health state data, and it cannot be extended to the full distribution range, limiting the application scenarios of the model;

[0117] The data analysis module combines the GAN model to generate extreme condition data and uses it for the correction and optimization of the shape parameter b. The generation method is as follows:

[0118] Collect the health index set of the standard data set S(t) at time t and label it as the real data set S real ; Based on the normal distribution range, construct boundary conditions and define the upper threshold TH real of the real data set S UB and the lower threshold TH NT , and the extreme conditions will fall outside the upper and lower thresholds;

[0119] Design a GNN model for generating the extreme health state distribution, including a generator G and a discriminator D;

[0120] The generator G is used to receive the random noise vector Ψ~N(0, 1) and the prior health state distribution P(S realTake it as input and then output the generated dataset S of the extreme health state gen , whose distribution is defined as the health state samples under the boundary conditions; the generator G is used to minimize the accuracy of the discriminator, making the generated dataset S gen closer to the true distribution;

[0121] The discriminator D is used to receive the standard dataset S(t) and the generated distribution from the real dataset as inputs, and discriminate the probability whether the standard dataset S(t) comes from the real distribution; the discriminator D is used to maximize the correct discrimination of the real dataset, making the generated dataset closer to the real distribution;

[0122] Define the loss function of the GNN model:

[0123]

[0124] In the formula, It means that the goal of the discriminator D in the model is to maximize V(D, G), judge that the real dataset is real (D(S real ) → 1), and at the same time judge that the generated dataset is fake (D(T gen ) → 0), and minimize V(D, G), making the generated dataset indistinguishable from true or false by the discriminator D (D(G(ψ)) → 1). Finally, let the discriminator D and the generator G compete to find the best balance point;

[0125] S real ~P(S real ) represents the real dataset sampled from the prior health state distribution; log D(S real represents the confidence of the real dataset, that is, the more convinced the discriminator is that the data is real, the closer the output D(S real ) of the discriminator is to 1, and at this time the value of log D(S real ) is larger;

[0126] Ψ~P Ψ represents a random vector sampled from the random noise distribution P Ψ as the input of the generator; G(Ψ) is the generated sample generated by the generator simulating the real data distribution; D(G(Ψ)) is the output of the discriminator for the generated sample, representing the probability that the discriminator thinks the generated sample comes from the real data distribution; log(1 - D(G(Ψ)) represents the ability of the discriminator to identify fake samples;

[0127] Add the boundary loss term L bnd to the loss function to ensure that the generated dataset falls within the boundary region; where hyp is a hyperparameter that controls the strength of the boundary constraint;

[0128] Fuse the generated dataset and the real dataset to construct an extended dataset Sext , and return the extended dataset S ext to the Bayesian dynamic update and correction model to optimize and correct the shape parameter b. Among them, the generated dataset makes up for the deficiencies of the real dataset, ensures that the correction of the shape parameter b covers the entire distribution, makes the correction of the shape parameter b more robust, and can cope with the state changes of the device under extreme aging or fault conditions.

[0129] The data confirmation module calculates the dynamic anti-interference threshold based on the comprehensive interference index and the health assessment index, and further verifies the abnormal data through the dynamic anti-interference threshold to obtain the confirmed data; it avoids misjudging the normal fluctuations of parameter data as abnormal under the influence of bad weather and animal and plant activities, or when the device is operating under changing health states; the dynamic anti-interference threshold The calculation formula of is:

[0130] In the formula, (δ 1 · CII) is the interference correction term, and δ 1 is the weight coefficient of the comprehensive interference index; (1 + δ 2 (1 - HAI x )) is the health correction term, and δ 2 is the weight coefficient of the health assessment index;

[0131] Compare all the abnormal data in the abnormal dataset A(t) with the corresponding dynamic anti-interference threshold in turn, mark the abnormal data greater than the dynamic anti-interference threshold as the confirmed data, and finally merge all the confirmed data to construct the confirmed dataset D(t).

[0132] By quantifying and calculating the dynamic and random external interferences, abnormal data can be effectively extracted from the data detected by the metering box. Compared with the traditional method of identifying static thresholds, abnormal data can be identified more accurately, and it can avoid misjudging normal fluctuations as abnormal when the external environment changes; at the same time, the dynamic threshold adjustment mechanism can achieve personalized optimization by constructing an analysis model for electrical components for different detected data, making the overall system more compatible, responding to environmental changes in real time, adapting to a variety of complex scenarios, and realizing unified dynamic anomaly detection.

[0133] The way for the data repair module to correct the confirmed data is:

[0134] Extract all the confirmed data from the confirmed dataset D(t), and then extract the abnormal sources of the confirmed data from the abnormal source collection AEC(t); correct the confirmed data based on the abnormal sources, and the correction formula is:

[0135]

[0136] In the formula, The parameter data after repair reflects the actual measured true value of the metering box during the measurement process under the influence of external interference and equipment problems, removes noise, and compensates for the abnormal data deviation caused by external interference and equipment problems; For the diagnosed data; For the abnormal source correction term, where: k Q Is the correction coefficient corresponding to the abnormal source, Q is the name index of the abnormal source, Is the current value of the abnormal source, Is the best value of the abnormal source, Is the limit upper value of the abnormal source, Is the limit upper value of the abnormal source; For the equipment health correction term, where: ΔHAI x Is the equipment health correction value, and the calculation formula is: Is the reference evaluation index in the ideal health state, Is the weight coefficient of the equipment health.

[0137] By specifically correcting the diagnosed data, removing the abnormal sources that cause the abnormality of the parameter data, and the deviation caused by the problems of electrical component equipment, the accuracy of the parameter data collected by the metering box can be effectively improved, ensuring that the data received by the user side is closer to the true value; and when correcting the diagnosed data, the specific reasons for the data abnormality are fully considered, providing a specific basis for the repair of the data, and effectively improving the credibility of the parameter data after repair.

[0138] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A remote operation control system for an outdoor metering box based on the Internet of Things, characterized in that: It includes a data acquisition module, which is used to collect the equipment parameters of the meter box and the environmental parameters where the meter box is installed, and merge the equipment parameters and the environmental parameters to construct a basic data set X(t); it also includes: A data processing module is used to preprocess the basic data set and convert it into a standard data set S(t); The data analysis module is used to filter abnormal data from the standard data set S(t) and merge them into an abnormal data set A(t); it is also used to extract a comprehensive interference index and a health assessment index from the standard data set S(t); based on the standard data set S(t), a device health assessment model is constructed to obtain the health assessment index of the electrical components; based on the abnormal data set A(t), a dynamic abnormal analysis model is constructed to obtain a comprehensive interference index; The abnormal diagnosis module calculates the dynamic anti-interference threshold based on the extracted comprehensive interference index and health assessment index, and then extracts the confirmed data in the abnormal data set A(t) based on the dynamic anti-interference threshold, and constructs it into the confirmed data set D(t); The data repair module is used to correct the confirmed data, obtain the repaired parameter data and transmit it to the user end.

2. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 1, characterized in that: The environmental parameters include temperature, humidity, wind speed, cabinet tilt angle, rainfall, lightning activity and whether there is biological infestation; the equipment parameters include current, voltage, and maintenance records and cumulative operating time of corresponding electrical components.

3. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 2, characterized in that: The method of filtering abnormal data is: Extract parameter data from the standard data set S(t), compare the parameter data with a preset static threshold range, and record the parameter data within the static threshold range as normal data; Compare the parameter data beyond the static threshold range with the pre-set dynamic threshold range, and record the parameter data beyond the dynamic threshold range as risk data; A collaborative analysis model is constructed based on risk data to obtain abnormal data and abnormal sources; the abnormal sources are merged to construct an abnormal source collection AEC(t).

4. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 3, characterized in that: The dynamic threshold range is set as follows: Collect parameter data within a fixed period of time in the past and construct it into a historical data set {x1, x2, ..., x N }; where x is the name index of the parameter data, and N is the number of samples of the parameter data; Calculate the mean of parameter data using the formula: In the formula, represents the mean of parameter data; x i is the i-th parameter data; Calculate the standard deviation of the parameter data using the formula: In the formula, σ x Indicates the standard deviation of parameter data; Set the dynamic threshold range of abnormal data based on the mean and standard deviation of parameter data in, is the lower limit of the dynamic threshold, is the upper limit of the dynamic threshold; ω is the threshold coefficient.

5. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 3, characterized in that: The method of constructing the collaborative analysis model is: Extract the parameter data with correlation between the basic data sets X(t) and construct it into the current state vector XV = [D x1 , D x2 , …, D xM ]; among them, D x1 , D x2 , …, D xM represents different parameter data with associated relationships, and M is the number of collected parameter data with associated relationships; Then extract the parameter data within a fixed time in the past and construct it into a historical state vector Calculate the deviation DEV between the current state vector and the historical state vector. The calculation formula is: In the formula, is the Euclidean distance between the current state vector and the historical state vector; D xi is the current value of the i-th parameter data, is the historical value of the i-th parameter data; Compare the deviation DEV with a preset deviation threshold, and mark the current state vector whose deviation DEV is greater than the deviation threshold as an abnormal state; Extract all parameter data in abnormal state and calculate the deviation ΔD of each parameter data xi , the calculation formula is: Then calculate the total deviation For each parameter data, calculate its contribution ratio in the total deviation value. The calculation formula is: Then calculate the contribution ratio threshold TH of each data parameter xi , the calculation formula is: In the formula, is the mean value of the historical contribution ratio of the i-th parameter data; μ xi is the standard deviation of the historical contribution ratio of the i-th parameter data; τ is the discrete range factor; The contribution ratio of all parameter data in abnormal state is Cont xi And the corresponding contribution ratio threshold TH xi Compare and convert the contribution ratio Cont xi Greater than contribution ratio threshold TH xi The parameter data is marked as abnormal data.

6. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 3, characterized in that: The method for determining the source of the abnormality is as follows: Extract the contribution ratio of all abnormal data Cont xi , calculate the main anomaly threshold and the associated anomaly threshold, the calculation formula is: Where TH main is the main anomaly threshold, and the abnormal data with a contribution ratio greater than the main anomaly threshold is the main anomaly source; TH rel is the associated anomaly threshold, and the abnormal data with a contribution ratio greater than the associated anomaly threshold and less than the main anomaly threshold is the secondary anomaly source; is the average contribution ratio of all abnormal data; is the contribution ratio of abnormal data in descending order; E is the total number of abnormal data; u is the dynamic adjustment coefficient; v is the fixed offset part.

7. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 3, characterized in that: The method of constructing the equipment health assessment model is: Extract disturbance parameters from the standard data set S(t), including the temperature variation amplitude ΔT env , Humidity change range ΔH env 、Wind speed fluctuation max(W speed ), accumulated rainfall ΔP rain , lightning frequency F lgt 、Frequency of animal and plant activities ap 、Area blocked by plants and animals A ap and meter box offset angle θ tilt ; All interference parameters are combined to form the interference set Z; The comprehensive interference index is calculated based on the interference set, and the calculation formula is: Where, CII is the comprehensive interference index; z is the name index of the interference parameter; S z is the normalized value of the interference parameter; λ is the nonlinear adjustment index; is the nonlinear adjustment factor; γ z is the importance coefficient of the interference parameter, and its value range is: 0<γ z ≤1;η z is the normalized weight of the interference parameter, satisfying ∑ z∈z η z =1; The method of constructing the dynamic anomaly analysis model is: Extract health score parameters from the standard dataset S(t), including the cumulative running time Historical maintenance times Based on cumulative running time Calculate the aging score of electrical components using the following formula: In the formula, To collect the aging scores of electrical components corresponding to the parameter data; The characteristic life of the corresponding parameter data is collected; b is the shape parameter; Based on the historical maintenance times N maint Calculate the maintenance score of electrical components using the following formula: In the formula, Provide maintenance scores for electrical components that collect corresponding parameter data; The maximum number of maintenance times for the electrical component; The time interval between the last maintenance of the electrical component; is the longest allowable maintenance time interval of the electrical component; α1 is the weight coefficient of the maintenance times, and α2 is the weight coefficient of the maintenance time interval; The formula for calculating the equipment health assessment index is: Where, HAI x is the health assessment index of the electrical component for which the corresponding parameter data is collected; β1 is the weight coefficient of the aging score, and β2 is the weight coefficient of the maintenance score; The formula for calculating the dynamic anti-interference threshold is: Where (δ1·CII) is the interference correction term, δ1 is the weight coefficient of the comprehensive interference index; (1+δ2(1-HAI x )) is the health correction term, and δ2 is the weight coefficient of the health assessment index.

8. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 7, characterized in that: The fitting correction method of the shape parameter b is: Under uncertain conditions, the shape parameter b is predicted by prior distribution and observed data. The prediction formula is: P(b)∝P(b prior )·P(S(t)|b prior ), where P(b) is the predicted distribution; P(b prior ) is the prior distribution; b prior is the assumed shape parameter; P(S(t)|b prior ) is the likelihood function; Combining the predicted likelihood value and prior distribution, the posterior distribution of shape parameter b is calculated as follows: P(b post )∝P(S(t)|b)·PP(b); Formula In the post ) is the posterior distribution; The updated posterior distribution P(b post ) is used as the prior distribution for the next round and the above steps are repeated; Among them, the calculation expression of the prior distribution is: The calculation expression of the likelihood function is: In the formula, μ b is the prior mean; is the prior variance; exp(...) is the exponential function; is the multiplication symbol, J is the total number of data points in the standard data set S(t); S(t)j is the jth data point in the standard data set S(t); σ ε represents the uncertainty of observation data; H(t j |b prior ) represents the assumed shape parameter b prior and the predicted j-th time point t j The corresponding health value.

9. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 8, characterized in that: The data analysis module combines the GAN model to generate extreme condition data in the following way: Collect the health indicator set of the standard data set S(t) at time t and mark it as the real data set S real ; Define the real data set S real The previous threshold TH UB and the next threshold TH NT ; The structure of the GAN model is defined as the generator G and the discriminator D; The generator G is used to receive the random noise vector Ψ~N(0,1) and the prior health state distribution P(S real ) as input and output a generated dataset S of extreme health states gen ; The discriminator D is used to receive the standard data set S(t) and the generated distribution from the real data set as input, and to determine the probability of whether the standard data set S(t) comes from the real distribution; Define the loss function of the GNN model: In the formula, S real ~P(S real ) represents the real data set sampled from the prior health state distribution; log D(S real ) represents the confidence of the real data set; Ψ~P Ψ Represents the random noise distribution P Ψ The random vector sampled in ; G(Ψ) is the generated sample generated by the generator simulating the real data distribution; D(G(Ψ)) is the output of the discriminator for the generated sample; Add the boundary loss term L to the loss function bnd , the boundary loss term is calculated as: Where hyp is a hyperparameter that controls the strength of boundary constraints; The generated dataset and the real dataset are fused to construct the extended dataset S ext , and expand the dataset S ext Return to the Bayesian dynamic update correction model.

10. The remote operation control system of an outdoor metering box based on the Internet of Things as claimed in claim 6, characterized in that: The method for correcting the confirmed data is as follows: Confirmed data are extracted from the confirmed data set D(t), and then the abnormal sources of the confirmed data are extracted from the abnormal source collection AEC(t); the correction formula is: In the formula, is the parameter data after repair; For confirmed data; is the anomaly source correction term, where: k Q is the correction coefficient of the corresponding abnormal source, Q is the name index of the abnormal source, is the current value of the exception source, is the optimal value of the anomaly source, is the upper limit value of the abnormal source, is the upper limit value of the abnormal source; is the equipment health correction term, where: ΔHAI x is the equipment health correction value, and the calculation formula is: It is a benchmark assessment index under health status. is the weight coefficient of the equipment health.

Images