Intelligent load side power utilization safety prevention and control system

Through the intelligent load-side power safety prevention and control system, the data collection, preprocessing and power consumption behavior identification models are used to solve the problem of low efficiency in power safety inspection on the load-side, and efficient power safety prevention and control is achieved, reducing the efficiency of power safety prevention and control. The incidence rate of accidents.

CN119030147APending Publication Date: 2024-11-26SHANGHAI AITAO INFORMATION TECH DEV CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411193655.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The efficiency of power safety inspection on the load side is inefficient, making it difficult to quickly and comprehensively check for power facilities failures, resulting in low efficiency of power safety prevention and control.

Method used

An intelligent load-side power safety prevention and control system was designed, including a data acquisition module, a data preprocessing module and an electricity behavior recognition model. By collecting and preprocessing power data, an electricity behavior recognition model is constructed and a different use is generated. The probability value of electrical behavior and trigger corresponding prevention and control measures based on the probability value.

Benefits of technology

It improves the efficiency of power safety prevention and control on the load side, accurately identify abnormal power usage behaviors, and reduces the incidence of power safety accidents.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119030147A_ABST
    Figure CN119030147A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of power utilization safety prevention and control, and discloses an intelligent load side power utilization safety prevention and control system, which comprises a data acquisition module used for acquiring power data of a power utilization scene according to a preset time interval t in a preset time period T, and constructing a first feature sequence; the data preprocessing module is used for preprocessing the first feature sequence to generate a second feature sequence; a power consumption behavior recognition model construction module which is used for constructing a power consumption behavior recognition model; the input of the electricity consumption behavior recognition model is a second feature sequence, the output is probability values of different electricity consumption behaviors, and the electricity consumption behaviors comprise normal behaviors, overload electricity consumption, privately pulling and disorderly connecting wires, winding over or changing a metering device and using poor-quality electrical equipment; and the prevention and control measure generation module is used for generating prevention and control measures for the power consumption scene according to the probability values of the different power consumption behaviors.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of electricity safety prevention and control, and more specifically, to an intelligent load-side electricity safety prevention and control system. Background Art

[0002] With the development of society and the advancement of technology, electricity has become an indispensable part of modern life. Both daily life and industrial production cannot be separated from the support of electricity. In the past decade, with the access of new energy to the power grid on the load side, and the increase in social demand for electricity and the increase in power, these changes are quietly changing the structure of load-side electricity consumption and bringing more challenges to the safety prevention and control of electricity consumption. At present, the safety of electricity consumption on the load side is usually checked by professional electricians for safety inspections of power facilities. However, in the face of the complexity and diversity of current power distribution equipment and power-consuming equipment, it is difficult to quickly and comprehensively check the safety of electricity consumption on the load side by relying solely on electricians, resulting in low efficiency in the safety inspection of electricity consumption on the load side.

[0003] Therefore, there is an urgent need for a load-side power safety prevention and control system to solve the above problems. Summary of the invention

[0004] The present invention provides an intelligent load-side electricity safety prevention and control system to solve the technical problems of low efficiency and incomplete inspection of load-side electricity safety in related technologies.

[0005] The present invention provides an intelligent load-side electricity safety prevention and control system, comprising:

[0006] A data collection module, which is used to collect power data of power usage scenarios within a preset time period T and at preset time intervals t, and construct a first feature sequence;

[0007] The first characteristic sequence includes N sequence units, and the nth sequence unit represents the voltage, current, power and power consumption at the nth time point;

[0008] The power consumption is represented by a time domain waveform, where the horizontal axis represents the time point and the vertical axis represents the amplitude;

[0009] Electricity usage scenarios include: factory production, corporate office, household electricity, vehicle charging and temporary electricity use;

[0010] A data preprocessing module, which is used to preprocess the first feature sequence to generate a second feature sequence;

[0011] The second feature sequence includes N sequence units, and the nth sequence unit represents the feature data obtained after preprocessing at the nth time point;

[0012] A power consumption behavior recognition model building module, which is used to build a power consumption behavior recognition model;

[0013] The input of the electricity consumption behavior recognition model is the second feature sequence, and the output is the probability value of different electricity consumption behaviors;

[0014] Electricity usage behaviors include: normal behavior, overloading, unauthorized wiring, bypassing or modifying metering devices, and using inferior electrical equipment;

[0015] The prevention and control measures generation module is used to generate prevention and control measures for power usage scenarios based on the probability values ​​of different power usage behaviors.

[0016] Furthermore, N, n, T and t are all custom parameters, 1≤n≤N, N=T / t.

[0017] Furthermore, the pre-processing step includes:

[0018] Step S201, for missing values ​​in the power data, filling processing is performed by taking the average value of the value of the previous sequence unit and the value of the next sequence unit;

[0019] Step S202, normalizing the power data using a maximum and minimum normalization method;

[0020] Step S203, creating a binary feature of "whether it is daytime";

[0021] Step S204, creating a binary feature of "whether it is the weekend";

[0022] Step S205, creating a binary feature of "whether it is a legal holiday";

[0023] Step S206, obtaining the maximum and minimum values ​​of electricity consumption and power every day and the average values ​​in the morning and afternoon, and creating a load characteristic;

[0024] Step S207, using fast Fourier transform to convert the time domain waveform representation of the power consumption into a frequency domain waveform representation, the horizontal axis of the frequency domain waveform represents the frequency, and the vertical axis represents the amplitude, extracting the power spectrum density of the low-frequency component from the frequency domain waveform, and creating a periodic feature, specifically selecting the first K frequency values ​​from the frequency domain waveform, and calculating the power spectrum density corresponding to the first K frequency values;

[0025] The formula for the number of frequency values ​​of a frequency domain waveform is:

[0026]

[0027] Among them, Fre N Represents the number of frequency values ​​of the frequency domain waveform, N is the number of sampling time points,

[0028] The power spectral density formula is:

[0029] Pf =|X f | 2 ;

[0030] Where P f represents the power spectral density, X f represents the complex value corresponding to frequency f, |X f | represents the modulus of a complex value. The formula for a complex value is:

[0031]

[0032] Where a and b represent the real part and imaginary part of the complex number a+bi obtained by fast Fourier transform of the time domain waveform respectively;

[0033] Step S208, creating statistical features by calculating the mean, standard deviation and median of power consumption;

[0034] The above preprocessed data are concatenated and represented by feature data.

[0035] Furthermore, the electricity consumption behavior recognition model includes a segmentation layer, L high The first hidden layer, L low A second hidden layer, a concatenation layer, and a classifier;

[0036] The segmentation layer is used to segment the second feature sequence into a third feature sequence and a fourth feature sequence, and the number of sequence units of the third feature sequence and the fourth feature sequence is the same as the number of sequence units of the second feature sequence;

[0037] The segmentation layer first calculates the standard deviation of each sequence unit of the second feature sequence, and marks the sequence unit as 1 when the standard deviation is greater than or equal to the standard deviation threshold, and marks the remaining sequence units as 0 as the third feature sequence. When the standard deviation is less than the standard deviation threshold, the sequence unit is marked as 2, and the remaining sequence units are marked as 0 as the fourth feature sequence, where the standard deviation threshold is a custom parameter;

[0038] That is, the sequence units of the third characteristic sequence only include unmarked sequence units and sequence units marked as 1;

[0039] That is, the sequence units of the fourth characteristic sequence only include unmarked sequence units and sequence units marked as 2;

[0040] For the first hidden layer and the second hidden layer number L high and L low ,satisfy:

[0041] L high =min(NI,3);

[0042] L low=min(NJ,2);

[0043] Where I represents the number of sequence units marked as 1 in the third feature sequence, and J represents the number of sequence units marked as 2 in the fourth feature sequence;

[0044] Each first hidden layer includes N first units, the nth first unit of the first hidden layer inputs the combination vector of the nth sequence unit of the third feature sequence, and outputs the first hidden feature;

[0045] No.1 high The first unit of the nth hidden layer inputs the lth high -1 The first hidden feature of the nth first unit output of the first hidden layer, output the first hidden feature, where 1≤l high ≤L high ;

[0046] Each second hidden layer includes N second units, the nth second unit of the first second hidden layer inputs the combination vector of the nth sequence unit of the fourth feature sequence, and outputs the second hidden feature;

[0047] No.1 low The nth second unit of the second hidden layer inputs the lth low -1 The second hidden feature output by the nth second unit of the second hidden layer, output the second hidden feature, where 1≤l low ≤L low ;

[0048] The splicing layer is used to combine the L high The first hidden feature output by the Nth first unit of the first hidden layer is concatenated to the Lth low After the second hidden feature output by the Nth second unit of the second hidden layer is obtained, the third hidden feature is obtained;

[0049] The third hidden feature is input into the classifier, and the classification space of the classifier represents the probability values ​​of different electricity usage behaviors.

[0050] Furthermore, the first high The formula for the first hidden layer is:

[0051]

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058] in and represents the nth sequence unit of the third feature sequence input by the nth first unit of the first hidden layer, and Respectively represent the first high The first hidden layer and the lth high-1 The first hidden feature output by the nth first unit of the first hidden layer, Indicates the first high The first hidden feature output by the n-1th first unit of the first hidden layer, and Respectively represent the first high The update gate, reset gate and candidate hidden features of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the update gate of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the reset gate of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the candidate hidden state of the nth first unit of the first hidden layer, ⊙ represents point-by-point multiplication, Indicates the first high Whether the nth first unit of the first hidden layer is activated. When it returns 1, the nth first unit is activated. When it returns 0, the nth first unit is skipped. Indicates the first high The state of the nth first unit of the first hidden layer, which takes the value of 1 or 0, indicating whether it is marked or unmarked, sigmoid represents the sigmoid activation function, tanh represents the hyperbolic tangent function, concat represents the concatenation operation, and MLP represents the multi-layer perceptron;

[0059] For the first low The formula for the second hidden layer is the same as above.

[0060] Furthermore, the formulas for the concatenation layer and the classifier are:

[0061]

[0062] in and Respectively represent the L high The first hidden feature of the first unit output of the first hidden layer and the first hidden feature of the Lth unit low The second hidden feature output by the Jth second unit of the second hidden layer, p b Represents the probability value of different power consumption behaviors of the output, E p represents the third weight parameter, b p It represents the bias parameter of behavior recognition, concat represents the concatenation operation, and softmax represents the softmax activation function.

[0063] Furthermore, the electricity usage behavior recognition model is pre-trained before training. During the pre-training process, the preprocessed second feature sequence is input into the pre-trained classifier. The classification space of the pre-trained classifier represents unauthorized power access, that is, the behavior of accessing the power line without the consent of the power company. The classification labels of the pre-trained classifier are manually labeled.

[0064] Furthermore, the loss function of the electricity usage behavior recognition model adopts the cross entropy loss function.

[0065] Furthermore, training samples of the electricity consumption behavior recognition model are obtained through manual labeling.

[0066] Furthermore, according to the probability value of each electricity usage behavior output by the electricity usage behavior recognition model, different initial thresholds are set to trigger corresponding prevention and control measures, among which the initial threshold of the probability of overload electricity usage is set to 0.2, the initial threshold of the probability of privately pulling and connecting wires is set to 0.1, the initial threshold of the probability of bypassing or modifying the metering device is set to 0.1, and the initial threshold of the probability of using inferior electrical equipment is set to 0.1. When the probability value output by the electricity usage behavior recognition model exceeds the initial threshold, the user is first notified by phone or text message to check the relevant equipment and remind them to pay attention to safe electricity use. If the behavior continues to occur within one day after the notification, staff will be dispatched to the site for inspection. In addition, an emergency response mechanism is set up. When the sum of the probability values ​​of other abnormal electricity usage behaviors other than normal behaviors exceeds 0.8, staff will be immediately dispatched to the site for inspection, and necessary measures will be taken to ensure electricity safety.

[0067] The beneficial effects of the present invention are as follows: the present invention accurately identifies abnormal electricity usage behavior through the probability value generated by the electricity usage behavior identification model, and takes corresponding prevention and control measures according to the set threshold value, thereby improving the efficiency of load-side electricity safety prevention and control and reducing the occurrence rate of electricity safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of an intelligent load-side electricity safety prevention and control system of the present invention;

[0069] Figure 2 is a flow chart of preprocessing power data of the present invention;

[0070] In the figure: data acquisition module 101, data preprocessing module 102, electricity consumption behavior recognition model building module 103 and prevention and control measures generation module 104. DETAILED DESCRIPTION

[0071] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0072] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0073] like Figure 1-2 As shown, an intelligent load-side power safety prevention and control system includes:

[0074] The data collection module 101 is used to collect power data of power usage scenarios within a preset time period T and at preset time intervals t, and construct a first feature sequence;

[0075] The first characteristic sequence includes N sequence units, and the nth sequence unit represents the voltage, current, power and power consumption at the nth time point;

[0076] The power consumption is represented by a time domain waveform, where the horizontal axis represents the time point and the vertical axis represents the amplitude;

[0077] Electricity usage scenarios include: factory production, corporate office, household electricity, vehicle charging and temporary electricity use;

[0078] A data preprocessing module 102, which is used to preprocess the first feature sequence to generate a second feature sequence;

[0079] The second feature sequence includes N sequence units, and the nth sequence unit represents the feature data obtained after preprocessing at the nth time point;

[0080] The power consumption behavior recognition model building module 103 is used to build a power consumption behavior recognition model;

[0081] The input of the electricity consumption behavior recognition model is the second feature sequence, and the output is the probability value of different electricity consumption behaviors;

[0082] Electricity usage behaviors include: normal behavior, overloading, unauthorized wiring, bypassing or modifying metering devices, and using inferior electrical equipment;

[0083] The prevention and control measures generating module 104 is used to generate prevention and control measures for power usage scenarios according to the probability values ​​of different power usage behaviors.

[0084] In one embodiment of the present invention, N, n, T and t are all custom parameters, 1≤n≤N, N=T / t, preferably, T is set to 1 minute, and t is set to 2 seconds.

[0085] In one embodiment of the present invention, a power data collection device is added to the molded case switch branch of the distribution side circuit branch, and the old molded case circuit breaker is replaced with an intelligent switch on the power consumption side, and a power data collection device is configured.

[0086] In one embodiment of the present invention, an electricity consumption prevention and control device is used to collect electricity data and perform safety prevention and control. This device is different from the Internet of Things gateway. While collecting electricity data, it monitors the safety status of the circuit in real time. This device has an embedded prevention and control strategy algorithm and selects different prevention and control strategy modes according to different electricity consumption scenarios.

[0087] In one embodiment of the present invention, the collected power data is transmitted to the power safety management platform for data preprocessing and input into the power consumption behavior recognition model.

[0088] In one embodiment of the present invention, Figure 2 As shown, the preprocessing steps include:

[0089] Step S201, for missing values ​​in the power data, filling processing is performed by taking the average value of the value of the previous sequence unit and the value of the next sequence unit;

[0090] Step S202, normalizing the power data using a maximum and minimum normalization method;

[0091] Step S203, creating a binary feature of "whether it is daytime", preferably, setting 6 to 18 o'clock as daytime, represented by 1, and setting 18 to 6 o'clock as night, represented by 0;

[0092] Step S204, creating a binary feature of "whether it is the weekend", for example, the weekend is represented by 1;

[0093] Step S205, creating a binary feature of "whether it is a legal holiday", for example, legal holidays are represented by 1;

[0094] Step S206, obtaining the maximum and minimum values ​​of electricity consumption and power and the average values ​​in the morning and afternoon, and creating a load feature. Preferably, 6 to 12 o'clock is set as the morning, and 12 to 6 is set as the afternoon;

[0095] Step S207, using fast Fourier transform to convert the time domain waveform representation of the power consumption into the frequency domain waveform representation, the horizontal axis of the frequency domain waveform represents the frequency, and the vertical axis represents the amplitude, extracting the power spectrum density of the low-frequency component from the frequency domain waveform, and creating a periodic feature, specifically selecting the first K frequency values ​​from the frequency domain waveform, preferably, K is set to Calculate the power spectral density corresponding to the first K frequency values;

[0096] The formula for the number of frequency values ​​of a frequency domain waveform is:

[0097]

[0098] Among them, Fre N Represents the number of frequency values ​​of the frequency domain waveform, N is the number of sampling time points,

[0099] The power spectral density formula is:

[0100] P f =|X f | 2 ;

[0101] Where P f represents the power spectral density, X f represents the complex value corresponding to frequency f, |X f | represents the modulus of a complex value. The formula for a complex value is:

[0102]

[0103] Where a and b represent the real part and imaginary part of the complex number a+bi obtained by fast Fourier transform of the time domain waveform respectively;

[0104] Step S208, creating statistical features by calculating the mean, standard deviation and median of power consumption;

[0105] The above preprocessed data are concatenated and represented by feature data.

[0106] In one embodiment of the present invention, the electricity consumption behavior recognition model includes a segmentation layer, L high The first hidden layer, L low A second hidden layer, a concatenation layer, and a classifier;

[0107] The segmentation layer is used to segment the second feature sequence into a third feature sequence and a fourth feature sequence, and the number of sequence units of the third feature sequence and the fourth feature sequence is the same as the number of sequence units of the second feature sequence;

[0108] The segmentation layer first calculates the standard deviation of each sequence unit of the second feature sequence, and determines that when the standard deviation is greater than or equal to the standard deviation threshold, the sequence unit is marked as 1, and the remaining sequence units are marked as 0 as the third feature sequence. When the standard deviation is less than the standard deviation threshold, the sequence unit is marked as 2, and the remaining sequence units are marked as 0 as the fourth feature sequence, wherein the standard deviation threshold is a custom parameter, and preferably, the standard deviation threshold is set to 0.3;

[0109] That is, the sequence units of the third characteristic sequence only include unmarked sequence units and sequence units marked as 1;

[0110] That is, the sequence units of the fourth characteristic sequence only include unmarked sequence units and sequence units marked as 2;

[0111] For the first hidden layer and the second hidden layer number L high and L low ,satisfy:

[0112] L high =min(NI,3);

[0113] L low =min(NJ,2);

[0114] Where I represents the number of sequence units marked as 1 in the third feature sequence, and J represents the number of sequence units marked as 2 in the fourth feature sequence;

[0115] Each first hidden layer includes N first units, the nth first unit of the first hidden layer inputs the combination vector of the nth sequence unit of the third feature sequence, and outputs the first hidden feature;

[0116] No.1 high The first unit of the nth hidden layer inputs the lth high -1 The first hidden feature of the nth first unit output of the first hidden layer, output the first hidden feature, where 1≤l high ≤L high ;

[0117] Each second hidden layer includes N second units, the nth second unit of the first second hidden layer inputs the combination vector of the nth sequence unit of the fourth feature sequence, and outputs the second hidden feature;

[0118] No.1 low The nth second unit of the second hidden layer inputs the lth low -1 The second hidden feature output by the nth second unit of the second hidden layer, output the second hidden feature, where 1≤l low ≤L low ;

[0119] The splicing layer is used to combine the L high The first hidden feature output by the Nth first unit of the first hidden layer is concatenated to the Lth low After the second hidden feature output by the Nth second unit of the second hidden layer is obtained, the third hidden feature is obtained;

[0120] The third hidden feature is input into the classifier, and the classification space of the classifier represents the probability values ​​of different electricity usage behaviors.

[0121] In one embodiment of the present invention, the first high The formula for the first hidden layer is:

[0122]

[0123]

[0124]

[0125]

[0126]

[0127]

[0128]

[0129] in and represents the nth sequence unit of the third feature sequence input by the nth first unit of the first hidden layer, and Respectively represent the first high The first hidden layer and the lth high-1 The first hidden feature output by the nth first unit of the first hidden layer, Indicates the first high The first hidden feature output by the n-1th first unit of the first hidden layer, and Respectively represent the first high The update gate, reset gate and candidate hidden features of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the update gate of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the reset gate of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the candidate hidden state of the nth first unit of the first hidden layer, ⊙ represents point-by-point multiplication, Indicates the first high Whether the nth first unit of the first hidden layer is activated. When it returns 1, the nth first unit is activated. When it returns 0, the nth first unit is skipped. Indicates the first high The state of the nth first unit of the first hidden layer, which takes the value of 1 or 0, indicating whether it is marked or unmarked, sigmoid represents the sigmoid activation function, tanh represents the hyperbolic tangent function, concat represents the concatenation operation, and MLP represents the multi-layer perceptron;

[0130] For the first low The formula for the second hidden layer is the same as above.

[0131] In one embodiment of the present invention, the formulas for the concatenated layer and the classifier are:

[0132]

[0133] in and Respectively represent the L high The first hidden feature of the first unit output of the first hidden layer and the first hidden feature of the Lth unit low The second hidden feature output by the Jth second unit of the second hidden layer, p b Represents the probability value of different power consumption behaviors of the output, E p represents the third weight parameter, b p It represents the bias parameter of behavior recognition, concat represents the concatenation operation, and softmax represents the softmax activation function.

[0134] In one embodiment of the present invention, the electricity usage behavior recognition model is pre-trained before being trained. During the pre-training process, the preprocessed second feature sequence is input into a pre-trained classifier. The classification space of the pre-trained classifier represents unauthorized power access, that is, the behavior of accessing power lines without the consent of the power company. The classification labels of the pre-trained classifier are manually labeled.

[0135] In one embodiment of the present invention, the loss function of the electricity usage behavior recognition model adopts a cross entropy loss function.

[0136] In one embodiment of the present invention, training samples of the electricity usage behavior recognition model are obtained by manual labeling.

[0137] In one embodiment of the present invention, different initial thresholds are set according to the probability values ​​of each electricity usage behavior output by the electricity usage behavior recognition model to trigger corresponding prevention and control measures, wherein the initial threshold of the probability of overload electricity usage is set to 0.2, the initial threshold of the probability of privately pulling and connecting wires is set to 0.1, the initial threshold of the probability of bypassing or modifying the metering device is set to 0.1, and the initial threshold of the probability of using inferior electrical equipment is set to 0.1. When the probability value output by the electricity usage behavior recognition model exceeds the initial threshold, the user is first notified by phone or text message to check the relevant equipment and remind them to pay attention to safe electricity use. If the behavior continues to occur within one day after the notification, staff are dispatched to the site for inspection. In addition, an emergency response mechanism is set. When the sum of the probability values ​​of other abnormal electricity usage behaviors other than normal behaviors exceeds 0.8, staff are immediately dispatched to the site for inspection and necessary measures are taken to ensure electricity safety.

[0138] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. An intelligent load-side power safety prevention and control system, characterized in that: include: A data collection module, which is used to collect power data of power usage scenarios within a preset time period T and at preset time intervals t, and construct a first feature sequence; The first characteristic sequence includes N sequence units, and the nth sequence unit represents the voltage, current, power and power consumption at the nth time point; The power consumption is represented by a time domain waveform, where the horizontal axis represents the time point and the vertical axis represents the amplitude; Electricity usage scenarios include: factory production, corporate office, household electricity, vehicle charging and temporary electricity use; A data preprocessing module, which is used to preprocess the first feature sequence to generate a second feature sequence; The second feature sequence includes N sequence units, and the nth sequence unit represents the feature data obtained after preprocessing at the nth time point; A power consumption behavior recognition model building module, which is used to build a power consumption behavior recognition model; The input of the electricity consumption behavior recognition model is the second feature sequence, and the output is the probability value of different electricity consumption behaviors; Electricity usage behaviors include: normal behavior, overloading, unauthorized wiring, bypassing or modifying metering devices, and using inferior electrical equipment; The prevention and control measures generation module is used to generate prevention and control measures for power usage scenarios based on the probability values ​​of different power usage behaviors.

2. According to claim 1, an intelligent load-side power safety prevention and control system is characterized in that: N, n, T and t are all custom parameters, 1≤n≤N, N=T / t.

3. The intelligent load-side power safety prevention and control system according to claim 1 is characterized in that: The preprocessing steps include: Step S201, for missing values ​​in the power data, filling processing is performed by taking the average value of the value of the previous sequence unit and the value of the next sequence unit; Step S202, normalizing the power data using a maximum and minimum normalization method; Step S203, creating a binary feature of "whether it is daytime"; Step S204, creating a binary feature of "whether it is the weekend"; Step S205, creating a binary feature of "whether it is a legal holiday"; Step S206, obtaining the maximum and minimum values ​​of electricity consumption and power every day and the average values ​​in the morning and afternoon, and creating a load characteristic; Step S207, using fast Fourier transform to convert the time domain waveform representation of the power consumption into a frequency domain waveform representation, the horizontal axis of the frequency domain waveform represents the frequency, and the vertical axis represents the amplitude, extracting the power spectrum density of the low-frequency component from the frequency domain waveform, and creating a periodic feature, specifically selecting the first K frequency values ​​from the frequency domain waveform, and calculating the power spectrum density corresponding to the first K frequency values; The formula for the number of frequency values ​​of a frequency domain waveform is: Among them, Fre N Represents the number of frequency values ​​of the frequency domain waveform, N is the number of sampling time points, The power spectral density formula is: P f =|X f | 2 ; Where P f represents the power spectral density, X f represents the complex value corresponding to frequency f, |X f | represents the modulus of a complex value. The formula for a complex value is: Where a and b represent the real part and imaginary part of the complex number a+bi obtained by fast Fourier transform of the time domain waveform respectively; Step S208, creating statistical features by calculating the mean, standard deviation and median of power consumption; The above preprocessed data are concatenated and represented by feature data.

4. The intelligent load-side power safety prevention and control system according to claim 1 is characterized in that: The electricity consumption behavior recognition model includes the segmentation layer, L high The first hidden layer, L low A second hidden layer, a concatenation layer, and a classifier; The segmentation layer is used to segment the second feature sequence into a third feature sequence and a fourth feature sequence, and the number of sequence units of the third feature sequence and the fourth feature sequence is the same as the number of sequence units of the second feature sequence; The segmentation layer first calculates the standard deviation of each sequence unit of the second feature sequence, and marks the sequence unit as 1 when the standard deviation is greater than or equal to the standard deviation threshold, and marks the remaining sequence units as 0 as the third feature sequence. When the standard deviation is less than the standard deviation threshold, the sequence unit is marked as 2, and the remaining sequence units are marked as 0 as the fourth feature sequence, where the standard deviation threshold is a custom parameter; That is, the sequence units of the third characteristic sequence only include unmarked sequence units and sequence units marked as 1; That is, the sequence units of the fourth characteristic sequence only include unmarked sequence units and sequence units marked as 2; For the first hidden layer and the second hidden layer number L high and L low ,satisfy: L high =min(NI,3); L low =min(NJ,2); Where I represents the number of sequence units marked as 1 in the third feature sequence, and J represents the number of sequence units marked as 2 in the fourth feature sequence; Each first hidden layer includes N first units, the nth first unit of the first first hidden layer inputs the combination vector of the nth sequence unit of the third feature sequence, and outputs the first hidden feature; No.1 high The first unit of the nth hidden layer inputs the lth high -1 The first hidden feature of the nth first unit output of the first hidden layer, output the first hidden feature, where 1≤l high ≤L high ; Each second hidden layer includes N second units, the nth second unit of the first second hidden layer inputs the combination vector of the nth sequence unit of the fourth feature sequence, and outputs the second hidden feature; No.1 low The nth second unit of the second hidden layer inputs the lth low -1 The second hidden feature output by the nth second unit of the second hidden layer, output the second hidden feature, where 1≤l low ≤L low ; The splicing layer is used to combine the L high The first hidden feature output by the Nth first unit of the first hidden layer is concatenated to the Lth low After the second hidden feature output by the Nth second unit of the second hidden layer is obtained, the third hidden feature is obtained; The third hidden feature is input into the classifier, and the classification space of the classifier represents the probability values ​​of different electricity usage behaviors.

5. The intelligent load-side power safety prevention and control system according to claim 4 is characterized in that: No.1 high The formula for the first hidden layer is: in and represents the nth sequence unit of the third feature sequence input by the nth first unit of the first hidden layer, and Respectively represent the first high The first hidden layer and the lth high-1 The first hidden feature output by the nth first unit of the first hidden layer, Indicates the first high The first hidden feature output by the n-1th first unit of the first hidden layer, and Respectively represent the first high The update gate, reset gate and candidate hidden features of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the update gate of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the reset gate of the nth first unit of the first hidden layer, and Respectively represent the first high The first weight parameter, the second weight parameter and the bias parameter of the candidate hidden state of the nth first unit of the first hidden layer, ⊙ represents point-by-point multiplication, Indicates the first high Whether the nth first unit of the first hidden layer is activated. When it returns 1, the nth first unit is activated. When it returns 0, the nth first unit is skipped. Indicates the first high The state of the nth first unit of the first hidden layer, which takes the value of 1 or 0, indicating whether it is marked or unmarked, sigmoid represents the sigmoid activation function, tanh represents the hyperbolic tangent function, concat represents the concatenation operation, and MLP represents the multi-layer perceptron; For the first low The formula for the second hidden layer is the same as above.

6. The intelligent load-side power safety prevention and control system according to claim 4 is characterized in that: The formulas for the concatenation layer and the classifier are: in and Respectively represent the L high The first hidden feature of the first unit output of the first hidden layer and the first hidden feature of the Lth unit low The second hidden feature output by the Jth second unit of the second hidden layer, p b Represents the probability value of different power consumption behaviors of the output, E p represents the third weight parameter, b p It represents the bias parameter of behavior recognition, concat represents the concatenation operation, and softmax represents the softmax activation function.

7. The intelligent load-side power safety prevention and control system according to claim 1 is characterized in that: Before training the electricity usage behavior recognition model, it is pre-trained. During the pre-training process, the preprocessed second feature sequence is input into the pre-trained classifier. The classification space of the pre-trained classifier represents unauthorized power access, that is, the behavior of accessing the power line without the consent of the power company. The classification labels of the pre-trained classifier are manually labeled.

8. The intelligent load-side power safety prevention and control system according to claim 1 is characterized in that: The loss function of the electricity consumption behavior recognition model adopts the cross entropy loss function.

9. The intelligent load-side power safety prevention and control system according to claim 1 is characterized in that: The training samples of the electricity consumption behavior recognition model are obtained through manual labeling.

10. The intelligent load-side power safety prevention and control system according to claim 1, characterized in that: According to the probability value of each electricity usage behavior output by the electricity usage behavior recognition model, different initial thresholds are set to trigger corresponding prevention and control measures. The initial threshold of the probability of overload electricity usage is set to 0.2, the initial threshold of the probability of private wiring is set to 0.1, the initial threshold of the probability of bypassing or modifying the metering device is set to 0.1, and the initial threshold of the probability of using inferior electrical equipment is set to 0.

1. When the probability value output by the electricity usage behavior recognition model exceeds the initial threshold, the user is first notified by phone or text message to check the relevant equipment and remind them to pay attention to safe electricity use. If the behavior continues to occur within one day after the notification, staff will be dispatched to the site for inspection. In addition, an emergency response mechanism is set up. When the sum of the probability values ​​of other abnormal electricity usage behaviors other than normal behaviors exceeds 0.8, staff will be immediately dispatched to the site for inspection and necessary measures will be taken to ensure electricity safety.