Electrical fire early warning method based on high and low frequency recurrent neural network

Abstract

The invention provides an electrical fire early warning method based on a high and low frequency recurrent neural network, and relates to the technical field of electrical fire prediction and household electricity safety. The early warning method comprises the following steps: training a recurrent neural network based on electrical parameter data to obtain a high and low frequency electrical parameter recurrent neural network; inputting the high-frequency electrical parameter data into a high-frequency electrical parameter recurrent neural network to obtain a high-frequency prediction result, inputting the low-frequency electrical parameter data into a low-frequency electrical parameter recurrent neural network to obtain a low-frequency prediction result, and obtaining a line prediction temperature based on the high-frequency prediction result and the low-frequency prediction result; and obtaining a temperature early-warning quantile based on the prediction error of the line prediction temperature and the probability distribution, monitoring the real temperature in real time based on the line prediction temperature, and comparing the temperature early-warning quantile to select whether to give an alarm or not. According to the early warning method provided by the invention, the relevance between the prediction result and the time sequence is enhanced, so that the early warning threshold can be dynamically adjusted.

Description

Technical Field

[0001] The present invention relates to the field of electrical fire prediction and household electricity safety technology, and in particular to an electrical fire early warning method based on high- and low-frequency recurrent neural networks. Background Art

[0002] Electrical fires account for the largest proportion of all fires, posing a huge threat to people's lives and property. The causes of electrical fires are complex and varied, including equipment aging, line failures, overload use and other factors. The occurrence process of electrical fires is often hidden and sudden, and the early characteristics are difficult to accurately capture and quantify, which increases the difficulty of fire prevention and control.

[0003] Traditional electrical fire warning systems mainly rely on fixed threshold judgments, that is, when the monitored electrical parameters such as current, power, temperature, etc. exceed the preset thresholds, an alarm is triggered. However, this method has obvious limitations. On the one hand, the fixed threshold cannot adapt to the dynamic characteristics of electrical parameters changing over time, which may cause high-risk states that are close to the threshold for a long time to be ignored; on the other hand, environmental factors or short-term equipment abnormalities may cause electrical parameters to briefly exceed the threshold, thereby triggering false alarms, increasing unnecessary panic and waste of resources. Therefore, it is urgent to provide a solution to improve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide an electrical fire early warning method based on high and low frequency recurrent neural networks, which improves the shortcomings of existing early warning methods that only judge line risks by comparing electrical parameters with fixed thresholds.

[0005] The present invention provides an electrical fire early warning method based on high and low frequency recurrent neural networks, which adopts the following technical solutions: When the line temperature is in a fault heating state, the current, power and temperature data in the line are collected based on a sampling frequency of 1s / time to obtain high-frequency electrical parameter data. Conversely, the current, power and temperature data in the line are collected based on a sampling frequency of 3min / time to obtain low-frequency electrical parameter data. Based on the high-frequency electrical parameter data, a recurrent neural network is trained to obtain a high-frequency electrical parameter recurrent neural network, and based on the low-frequency electrical parameter data, a recurrent neural network is trained to obtain a low-frequency electrical parameter recurrent neural network; Inputting high-frequency electrical parameter data into a high-frequency electrical parameter recurrent neural network to obtain a high-frequency prediction result, inputting low-frequency electrical parameter data into a low-frequency electrical parameter recurrent neural network to obtain a low-frequency prediction result, and obtaining a line prediction temperature based on the high-frequency prediction result and the low-frequency prediction result; The temperature warning quantile is obtained based on the prediction error and probability distribution of the line predicted temperature. The actual temperature is monitored in real time based on the line predicted temperature and compared with the temperature warning quantile to decide whether to alarm.

[0006] Optionally, the fault temperature rise state is that the actual temperature rise amplitude collected twice in succession exceeds a set percentage threshold.

[0007] Optionally, the percentage threshold is: when the temperature is between 20 degrees and 30 degrees, the percentage threshold is 30%; when the temperature is between 30 degrees and 40 degrees, the percentage threshold is 20%; when the temperature is between 40 degrees and 50 degrees, the percentage threshold is 15%; when the temperature exceeds 50 degrees, the percentage threshold is 10%.

[0008] Optionally, obtaining the line prediction temperature based on the high-frequency prediction result and the low-frequency prediction result includes: obtaining the line prediction temperature by weighted calculation based on the high-frequency prediction result and the low-frequency prediction result; The calculation formula is as follows: ; in, Predict the temperature for the line, is the high-frequency prediction result, is the weight of the high-frequency prediction results, is the low-frequency prediction result, is the weight of the low-frequency prediction results.

[0009] Optionally, the weight of the high-frequency prediction result and the weight of the low-frequency prediction result are calculated as follows: ; ; in, is the weight of the high-frequency prediction results, for The actual temperature value at the moment, for The maximum historical real temperature in k consecutive time steps before the moment, is the weight of the low-frequency prediction results.

[0010] Optionally, a temperature warning quantile is obtained based on the prediction error and probability distribution of the line predicted temperature, including: calculating the confidence level that the actual temperature falls within the interval based on the standard deviation of the prediction error of the line predicted temperature and the probability distribution, and dividing the warning quantile into 70% quantile, 80% quantile, 90% quantile, and 99% quantile; the probability distribution is a t distribution.

[0011] Optionally, the confidence level that the actual temperature falls within the interval is calculated based on the standard deviation of the prediction error of the line predicted temperature and the probability distribution. The calculation formula is as follows: ; in, Temperature warning quantiles, Predict the temperature for the line, for The critical value of the quantile, The standard deviation of the prediction error for the predicted temperature for the line.

[0012] Optionally, the standard deviation of the prediction error of the line predicted temperature is calculated as follows: m time windows of size 3 are set, the difference between the actual temperature value and the predicted temperature value at the 3 moments is calculated respectively, the squares of the obtained differences are summed, and the variance of the differences is obtained by dividing by m-1 time windows, and the square root of the variance is taken to obtain the standard variance of the prediction error of the line predicted temperature.

[0013] Optionally, choose whether to alarm based on the actual temperature and the temperature warning quantile, including: Sampling is performed at a sampling frequency of 3 minutes / time. When the actual temperature collected twice in a row exceeds the percentage threshold of the predicted temperature, sampling is performed at a sampling frequency of 1 second / time. When the actual temperature obtained by sampling is below the 70th percentile, the line fire safety factor is excellent; between the 70th and 80th percentiles, the line fire safety factor is good; between the 80th and 90th percentiles, the line fire safety factor is qualified; between the 90th and 99th percentiles, the line fire safety factor is poor; and an alarm is triggered when the actual temperature exceeds the 99th percentile.

[0014] The invention proposes an electrical fire early warning method based on high- and low-frequency recurrent neural networks, which has the following beneficial effects: 1. The present invention combines high- and low-frequency recurrent neural networks to predict line temperature, thereby enhancing the correlation between the prediction results and the time series, and enabling dynamic adjustment of the warning threshold, thus solving the problem of the traditional threshold being fixed; 2. The warning quantile calculated by the present invention can be used to classify the danger level according to the warning quantile of the line temperature, thereby improving the high-risk state of the line being close to the threshold for a long time, which is ignored in the prior art; 3. The recurrent neural network used in the present invention has powerful time series processing capabilities, can accurately capture the temporal relationship and dynamic changes between data, and improve the accuracy of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flow chart of an electrical fire warning method based on high- and low-frequency recurrent neural networks is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article 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.

[0017] The embodiment of the present invention provides an electrical fire warning method based on high and low frequency recurrent neural network, see Figure 1 ,include: S1. When the line temperature is in a fault heating state, the current, power and temperature data in the line are collected based on a sampling frequency of 1s / time to obtain high-frequency electrical parameter data. Conversely, the current, power and temperature data in the line are collected based on a sampling frequency of 3min / time to obtain low-frequency electrical parameter data. S2, training a recurrent neural network based on high-frequency electrical parameter data to obtain a high-frequency electrical parameter recurrent neural network, and training a recurrent neural network based on low-frequency electrical parameter data to obtain a low-frequency electrical parameter recurrent neural network; S3, inputting the high-frequency electrical parameter data into the high-frequency electrical parameter recurrent neural network to obtain a high-frequency prediction result, inputting the low-frequency electrical parameter data into the low-frequency electrical parameter recurrent neural network to obtain a low-frequency prediction result, and obtaining the line prediction temperature based on the high-frequency prediction result and the low-frequency prediction result; S4. Obtain the temperature warning quantile based on the prediction error and probability distribution of the line predicted temperature, monitor the actual temperature in real time based on the line predicted temperature and compare the temperature warning quantile to choose whether to alarm.

[0018] In some embodiments, the process of executing step S1 includes: S1.1. Collect high-frequency electrical parameters; S1.2. Collect low-frequency electrical parameters.

[0019] Specifically, when executing step S1.1, collecting high-frequency electrical parameters, it includes: when the line temperature is in a fault heating state, collecting current, power and temperature data in the line based on a sampling frequency of 1s / time to obtain high-frequency electrical parameter data.

[0020] Furthermore, the fault temperature rise state is that the actual temperature rise amplitude collected twice in succession exceeds a set percentage threshold.

[0021] In fact, the percentage threshold is: when the temperature is between 20 degrees and 30 degrees, the percentage threshold is 30%, when the temperature is between 30 degrees and 40 degrees, the percentage threshold is 20%, when the temperature is between 40 degrees and 50 degrees, the percentage threshold is 15%, and when the temperature exceeds 50 degrees, the percentage threshold is 10%.

[0022] Specifically, when executing step S1.2, collecting low-frequency electrical parameters, it includes: when the line temperature is in a non-fault heating state, collecting current, power and temperature data in the line based on a sampling frequency of 3 minutes / time to obtain low-frequency electrical parameter data.

[0023] In some embodiments, the process of executing step S2 includes: S2.1, training a recurrent neural network based on high-frequency electrical parameter data to obtain a high-frequency electrical parameter recurrent neural network; S2.2. Training a recurrent neural network based on low-frequency electrical parameter data to obtain a low-frequency electrical parameter recurrent neural network.

[0024] Specifically, when executing step S2.1, training a recurrent neural network based on high-frequency electrical parameter data to obtain a high-frequency electrical parameter recurrent neural network includes: sending the high-frequency current, power and temperature data collected at a sampling frequency of 1s / time as input features into the recurrent neural network, and training to obtain a high-frequency electrical parameter recurrent neural network.

[0025] In fact, the model trained by using high-frequency collected current, power and temperature electrical parameter data as input can explore the relationship between line temperature rise and electrical parameters, and be used to predict high-risk line states in advance.

[0026] Specifically, when executing step S2.2, training a recurrent neural network based on low-frequency electrical parameter data to obtain a low-frequency electrical parameter recurrent neural network includes: sending the low-frequency current, power and temperature data collected at a sampling frequency of 3 minutes / time as input features into the recurrent neural network, and training to obtain a low-frequency electrical parameter recurrent neural network.

[0027] In fact, the model trained by using low-frequency collected current, power and temperature electrical parameter data as input can monitor the usage status of long-term electrical appliances and predict the overall electricity consumption pattern.

[0028] In some embodiments, the process of executing step S3 includes: S3.1. inputting high-frequency electrical parameter data into a high-frequency electrical parameter recurrent neural network to obtain a high-frequency prediction result; S3.2, inputting the low-frequency electrical parameter data into the low-frequency electrical parameter recurrent neural network to obtain a low-frequency prediction result; S3.3. Obtain the line predicted temperature based on the high-frequency prediction results and the low-frequency prediction results.

[0029] Specifically, when executing step S3.1, the high-frequency electrical parameter data is input into the high-frequency electrical parameter recurrent neural network to obtain the high-frequency prediction result, including: inputting the high-frequency current, power and temperature data collected at a sampling frequency of 1s / time into the high-frequency electrical parameter recurrent neural network to obtain the high-frequency prediction result.

[0030] Specifically, when executing step S3.2, based on the low-frequency electrical parameter recurrent neural network, real-time monitoring of line electrical parameters and output of low-frequency prediction results are performed, it includes: inputting the low-frequency current, power and temperature data collected at a sampling frequency of 3 minutes / time into the low-frequency electrical parameter recurrent neural network to obtain the low-frequency prediction results.

[0031] Specifically, when executing step S3.3, obtaining the line predicted temperature based on the high-frequency prediction result and the low-frequency prediction result includes: The line predicted temperature is obtained by weighted calculation based on the high-frequency prediction results and the low-frequency prediction results; The calculation formula is as follows: ; in, Predict the temperature for the line, is the high-frequency prediction result, is the weight of the high-frequency prediction results, is the low-frequency prediction result, is the weight of the low-frequency prediction results.

[0032] Furthermore, the weights of the high-frequency prediction results and the low-frequency prediction results are calculated as follows: ; ; in, is the weight of the high-frequency prediction results, for The actual temperature value at the moment, for The maximum historical real temperature in k consecutive time steps before the moment, is the weight of the low-frequency prediction results.

[0033] In fact, when T increases, it means that the fire risk of the line also increases, and the high-frequency prediction results can more accurately predict the data before the disaster, so it is necessary to increase the weight corresponding to the high-frequency prediction results.

[0034] In some embodiments, the process of executing step S4 includes: S4.1. Obtain the temperature warning quantile based on the prediction error and probability distribution of the line predicted temperature; S4.2. Monitor the actual temperature in real time based on the line predicted temperature and compare it with the temperature warning quantile to choose whether to alarm.

[0035] Specifically, when executing step S4.1, obtaining the temperature warning quantile based on the prediction error and probability distribution of the line predicted temperature, it includes: calculating the confidence level that the actual temperature falls within the interval based on the standard deviation of the prediction error of the line predicted temperature and the probability distribution, and dividing the warning quantile into 70% quantile, 80% quantile, 90% quantile, and 99% quantile; the probability distribution is t distribution.

[0036] Furthermore, the confidence level that the actual temperature falls within the interval is calculated based on the standard deviation and probability distribution of the prediction error of the line predicted temperature. The calculation formula is as follows: ; in, Temperature warning quantile, Predict the temperature for the line, for The critical value of the quantile, is the standard deviation of the prediction error for the predicted temperature for the line.

[0037] Furthermore, the process of calculating the standard deviation of the prediction error of the line predicted temperature includes: setting m time windows of size 3 moments, respectively calculating the difference between the actual value and the predicted value at the three moments, summing the squares of the differences, and dividing by m-1 time windows to obtain the variance of the differences, taking the square root of the variance, and obtaining the standard deviation of the prediction error of the line predicted temperature.

[0038] Specifically, when executing step S4.2, selecting whether to alarm based on the actual temperature and the temperature warning quantile includes: S4.2.1. Real-time monitoring of actual temperature based on line predicted temperature; S4.2.2. Choose whether to alarm based on the actual temperature and temperature warning.

[0039] Specifically, when executing step S4.2.1, when the actual temperature is monitored in real time based on the line predicted temperature, sampling is performed at a sampling frequency of 3 minutes / time. When the actual temperature collected twice in succession exceeds the percentage threshold of the predicted temperature, sampling is performed at a sampling frequency of 1 second / time.

[0040] Furthermore, when the temperature is between 20 degrees and 30 degrees, and the actual temperature collected twice in a row exceeds 30% of the predicted temperature, it is regarded as a fault temperature rise; when the temperature is between 30 degrees and 40 degrees, and the actual temperature collected twice in a row exceeds 20% of the predicted temperature, it is regarded as a fault temperature rise; when the temperature is between 40 degrees and 50 degrees, and the actual temperature collected twice in a row exceeds 15% of the predicted temperature, it is regarded as a fault temperature rise; when the temperature exceeds 50 degrees, and the actual temperature collected twice in a row exceeds 10% of the predicted temperature, it is regarded as a fault temperature rise.

[0041] In fact, since the higher the line temperature, the more dangerous it is, and the higher the sampling frequency, the safer it is, a dynamic percentage is used as the threshold, and the percentage decreases as the temperature increases, so as to track the actual line temperature more effectively.

[0042] Specifically, when executing step S4.2.2, the choice of whether to alarm is based on the actual temperature and the temperature warning, including: when the actual temperature obtained by sampling is below the 70th percentile, the line fire safety factor is excellent; between the 70th and 80th percentiles, the line fire safety factor is good; between the 80th and 90th percentiles, the line fire safety factor is qualified; between the 90th and 99th percentiles, the line fire safety factor is poor; and when it exceeds the 99th percentile, an alarm is triggered.

[0043] In fact, during the execution of the entire solution, the real-time monitored line electrical parameters will be logged, and the high-frequency electrical parameter recurrent neural network and the low-frequency electrical parameter recurrent neural network will be regularly updated based on the line electrical parameters in the log records, and the corresponding warning quantiles will be adjusted, so that the entire solution is more suitable for the applied electrical lines.

[0044] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. An electrical fire early warning method based on high and low frequency recurrent neural networks, characterized in that: include: When the line temperature is in a fault heating state, the current, power and temperature data in the line are collected based on a sampling frequency of 1s / time to obtain high-frequency electrical parameter data. Conversely, the current, power and temperature data in the line are collected based on a sampling frequency of 3min / time to obtain low-frequency electrical parameter data. Based on the high-frequency electrical parameter data, a recurrent neural network is trained to obtain a high-frequency electrical parameter recurrent neural network, and based on the low-frequency electrical parameter data, a recurrent neural network is trained to obtain a low-frequency electrical parameter recurrent neural network; Inputting high-frequency electrical parameter data into a high-frequency electrical parameter recurrent neural network to obtain a high-frequency prediction result, inputting low-frequency electrical parameter data into a low-frequency electrical parameter recurrent neural network to obtain a low-frequency prediction result, and obtaining a line prediction temperature based on the high-frequency prediction result and the low-frequency prediction result; The temperature warning quantile is obtained based on the prediction error and probability distribution of the line predicted temperature. The actual temperature is monitored in real time based on the line predicted temperature and compared with the temperature warning quantile to decide whether to alarm.

2. The electrical fire early warning method based on high and low frequency recurrent neural network according to claim 1 is characterized in that: The fault temperature rise state is that the actual temperature rise amplitude collected twice in succession exceeds the set percentage threshold.

3. The electrical fire early warning method based on high and low frequency recurrent neural network according to claim 2 is characterized in that: The percentage threshold is: when the temperature is between 20 degrees and 30 degrees, the percentage threshold is 30%, when the temperature is between 30 degrees and 40 degrees, the percentage threshold is 20%, when the temperature is between 40 degrees and 50 degrees, the percentage threshold is 15%, and when the temperature exceeds 50 degrees, the percentage threshold is 10%.

4. The electrical fire early warning method based on high and low frequency recurrent neural network according to claim 1 is characterized in that: The line predicted temperature is obtained based on the high-frequency prediction results and the low-frequency prediction results, including: The line predicted temperature is obtained by weighted calculation based on the high-frequency prediction results and the low-frequency prediction results; The calculation formula is as follows: ； in, Predict the temperature for the line, is the high-frequency prediction result, is the weight of the high-frequency prediction results, is the low-frequency prediction result, is the weight of the low-frequency prediction results.

5. The electrical fire early warning method based on high and low frequency recurrent neural network according to claim 4 is characterized in that: The weight of the high-frequency prediction result and the weight of the low-frequency prediction result are calculated as follows: ； ； in, is the weight of the high-frequency prediction results, for The actual temperature value at the moment, for The maximum historical real temperature in k consecutive time steps before the moment, is the weight of the low-frequency prediction results.

6. The electrical fire early warning method based on high and low frequency recurrent neural network according to claim 1, characterized in that: The temperature warning quantile is obtained based on the prediction error and probability distribution of the line predicted temperature, including: The confidence level that the actual temperature falls within the interval is calculated based on the standard deviation of the prediction error of the line predicted temperature and the probability distribution, and the warning quantile is divided into 70% quantile, 80% quantile, 90% quantile, and 99% quantile; the probability distribution is t distribution.

7. The electrical fire early warning method based on high and low frequency recurrent neural network according to claim 6 is characterized in that: The confidence level that the actual temperature falls within the interval is calculated based on the standard deviation of the prediction error of the line predicted temperature and the probability distribution. The calculation formula is as follows: ； in, Temperature warning quantiles, Predict the temperature for the line, for The critical value of the quantile, The standard deviation of the prediction error for the predicted temperature for the line.

8. The electrical fire early warning method based on high and low frequency recurrent neural network according to claim 7 is characterized in that: The standard deviation of the prediction error of the line predicted temperature is calculated as: Set m time windows of size 3 moments, calculate the difference between the actual temperature value and the predicted temperature value at the three moments respectively, sum the squares of the obtained differences, and divide by the number of m-1 time windows to obtain the variance of the differences, take the square root of the variance, and obtain the standard deviation of the prediction error of the line predicted temperature.

9. The electrical fire early warning method based on high and low frequency recurrent neural network according to claim 6, characterized in that: Based on the line predicted temperature, the actual temperature is monitored in real time and compared with the temperature warning quantile to select whether to alarm, including: Sampling is performed at a sampling frequency of 3 minutes / time. When the actual temperature collected twice in a row exceeds the percentage threshold of the predicted temperature, sampling is performed at a sampling frequency of 1 second / time. When the actual temperature obtained by sampling is below the 70th percentile, the line fire safety factor is excellent; between the 70th and 80th percentiles, the line fire safety factor is good; between the 80th and 90th percentiles, the line fire safety factor is qualified; between the 90th and 99th percentiles, the line fire safety factor is poor; and an alarm is triggered when the actual temperature exceeds the 99th percentile.

Images