A method for identifying voltage levels of power equipment with dynamic adjustment of neural network levels

Abstract

The present invention proposes a method for identifying the voltage level of power equipment with dynamic adjustment of neural network levels. The present invention introduces multiple training samples, marks the real voltage level of each group of samples, and then introduces the trained BP neural network model as the initial BP neural network model, inputs each group of samples into the initial BP neural network model in turn to obtain the initial predicted voltage level of each sample, calculates the classification error index and the estimation error index according to the sample predicted voltage level and the real voltage level; judges whether the classification error index and the estimation error index meet the convergence; if the indicators converge, the number of model layers described in step 2 is kept unchanged; if the classification error index or the estimation error index does not meet the convergence, jumps to step 5; if the number of layers of the initial neural network model reaches the maximum value, the number of layers remains unchanged; otherwise, the number of layers is increased by one layer, the weight and bias of the newly added layer are adjusted, and jumps to step 2; the present invention improves the training speed and efficiency, and improves the warning level of live operation of power equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and in particular relates to a method for identifying voltage levels of power equipment by dynamically adjusting neural network hierarchies. Background Art

[0002] Live working of power equipment is an important part of power work. Due to the large number of mixed access of equipment of different voltage levels in the current power grid, workers mainly rely on existing anti-electric shock warning equipment for warning during live working. However, it is difficult for existing anti-electric shock warning equipment to actively adjust the warning level for lines of different voltage levels. "Mao Kai, Research on group-type high-voltage electric field measurement and safety distance alarm system, Chongqing University, 2014" proposed to develop a group-type high-voltage electric field measurement safety distance warning instrument based on wireless transmission, but did not consider the influence of different electric field strengths of different voltage levels. "Ding Yueming, New technology and prototype development of high-voltage on-site construction safety distance alarm, Huazhong University of Science and Technology, 2011" proposed a high-voltage on-site construction safety distance alarm technology and prototype development. This paper mainly focuses on a solution for construction machinery to get too close to high-voltage equipment and cause failure, and does not involve adaptive warning of different voltage levels. In addition, some scholars proposed sampling neural networks for voltage level classification. However, none of the above methods considers the need to quickly judge the voltage level requirements when live working on the project site. At the same time, the use of neural network algorithms for voltage level classification may not converge. Therefore, it is necessary to quickly and accurately identify equipment at different voltage levels, thereby laying the foundation for the power early warning equipment to adaptively adjust the warning thresholds and levels.

[0003] In response to this, this patent aims at the actual need of quickly finding safety warnings for live operations of power equipment with different voltage levels, and proposes a method for quickly identifying different voltage levels for live operations that takes into account the dynamic adjustment of the neural network hierarchy, so as to achieve fast and accurate identification of different voltage levels, and then adaptively adjust the warning for the power warning equipment, thereby improving the warning level of the power operation warning equipment. Summary of the invention

[0004] In view of the above defects or improvement needs of the prior art, and in view of the background of power equipment operating across voltage levels, the present invention provides a method for quickly identifying different voltage levels for live working that takes into account the dynamic adjustment of the neural network level. The method aims to comprehensively consider the influence of factors such as electric field strength, height above the ground, temperature and humidity, solve the problem that it is difficult to quickly adaptively warn equipment operating across voltage levels and the problem that the algorithm does not converge in some cases, achieve rapid warning and comprehensive judgment of the voltage level of power equipment, and improve the level of live working.

[0005] To achieve the above object, according to one aspect of the present invention, a method for identifying voltage levels of power equipment with dynamic adjustment of neural network hierarchical levels is provided, which specifically includes the following steps:

[0006] Step 1: Introduce multiple groups of training samples, manually mark the real voltage level of each group of training samples, introduce the BP neural network model, set the initial number of neural network layers, input each group of training samples into the BP neural network model in turn to predict the predicted voltage level of each group of training samples, build a loss function model based on the real voltage level of each group of training samples, optimize the training through the Adma algorithm to obtain the initial optimized neural network weights and the initial optimized neural network bias, and build the initial BP neural network model based on the initial optimized neural network weights, the initial optimized neural network bias and the initial number of neural network layers;

[0007] Step 2: Input each group of training samples into the initial BP neural network model in turn to predict the initial predicted voltage level of each group of training samples, calculate the classification error index according to the initial predicted voltage level of each group of training samples and the actual voltage level of each group of training samples, and calculate the estimation error index according to the initial predicted voltage level of each group of training samples and the actual voltage level of each group of training samples;

[0008] Step 3: Determine whether the classification error index meets the convergence and whether the estimation error index meets the convergence;

[0009] Step 4: If both the classification error index and the estimation error index converge, the number of layers of the initial neural network model described in step 2 remains unchanged; if the classification error index does not meet the convergence requirement or the estimation error index does not meet the convergence requirement, jump to step 5;

[0010] Step 5: If the number of layers of the initial neural network model reaches the maximum number of layers, the number of layers of the initial neural network model remains unchanged; otherwise, the number of layers of the initial neural network model increases by one layer, and the weight and bias of the newly added neural network layer are adjusted, and jump to step 2;

[0011] Preferably, each group of training samples in step 1 is specifically defined as:

[0012] X t =(B t , W t , S t , Q t , G t , D t ), t∈t1ttt

[0013] Among them, R represents the number of training samples, X t represents the tth group of training samples, B t represents the altitude of the t-th group of training samples, Wt represents the temperature of the tth group of training samples, S t represents the humidity of the t-th group of training samples, Q t represents the airflow of the tth group of training samples, G t represents the illumination of the t-th group of training samples, D t represents the electric field strength of the tth group of training samples;

[0014] The actual voltage level of each group of training samples in step 1 is specifically defined as:

[0015] Level t , t∈[1,R]

[0016] Among them, R represents the number of training samples, level t represents the true voltage level of the tth group of training samples;

[0017] Preferably, the classification error index calculated in step 2 is specifically defined as:

[0018]

[0019] Among them, H(ptq) i is the classification error index obtained from the i-th training. Where i represents the i-th training, x is the number of equipment voltage level types, and p(x) i is the expected value of the actual voltage value X in the i-th training is the predicted value of the xth class in the i-th training sample, E tti It represents the predicted expected value of the t-th sample in the i-th training, which is obtained by the constructed neural network model.

[0020] The estimation error index calculated in step 2 is specifically defined as:

[0021] The estimation error indicator model obtained from the i-th training As shown below:

[0022]

[0023] Among them, E tti represents the predicted expected value of the t-th sample in the i-th training; U i represents the actual expected value of the i-th training in this training sample, U i =∑p(X) i ; α∈t0.1t1t is the random sampling value in this training, R is the number of samples in this training, Q i Represents the classification accuracy obtained from the i-th training;

[0024]

[0025] Among them, the typical value of Q1 is generally 0.55, Q i ∈(Q1t1), l i is the number of neural network model layers trained for the i-th time, express The minimum value of

[0026] Preferably, the step 3 is to determine whether the classification error index meets the convergence requirement, as follows:

[0027] If the classification error index is less than the classification error index threshold, it means that the classification error index converges, otherwise the classification error index does not meet the convergence;

[0028] Step 3 determines whether the estimated error index meets the convergence requirement, as follows:

[0029] If the estimated error index is less than the estimated error index threshold, it means that the estimated error index converges, otherwise the estimated error index does not meet the convergence;

[0030] Preferably, the weights of the initial neural network model and the bias of the initial neural network model are adjusted in step 5 as follows:

[0031]

[0032]

[0033] in, and represents the weight and bias of the neural network model between layers i and j after the s+1th training. s represents the sth training; η represents the efficiency, which is a random value between (0,1); represents the back propagation error between the i-th layer and the j-th layer; Represents the output value of the i-th layer;

[0034] The advantage of the present invention is that it comprehensively considers the influence of factors such as height above the ground, electric field strength, temperature, humidity, etc., and especially considers the sample parameters obtained by using the aforementioned neural network training model, which saves a lot of training time and improves training speed and efficiency. At the same time, it avoids the problem of misjudgment of the voltage level classification results of the operating object caused by the maximum number of training times, which will effectively improve the early warning level of live operations of power equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 : is a flow chart of a method for identifying different voltage levels of live working of electric power equipment considering dynamic adjustment of levels provided by an embodiment of the present invention;

[0036] Figure 2 : It is a schematic diagram of the BP neural network model of the method for identifying different voltage levels of power equipment using the scheme taking into account the dynamic adjustment of the number of layers;

[0037] Figure 3 : is the correlation between the number of neural network layers and classification prediction accuracy;

[0038] Figure 4 : The relationship between the number of neural network iterations and classification prediction accuracy. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0040] The following is a detailed description of the method for identifying different voltage levels for live operations on power equipment taking into account the dynamic adjustment of levels provided by the present invention in conjunction with an embodiment. The embodiment comprehensively considers the influence of factors such as height above the ground, electric field strength, temperature, humidity, etc., and especially considers the sample parameters obtained by establishing the neural network training model as described above, thereby saving a large amount of training time, improving training speed and efficiency, and avoiding the problem of misjudgment of the voltage level classification results of the operating object caused by the maximum number of training times.

[0041] Combine the following Figures 1 to 4 The specific implementation of the present invention is a method for identifying the voltage level of power equipment by dynamically adjusting the neural network level. The method flow chart is as follows: Figure 1 As shown, the specific steps include:

[0042] Step 1: Introduce multiple groups of training samples, manually mark the real voltage level of each group of training samples, introduce the BP neural network model, set the initial number of neural network layers, input each group of training samples into the BP neural network model in turn to predict the predicted voltage level of each group of training samples, build a loss function model based on the real voltage level of each group of training samples, optimize the training through the Adma algorithm to obtain the initial optimized neural network weights and the initial optimized neural network bias, and build the initial BP neural network model based on the initial optimized neural network weights, the initial optimized neural network bias and the initial number of neural network layers;

[0043] Each group of training samples in step 1 is specifically defined as:

[0044] X t =(B t , W t, S t , Q t , G t , D t ), t∈t1ttt

[0045] Where R = 1024 represents the number of training samples; X t represents the t-th group of training samples; Bt = 2000 meters, represents the altitude of the t-th group of training samples; Wt = 25 ° C, represents the temperature of the t-th group of training samples; St = 45% rh, represents the humidity of the t-th group of training samples; Qt = 0.06m3 / h, represents the airflow of the t-th group of training samples; Gt = 60000~100000lx, represents the illumination of the t-th group of training samples; Dt = 4000v / m, represents the electric field strength of the t-th group of training samples;

[0046] The actual voltage level of each group of training samples in step 1 is specifically defined as:

[0047] Level t , t∈[1,R]

[0048] Among them, R represents the number of training samples, level t represents the true voltage level of the tth group of training samples;

[0049] Step 2: Input each group of training samples into the initial BP neural network model in turn to predict the initial predicted voltage level of each group of training samples, calculate the classification error index according to the initial predicted voltage level of each group of training samples and the actual voltage level of each group of training samples, and calculate the estimation error index according to the initial predicted voltage level of each group of training samples and the actual voltage level of each group of training samples;

[0050] The classification error index calculated in step 2 is specifically defined as:

[0051]

[0052] Among them, H(ptq) i is the classification error index obtained from the i-th training. Where i represents the i-th training, x is the number of equipment voltage level types, and p(x) i is the expected value of the actual voltage value X in the i-th training is the predicted value of the xth class in the i-th training sample, E tti It represents the predicted expected value of the t-th sample in the i-th training, which is obtained by the constructed neural network model.

[0053] The estimation error index calculated in step 2 is specifically defined as:

[0054] The estimation error indicator model obtained from the i-th training As shown below:

[0055]

[0056] Among them, E tti represents the predicted expected value of the t-th sample in the i-th training; U i represents the actual expected value of the i-th training in this training sample, U i =∑p(X) i ; α∈t0.1t1t is the random sampling value in this training, R is the number of samples in this training, Q i Represents the classification accuracy obtained from the i-th training;

[0057]

[0058] Among them, the typical value of Q1 is generally 0.55, Q i ∈(Q1t1), l i is the number of neural network model layers trained for the i-th time, express The minimum value of

[0059] Step 3: Determine whether the classification error index meets the convergence and whether the estimation error index meets the convergence;

[0060] The classification error index described in step 3 is determined to be convergent, as follows:

[0061] If the classification error index is less than the classification error index threshold, it means that the classification error index converges, otherwise the classification error index does not meet the convergence;

[0062] Step 3 determines whether the estimated error index meets the convergence requirement, as follows:

[0063] If the estimated error index is less than the estimated error index threshold, it means that the estimated error index converges, otherwise the estimated error index does not meet the convergence;

[0064] Step 4: If both the classification error index and the estimation error index converge, the number of layers of the initial neural network model described in step 2 remains unchanged; if the classification error index does not meet the convergence requirement or the estimation error index does not meet the convergence requirement, jump to step 5;

[0065] Step 5: If the number of layers of the initial neural network model reaches the maximum number of layers, the number of layers of the initial neural network model remains unchanged, and the maximum number of layers is 1000; otherwise, the number of layers of the initial neural network model increases by one layer, and the weight and bias of the newly added neural network layer are adjusted, and jump to step 2;

[0066] Step 5 adjusts the weights of the initial neural network model and the bias of the initial neural network model as follows:

[0067]

[0068]

[0069] in, and represents the weight and bias of the neural network model between layers i and j after the s+1th training. s represents the sth training; η represents the efficiency, which is a random value between (0,1); represents the back propagation error between the i-th layer and the j-th layer; Represents the output value of the i-th layer;

[0070] The method for identifying different voltage levels by dynamically adjusting the neural network layer proposed in the embodiment of the present invention comprehensively considers the influence of factors such as height above the ground, electric field strength, temperature, humidity, etc., and especially considers the sample parameters obtained by using the aforementioned neural network training model to save a lot of training time, improves training speed and efficiency, and avoids the problem of misjudgment of the voltage level classification results of the operating object caused by the maximum number of training times. It can provide a reference for voltage level warning of power equipment and improve the voltage level prediction level in live operating scenarios.

[0071] Figure 1 As shown, it is a flow chart of a method for identifying different voltage levels of live working of power equipment taking into account dynamic adjustment of levels. Figure 2 Schematic diagram of the BP neural network model constructed for identifying different voltage levels taking into account dynamic adjustment levels. Figure 3 is the correlation between the number of neural network layers and classification prediction accuracy, Figure 4 is the relationship between the number of neural network iterations and classification prediction accuracy. Figure 3 and Figure 4 It can be seen that the method for identifying different voltage levels for live working of power equipment taking into account the dynamic adjustment of the number of layers proposed in the present invention can actively adjust the number of neural network layers, thereby improving the training speed and efficiency, while avoiding the problem of misjudgment of the voltage level classification results of the working object caused by the maximum number of training times, thereby improving the early warning level of live working of power equipment.

[0072] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying voltage levels of power equipment with dynamic adjustment of neural network hierarchical levels, characterized in that: The following steps are involved: Step 1: Introduce multiple groups of training samples, manually mark the real voltage level of each group of training samples, introduce the BP neural network model, set the initial number of neural network layers, input each group of training samples into the BP neural network model in turn to predict the predicted voltage level of each group of training samples, build a loss function model based on the real voltage level of each group of training samples, optimize the training through the Adma algorithm to obtain the initial optimized neural network weights and the initial optimized neural network bias, and build the initial BP neural network model based on the initial optimized neural network weights, the initial optimized neural network bias and the initial number of neural network layers; Step 2: Input each group of training samples into the initial BP neural network model in turn to predict the initial predicted voltage level of each group of training samples, calculate the classification error index according to the initial predicted voltage level of each group of training samples and the actual voltage level of each group of training samples, and calculate the estimation error index according to the initial predicted voltage level of each group of training samples and the actual voltage level of each group of training samples; Step 3: Determine whether the classification error index meets the convergence and whether the estimation error index meets the convergence; Step 4: If both the classification error index and the estimation error index converge, the number of layers of the initial neural network model described in step 1 remains unchanged; if the classification error index does not meet the convergence requirement or the estimation error index does not meet the convergence requirement, jump to step 5; Step 5: If the number of layers of the initial neural network model reaches the maximum number of layers, the number of layers of the initial neural network model remains unchanged; otherwise, the number of layers of the initial neural network model increases by one layer, and the weight and bias of the newly added neural network layer are adjusted, and jump to step 2; Each group of training samples in step 1 is specifically defined as: X t =(B t ,W t 、S t 、Q t 、G t 、D t ),t∈[1,R] Among them, R represents the number of training samples, X t represents the tth group of training samples, B t represents the altitude of the t-th group of training samples, W t represents the temperature of the tth group of training samples, S t represents the humidity of the t-th group of training samples, Q t represents the airflow of the tth group of training samples, G t represents the illumination of the t-th group of training samples, D t represents the electric field strength of the tth group of training samples; The actual voltage level of each group of training samples in step 1 is specifically defined as: Level t， t∈[1,R] Among them, R represents the number of training samples, level t Represents the true voltage level of the tth group of training samples.

2. The method for identifying voltage levels of power equipment by dynamically adjusting the neural network level according to claim 1 is characterized in that: The following steps are involved: The classification error index calculated in step 2 is specifically defined as: Among them, H(p,q) i is the classification error index obtained from the i-th training; where i represents the i-th training, x is the number of equipment voltage level types, and p(x) i is the expected value of the actual voltage value X in the i-th training is the predicted value of the xth class in the i-th training sample, E t,i It represents the predicted expected value of the t-th sample in the i-th training, which is obtained by the constructed neural network model; The estimation error index calculated in step 2 is specifically defined as: The estimated error indicator model H obtained from the i-th training Ti , as shown below: Among them, E t,i represents the predicted expected value of the t-th sample in the i-th training; U i represents the actual expected value of the i-th training in this training sample, U i =∑p(X) i ; α∈[0.1,1] is the random sampling value in this training, R is the number of samples in this training, Q i Represents the classification accuracy obtained from the i-th training; Among them, the typical value of Q1 is 0.55, Q i ∈(Q1,1), l i is the number of neural network model layers trained for the i-th time, express The minimum value of .

3. The method for identifying voltage levels of power equipment by dynamically adjusting the neural network level according to claim 1 is characterized in that: The following steps are involved: The determination of whether the classification error index meets the convergence requirement in step 3 is as follows: If the classification error index is less than the classification error index threshold, it means that the classification error index converges, otherwise the classification error index does not meet the convergence; Step 3 determines whether the estimated error index meets the convergence requirement, as follows: If the estimated error index is less than the estimated error index threshold, it means that the estimated error index converges, otherwise the estimated error index does not meet the convergence.

4. The method for identifying voltage levels of power equipment by dynamically adjusting the neural network level according to claim 1 is characterized in that: The following steps are involved: Step 5 adjusts the weights and biases of the newly added neural network layer as follows: in, and represents the weight and bias of the neural network model between layers i and j after the s+1th training; s represents the sth training; η represents the efficiency, which is a random value between (0,1); represents the back propagation error between the i-th layer and the j-th layer; Represents the output value of the i-th layer.

Images