A method and system for optimizing neural network topology recognition based on a data acquisition terminal

Abstract

This invention relates to the field of power grid metering technology, and more specifically, to an optimized neural network topology identification method and system based on a data acquisition terminal. This invention uses a BP neural network model for topology identification. It preprocesses data recorded by the target acquisition terminal itself and data read from all undetermined meters by the target acquisition terminal to obtain a preliminary result of the actual topology. This preliminary result is then used as input to a trained BP neural network model, improving the speed and accuracy of the model output. This solves the problem that existing methods for identifying electricity topology based on ordinary neural networks only involve simple data preprocessing, resulting in only a limited improvement in the speed and accuracy of network calculations.

Description

Technical Field

[0001] This invention relates to the field of power grid metering technology, and more specifically, to an optimized neural network topology recognition method and system based on a data acquisition terminal. Background Technology

[0002] The current method for identifying the power topology of low-voltage distribution areas involves calculating the electrical connections between various power nodes and power areas, identifying these connections, and storing them as data. This has profound significance for subsequent energy-saving analysis, anomaly calculation, and fault diagnosis.

[0003] The power outage method combined with manual inspection is a relatively traditional topology identification method. However, analyzing single electricity consumption data has significant errors. With the popularization of power grid technology, more and more complex physical power grid topologies have emerged, such as shared distribution equipment and half-wave power extraction equipment. At the same time, traditional power grid topologies cannot achieve carrier communication. Therefore, under complex current conditions, existing electricity consumption topology identification methods based on simple information have significant limitations and inaccuracies. Although some methods use ordinary neural networks to process electricity consumption data, they are prone to getting stuck in extreme points, leading to misjudgments. Moreover, current applications of ordinary neural networks to process electricity consumption data mostly involve simple preprocessing operations such as data cleaning, which generally do not significantly improve the speed and accuracy of network computing. Summary of the Invention

[0004] Therefore, it is necessary to provide an optimized neural network topology identification method and system based on a data acquisition terminal to address the problem that existing methods for identifying power topology based on ordinary neural networks only involve simple data preprocessing, which does not significantly improve the speed and accuracy of network calculations.

[0005] This invention is achieved using the following technical solution:

[0006] In a first aspect, the present invention discloses an optimized neural network topology recognition method based on a data acquisition terminal, which is used to obtain the actual topology of target data acquisition terminals and target meters located in the same target area.

[0007] The optimized neural network topology recognition method based on the acquisition terminal includes the following steps:

[0008] Step 1: Obtain the operating condition feature information recorded by the target acquisition terminal itself and use it as Data 1; Obtain the operating condition feature information of all M undetermined meters read by the target acquisition terminal and use it as Data 2;

[0009] Step 2: Based on Data 1 and Data 2, preprocessing is performed to obtain preliminary results of the actual topology; the preprocessing methods include:

[0010] S2.1, Obtain data from N times at the same time in Data 1 and Data 2, and obtain the operating condition change trend at the target terminal and the operating condition change trend at M undetermined meters;

[0011] S2.2, compare the operating condition change trends of the M undetermined meters with the operating condition change trends of the target terminal, calculate the average normalization value, and determine the P target meters that are in the same target area as the target acquisition terminal.

[0012] S2.3, sample one by one from P target meters and compare the operating condition change trends with other target meters to calculate the average similarity and obtain the preliminary results of the actual topology;

[0013] Step 3: Input the preliminary results of the actual topology into the trained BP neural network model to obtain the actual topology structure.

[0014] This optimized neural network topology recognition method based on the acquisition terminal implements the method or process according to embodiments of this disclosure.

[0015] Secondly, the present invention discloses an optimized neural network topology recognition system based on a data acquisition terminal, which uses the optimized neural network topology recognition method based on a data acquisition terminal disclosed in the first aspect.

[0016] The optimized neural network topology recognition system based on the acquisition terminal includes: a data acquisition unit, a data preprocessing unit, and a topology recognition unit.

[0017] The data acquisition unit acquires the operating condition characteristic information recorded by the target acquisition terminal itself, which is used as Data 1; it also acquires the operating condition characteristic information of all M undetermined meters read by the target acquisition terminal, which is used as Data 2. The data preprocessing unit performs preprocessing based on Data 1 and Data 2 to obtain a preliminary result of the actual topology. The topology identification unit inputs the preliminary result of the actual topology into a trained BP neural network model to obtain the actual topology structure.

[0018] This optimized neural network topology recognition system based on a data acquisition terminal implements the method or process according to embodiments of this disclosure.

[0019] Thirdly, the present invention discloses a readable storage medium. The readable storage medium stores computer program instructions, which, when read and executed by a processor, perform the steps of the optimized neural network topology recognition method based on a data acquisition terminal disclosed in the first aspect.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] 1. This invention uses a BP neural network model for topology identification. It preprocesses data recorded by the target acquisition terminal and data from all undetermined meters read by the target acquisition terminal to obtain a preliminary result of the actual topology. This preliminary result is then used as input to a trained BP neural network model, improving the speed and accuracy of the model's output. Specifically, the preprocessing first compares the operating condition change trends at the target terminal and at the M undetermined meters, calculating the average normalized value to identify the target meters located in the same target area as the target acquisition terminal. Then, it samples each target meter and compares its operating condition change trends with other target meters, calculating the average similarity to obtain the preliminary result of the actual topology. This method yields a highly accurate preliminary result of the actual topology, facilitating the model to obtain a more accurate actual topology structure.

[0022] 2. In training the BP neural network model, this invention uses a genetic algorithm for optimization. Through crossover and mutation, the optimal weights and thresholds are selected to avoid the model getting trapped in local minima and to ensure the accuracy of the BP neural network model.

[0023] 3. This invention has a dedicated NPU for the BP neural network model, which is used to load the trained BP neural network model for calculation, thereby reducing the computational burden on the target acquisition terminal itself. Attached Figure Description

[0024] Figure 1 This is a flowchart of the optimized neural network topology recognition method based on the acquisition terminal in Embodiment 1 of the present invention;

[0025] Figure 2 This is a flowchart of training the BP neural network model in Embodiment 1 of the present invention;

[0026] Figure 3 This is a structural diagram of the optimized neural network topology recognition system based on the acquisition terminal in Embodiment 2 of the present invention. Detailed Implementation

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention.

[0028] It should be noted that when a component is said to be "installed on" another component, it can be directly on the other component or it may be in a component that is centered on it. When a component is said to be "set on" another component, it can be directly set on the other component or it may also be in a component that is centered on it. When a component is said to be "fixed to" another component, it can be directly fixed to the other component or it may also be in a component that is centered on it.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0030] Example 1

[0031] Please see Figure 1 , Figure 1 This is a flowchart of the optimized neural network topology recognition method based on the acquisition terminal in this invention.

[0032] The optimized neural network topology recognition method based on the acquisition terminal disclosed in Embodiment 1 is used to obtain the actual topology of the target acquisition terminal and the target meter located in the same target area.

[0033] The optimized neural network topology recognition method based on the acquisition terminal includes the following steps:

[0034] Step 1: Obtain the working condition feature information recorded by the target acquisition terminal itself and use it as data 1;

[0035] Obtain the operating condition characteristic information of all M undetermined meters read by the target acquisition terminal, and use it as data two.

[0036] In general, operating condition characteristic information includes voltage, current, power factor, time, etc. In this embodiment 1, the acquired operating condition characteristic information is set as voltage, current, power factor, and time.

[0037] Generally, the time span for Data 1 and Data 2 is one day, and the recording timestamps are set to several specific times, which can be adjusted according to actual needs. For example, if it is required to be copied once every hour, then it is set to start at 00:00 and collect data every 60 minutes (i.e., 01:00, 02:00, ..., 23:00).

[0038] Specifically, the target acquisition terminal has the functions of acquisition and measurement; the target acquisition terminal itself records the operating condition characteristic information at the target terminal as data one.

[0039] The target acquisition terminal reads all M undetermined electricity meters, which have metering and calculation functions. Each of the M undetermined electricity meters records the operating condition characteristics information at that meter location, which serves as data two.

[0040] Of course, to ensure consistent timing during subsequent preprocessing, a step 0 can be added before step 1—synchronizing the target acquisition terminal and all M pending electricity meters read by the target acquisition terminal in advance. Specifically, it can be set to synchronize the time of the target acquisition terminal and all M pending electricity meters read by the target acquisition terminal once a day to ensure that the timestamps of Data 1 and Data 2 are consistent.

[0041] Step 2: Based on Data 1 and Data 2, perform preprocessing to obtain preliminary results of the actual topology.

[0042] Specifically, preprocessing methods include:

[0043] S2.1, Obtain data from N times at the same time in Data 1 and Data 2, and obtain the operating condition change trend at the target terminal and the operating condition change trend at M undetermined meters;

[0044] The operating condition characteristic information recorded by the target acquisition terminal at the same time and the operating condition characteristic information read by the target acquisition terminal at all M undetermined meters are selected from Data 1 and Data 2 to form the operating condition change trend at the target terminal and the operating condition change trend at the M undetermined meters.

[0045] Among them, N identical times include times t1, t2, ..., t N For the nth time t among N identical time points... n The target acquisition terminal itself records the operating condition feature information as D. n The target acquisition terminal reads the operating condition characteristic information of all M undetermined meters as follows: Among them, the operating condition characteristic information of the m-th undetermined meter out of M undetermined meters is as follows:

[0046] It should be noted that the value of N is chosen based on the actual situation—the larger N is, the higher the accuracy of the result, but the greater the computational load; the smaller N is, the lower the accuracy of the result, but the smaller the computational load. Generally, N is set to 12, that is, the operating condition feature information recorded by the target acquisition terminal itself at 12 identical times of the day, and the operating condition feature information read by the target acquisition terminal from all M undetermined meters.

[0047] S2.2, compare the operating condition change trends of the M undetermined meters with the operating condition change trends of the target terminal, calculate the average normalized value, and determine the P target meters that are in the same target area as the target acquisition terminal.

[0048] For time t n ,calculate With D n The ratio of parameters of the same type is calculated, and the ratio is then weighted to obtain a normalized value.

[0049] Specifically, Includes the current value recorded by the m-th undetermined meter. voltage value Power factor D n Includes target acquisition and recording of current values voltage value Power factor

[0050] At time t n The corresponding normalized value is In the formula, α, β, and ε are weighting coefficients. In this Example 1, α is 0.4, β is 0.4, and ε is 0.2.

[0051] Iterate through N identical moments to obtain Calculate the average normalized value of the corresponding N normalized values;

[0052] If the average normalized value is greater than the preset value 1, it means that the m-th undetermined meter and the target acquisition terminal are in the same distribution area and are classified as the target meter; otherwise, the m-th undetermined meter and the target acquisition terminal are not in the same distribution area. The preset value 1 is generally taken as 0.8.

[0053] After the above process, P target meters were obtained that are located in the same target area as the target acquisition terminal.

[0054] S2.3, sample from each of the P target meters and compare the operating condition change trends with other target meters to calculate the average similarity and obtain the preliminary results of the actual topology.

[0055] For the nth time t among N identical times... n The operating condition characteristics of P meters are as follows:

[0056] Obtain the P target meters obtained in S2.2, select one target meter (i.e., the p1th target meter among the P target meters), and group it with the other target meters (i.e., the p2th target meter among the P target meters), and obtain the operating condition characteristic information of this group of meters at the same time. Where p1∈[1,P], p2∈[1,P], and p1≠p2.

[0057] For time t n The operating condition characteristic information of the p1th target meter is: The operating condition characteristics of the p2th target meter are as follows: calculate and The ratio of parameters of the same type is used to calculate the similarity by weighting the ratios.

[0058] Specifically, Includes the current value recorded by the p1th target meter. voltage value Power factor Includes the current value recorded by the p2th target meter. voltage value Power factor

[0059] At time t n The corresponding similarity is In the formula, α′, β′, and ε′ are weighting coefficients. In this Example 1, α′ is 1 / 3, β′ is 1 / 3, and ε′ is 1 / 3.

[0060] Iterate through N identical moments to obtain Calculate the average similarity among the corresponding N similarities;

[0061] If the average similarity is greater than the preset value 2, it means that the p1th target meter and the p2th target meter are in the same topology branch; otherwise, the p1th target meter and the p2th target meter are in different topology branches. The preset value 2 is generally set to 0.95.

[0062] Step 3: Input the preliminary results of the actual topology into the trained BP neural network model to obtain the actual topology structure.

[0063] The BP neural network model consists of three layers: an input layer, hidden layers, and an output layer. The input layer receives preliminary topological results from sample samples during model training and preliminary results from the actual topological results when the model is used after training. The hidden layers use Sigmoid as the activation function, and the output layer uses Purelin as the activation function.

[0064] For BP neural network models, they need to be pre-trained before they can be used.

[0065] Specifically, the training methods for BP neural network models include:

[0066] S3.1, Obtain the sample dataset and process it to obtain preliminary results of the sample topology.

[0067] The sample dataset consists of historical operating data of sample acquisition terminals and sample meters located within the same sample area. The sample topology within the sample area is known.

[0068] For historical operating condition data, the same principle as steps 1 and 2 is used: obtain data three recorded by the sample acquisition terminal itself; obtain data four from all sample meters read by the sample acquisition terminal; preprocess data three and data four to obtain preliminary sample topology results.

[0069] S3.2 Input the preliminary results of the sample topology into the BP neural network model to be trained to obtain the initial neural network weights and initial thresholds; encode the initial neural network weights and initial thresholds as the initial population.

[0070] S3.3, conduct multiple rounds of training, calculate the fitness value of individuals in the current population, select individuals with larger fitness values ​​for crossover and mutation, and obtain a new population; among them, the current population in the first round is the initial population.

[0071] See Figure 2 Specifically, if we define the current round as round i, then the current population is the population to be processed in round i. First, calculate the fitness value of each individual in the population to be processed in round i, and sort them in descending order. Then, copy the structure of individuals with higher fitness values ​​to the transitional population in round i.

[0072] In the first transitional population of round i, individuals are randomly paired up, and one pair is selected. Then, a crossover point is randomly set, and the encoding values ​​of the two individuals in the pair are swapped after the crossover point to obtain two new individual codes, which also form a new pair of individuals. This process is called crossover. All pairs of individuals in the first transitional population of round i are crossovered to form the second transitional population of round i.

[0073] Take one individual from the second transitional population in the i-th round, and randomly generate a mutation value. If the mutation value of this individual is smaller than the mutation value, then the encoded value is reversed. This process is called mutation processing. Mutation processing is performed on all individuals in the second transitional population in the i-th round to obtain the population after the i-th round of processing, which is the new population.

[0074] S3.4 Decode the new population to obtain new neural network weights and new thresholds, and substitute them into the BP neural network model to be trained, then test and compare with the sample topology.

[0075] If the error meets the requirements, the new neural network weights and the new threshold are taken as the optimal weights and the optimal threshold, and the BP neural network model with the optimal weights and the optimal threshold is taken as the trained BP neural network model.

[0076] If the error does not meet the requirements, the new population is used as the current population, and the process returns to step S3.3.

[0077] The error metric is the accuracy, which is the similarity between the test result and the sample topology. In this Example 1, to ensure effectiveness, the error is set to an accuracy higher than 90%.

[0078] In other words, if the current round is set as round i, and if the population processed in round i meets the requirements, then the population processed in round i is decoded to obtain new neural network weights and new thresholds, which are used as the optimal weights and optimal thresholds. The BP neural network model with the optimal weights and optimal thresholds is then used as the trained BP neural network model.

[0079] If the population error after the i-th round of processing does not meet the requirements, then the population after the i-th round of processing is used as the population to be processed in the (i+1)-th round, and S3.3 and S3.4 are repeated. This process is repeated through multiple rounds of iteration until a BP neural network model that meets the error requirements is obtained.

[0080] S3.2 to S3.4 actually apply the genetic algorithm to the training process of the BP neural network to select the optimal weights and the optimal threshold, avoid the model from getting trapped in local minima, and ensure the accuracy of the BP neural network model.

[0081] After obtaining the trained BP neural network model, the preliminary results of the actual topology can be used as input, and the actual topology structure can be output.

[0082] It should be noted that the preliminary results of the actual topology obtained through steps one and two already have a high degree of accuracy. Using the trained BP neural network model for further calculation can output a more accurate actual topology structure.

[0083] Example 2

[0084] This embodiment 2 discloses an optimized neural network topology recognition system based on a data acquisition terminal, which uses the optimized neural network topology recognition method based on a data acquisition terminal from embodiment 1.

[0085] The optimized neural network topology recognition system based on the acquisition terminal includes: a data acquisition unit, a data preprocessing unit, and a topology recognition unit.

[0086] The data acquisition unit acquires the operating condition characteristic information recorded by the target acquisition terminal itself, which is used as Data 1; it also acquires the operating condition characteristic information of all M undetermined meters read by the target acquisition terminal, which is used as Data 2. The data preprocessing unit performs preprocessing based on Data 1 and Data 2 to obtain a preliminary result of the actual topology. The topology identification unit inputs the preliminary result of the actual topology into a trained BP neural network model to obtain the actual topology structure.

[0087] See Figure 3In this second embodiment, the data acquisition unit is a built-in functional module of the target acquisition terminal. The target acquisition terminal has its own data acquisition function, and can naturally acquire data recorded by the target acquisition terminal itself and data read from all pending electricity meters. The data acquisition unit includes acquisition units for voltage, current, and power factor, among others.

[0088] The data preprocessing unit is also a built-in functional module of the target acquisition terminal. The data preprocessing unit has certain computing capabilities and can perform preprocessing based on Data 1 and Data 2 to obtain preliminary results of the actual topology.

[0089] The topology recognition unit employs an NPU with 1T computing power and incorporates a pre-trained BP neural network model. This BP neural network model can be trained on a PC and then fed into the NPU. During training on the PC, the initial topology results obtained from processing the sample dataset, along with the sample topology structure, are input for model training and optimization based on a genetic algorithm. Typically, a programmable data acquisition app is used to input the sample dataset.

[0090] After the NPU is fed with the trained BP neural network model, it receives the preliminary results of the actual topology and calculates and outputs the actual topology structure.

[0091] It should be noted that the NPU is set independently of the target acquisition terminal's CPU, reducing the computational load on the target acquisition terminal's CPU.

[0092] Example 3

[0093] This embodiment 3 discloses a readable storage medium storing computer program instructions. When the computer program instructions are read and executed by a processor, the steps of the optimized neural network topology recognition method based on the acquisition terminal in embodiment 1 are performed.

[0094] When applying the method of Example 1, it can be applied in the form of software, such as by designing it as a program that can run independently on a computer-readable storage medium. The computer-readable storage medium can be a USB flash drive, designed as a USB security token, and the program can be designed to start the entire method through an external trigger.

[0095] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0096] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for optimizing neural network topology recognition based on a data acquisition terminal, used to obtain the actual topology of target data acquisition terminals and target meters located in the same target area, characterized in that, The optimized neural network topology recognition method based on the acquisition terminal includes the following steps: Step 1: Obtain the operating condition feature information recorded by the target acquisition terminal itself, and use it as data one; obtain all data read by the target acquisition terminal. M The operating condition characteristics of each undetermined electricity meter are used as data two. Step 2: Based on Data 1 and Data 2, preprocessing is performed to obtain preliminary results of the actual topology; wherein the preprocessing method includes: S2.1, Obtain data from data one and data two. N Data from the same time point is used to obtain the trend of operating conditions at the target terminal. M Trends in operating conditions at each pending electricity meter; S2.2, will M The operating condition change trends at each pending electricity meter are compared with those at the target terminal. The average normalized value is calculated to determine which meter is located in the same target area as the target data acquisition terminal. P One target electricity meter; In S2.2, calculate and D n The ratio of parameters of the same type is calculated, and the ratio is then weighted to obtain a normalized value. in, for M The first of the pending electricity meters m One pending electricity meter N At the same time, the first n At time t n Operating condition characteristic information; D n The data recorded by the target acquisition terminal itself N At the same time, the first n At time t n Operating condition characteristic information; Iterate through N identical moments to obtain corresponding N Calculate the average normalized value; If the average normalized value is greater than the preset value one, it indicates that the first... m If the pending electricity meter is in the same area as the target data acquisition terminal, it will be classified as the target electricity meter; otherwise, the first... m The pending electricity meter and the target data acquisition terminal are not located in the same distribution area; S2.3, from P Each target meter is sampled individually, and its operating condition change trend is compared with that of other target meters. The average similarity is calculated to obtain the preliminary results of the actual topology. Step 3: Input the preliminary results of the actual topology into the trained BP neural network model to obtain the actual topology structure.

2. The optimized neural network topology recognition method based on a data acquisition terminal according to claim 1, characterized in that, Before step 1, there is also step 0, which includes the target acquisition terminal and all data read by the target acquisition terminal. M The time of each pending electricity meter is synchronized in advance.

3. The optimized neural network topology recognition method based on a data acquisition terminal according to claim 1, characterized in that, The operating condition characteristics include voltage, current, power factor, and time.

4. The optimized neural network topology recognition method based on a data acquisition terminal according to claim 1, characterized in that, In S2.3, from P Take the first target meter p One target meter, and P The first target meter p Two target meters are grouped together, and the operating characteristics of this group of meters at the same time are acquired; among them, , , ; calculate and The ratio of parameters of the same type is used to calculate the similarity by weighting the ratios. in, For the first p One target meter in N At the same time, the first n At time t n Operating condition characteristic information; For the first p Two target meters at N At the same time, the first n At time t n Operating condition characteristic information; Traversal N At the same time, we get , corresponding N 1. Calculate the average similarity based on the similarity scores. If the average similarity is greater than the preset value of 2, it indicates that the first... p 1 target meter, the first p The two target meters are in the same topology branch; otherwise, the first p 1 target meter, the first p The two target meters are located in different topology branches.

5. The optimized neural network topology recognition method based on a data acquisition terminal according to claim 1, characterized in that, The BP neural network model consists of three layers: an input layer, a hidden layer, and an output layer. The input layer inputs preliminary sample topology results during the model training phase and preliminary actual topology results when the model is used after training is completed. The hidden layer uses Sigmoid as the activation function, and the output layer uses Purelin as the activation function.

6. The optimized neural network topology recognition method based on a data acquisition terminal according to claim 5, characterized in that, The training method for the BP neural network model includes: S3.1, Obtain the sample dataset and process it to obtain preliminary results of the sample topology; S3.2 Input the preliminary results of the sample topology into the BP neural network model to be trained to obtain the initial neural network weights and initial thresholds; encode the initial neural network weights and initial thresholds as the initial population; S3.3 Calculate the fitness value of individuals in the current population, select individuals with larger fitness values ​​for crossover and mutation, and obtain a new population; S3.4 Decode the new population to obtain new neural network weights and new thresholds, and substitute them into the BP neural network model to be trained, then test and compare with the sample topology. If the error meets the requirements, the new neural network weights and the new threshold are taken as the optimal weights and the optimal threshold, and the BP neural network model with the optimal weights and the optimal threshold is taken as the trained BP neural network model. If the error does not meet the requirements, the new population will be used as the current population for the next round, and the process will return to step S3.

3.

7. An optimized neural network topology recognition system based on a data acquisition terminal, characterized in that, It uses the optimized neural network topology recognition method based on the acquisition terminal as described in any one of claims 1-6; The optimized neural network topology recognition system based on the acquisition terminal includes: The data acquisition unit is used to acquire the operating condition characteristic information recorded by the target acquisition terminal itself, and use it as data one; it is also used to acquire all data read by the target acquisition terminal. M The operating condition characteristics of each undetermined electricity meter are used as data two. A data preprocessing unit is used to preprocess data based on data one and data two to obtain preliminary results of the actual topology; and The topology identification unit is used to input the preliminary results of the actual topology into the trained BP neural network model to obtain the actual topology structure.

8. The optimized neural network topology recognition system based on a data acquisition terminal according to claim 7, characterized in that, The topology identification unit is an NPU.

9. A readable storage medium, characterized in that, The readable storage medium stores computer program instructions, which are read and executed by a processor to perform the steps of the optimized neural network topology recognition method based on the acquisition terminal as described in any one of claims 1-6.

Images