Online prediction method and apparatus for metering error of electronic current transformer

Abstract

An online prediction method and apparatus for a metering error of an electronic current transformer. The method comprises: acquiring environmental data of an environment where an electronic current transformer is located (S101); and inputting target environmental data into a pre-trained target walk penetration neural network, and acquiring error prediction results outputted by a target error prediction model on the basis of the target environmental data, wherein the error prediction results comprise a ratio error prediction result and a phase error prediction result (S102). The method realizes online automatic error prediction for an electronic current transformer, thereby improving the efficiency of error determination for the electronic current transformer, and thus improving the calibration efficiency of the electronic current transformer. In addition, a bearing layer structure in the target walk penetration neural network can effectively improve the overall and local stability of the walk penetration neural network, thereby improving the error prediction accuracy of the electronic current transformer.

Description

An online prediction method and device for measurement error of electronic current transformer TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to an online prediction method and device for measurement error of electronic current transformer. BACKGROUND

[0002] Under the promotion of the digital transformation of the national power network, electronic current transformers, as one of the main digital intelligent measurement devices, are widely used in digital intelligent substations due to their simple insulation structure, wide measurement frequency band, large dynamic range and other advantages. The digital interface of the electronic current transformer is more conducive to realizing multi-mode data fusion sharing and meets the needs of the development of the digital substation. As the core equipment for collecting current signal data in the next generation of substations, the accuracy of its measurement is crucial for electric energy measurement.

[0003] Therefore, in actual application, it is necessary to calibrate electric energy measurement devices regularly. In related technologies, the error of the measured measurement device is determined by comparing the deviation between the measured measurement device and the high-accuracy standard device, so as to ensure the measurement performance accuracy and safety and reliability of all measurement devices. However, in this way, the connected current transformer will directly affect the calibration accuracy, and the current transformer with high accuracy requirement has high dependence.

[0004] Therefore, in the current power system, the power current transformer is periodically calibrated offline, but the offline calibration method has complex field wiring, high calibration difficulty and low work efficiency. In addition, compared with the past electromagnetic current transformer, the electronic current transformer is seriously affected by various environmental factors, and has low reliability and stability, so its calibration period is often shorter. Limited by the requirements of safe and economic operation of the power grid, it is difficult to arrange a short-period power-off plan, and the offline calibration work is further difficult to perform, so it is urgent to research a new online calibration method for electronic current transformer. SUMMARY

[0005] Therefore, the embodiments of the present application provide an online prediction method and device for measurement error of electronic current transformer to improve the calibration efficiency of the electronic current transformer.

[0006] According to an aspect of the present application, an online prediction method for measurement error of electronic current transformer is provided, and the method comprises:

[0007] Obtaining target environment data of an environment in which a target electronic current transformer is located, the target environment data comprising temperature data, humidity data, magnetic field data and corrosion data;

[0008] input the target environment data into a pre-trained target moving-penetration neural network, and obtain an error prediction result output by the target error prediction model based on the target environment data, the error prediction result including a ratio error prediction result and a phase error prediction result;

[0009] The target moving-penetration neural network is pre-trained through the following steps:

[0010] obtain an environment data set, each piece of environment data in the environment data set and an error label of an electronic current transformer corresponding to each piece of environment data;

[0011] input each piece of environment data in the environment data set into an initial moving-penetration neural network, and obtain a predicted error output by the initial moving-penetration neural network; wherein the initial moving-penetration neural network includes an input layer, a latent layer, a bearing layer and an output layer, the bearing layer is used to receive an output of the latent layer and store the output of the latent layer in a previous iteration;

[0012] train parameters in the initial moving-penetration neural network based on a difference between the predicted error and an error label of corresponding environment data until the difference converges, and obtain a target moving-penetration neural network; wherein the parameters include inter-layer connection weights between network layers and biases in each network layer.

[0013] In a possible embodiment, the environment data set is obtained by:

[0014] obtain electronic current transformer running state data and corresponding environment data collected historically, the running state data including a ratio difference and a phase difference of the electronic current transformer;

[0015] filter error values and outlier values of the electronic current transformer running state data according to a preset filtering rule, obtain filtered running state data and corresponding environment data, and constitute an environment data set, wherein the error values are running state data that appear continuously more than a preset number of times, and the outlier values are running state data that have a difference greater than a preset state difference threshold from other running state data.

[0016] In a possible embodiment, the initial moving-penetration neural network includes a plurality of latent layers and bearing layers, each latent layer and each bearing layer corresponding to each other, and the training of the parameters in the initial moving-penetration neural network based on the difference between the predicted error and the error label of the corresponding environment data includes:

[0017] calculate a quantity range of the latent layers based on a node number of the input layer, a node number of the output layer, and a preset node disturbance number, wherein the node number of the input layer is a data type number input to the input layer, the node number of the output layer is a preset data type number output by the output layer, and the preset node disturbance number is a constant in a preset value range;

[0018] obtain a latent layer candidate number based on the quantity range of the latent layers, the candidate number being each integer value in the quantity range;

[0019] calculate an error bias value based on the prediction error and an error label of corresponding environment data for each latent layer candidate number, and determine a latent layer candidate number corresponding to a maximum error bias value in the error bias values corresponding to the latent layer candidate numbers as the target quantity of the latent layers and the bearing layers.

[0020] In a possible embodiment, the training of the parameters in the initial mobile penetration neural network based on the difference between the prediction error and the error label of the corresponding environment data further includes:

[0021] set the inter-layer connection weight and bias particles and the spatial dimension, and determine the optimal value of the inter-layer connection weight and bias by using a particle swarm algorithm, wherein the spatial dimension and the total number of particles of the inter-layer connection weight and bias particles are determined by the following formula: N (t)=n h (t)(r1n in (t)+r2n out (t)+n h (t)+1)+r2n out (t)

[0022] wherein S N (t) is the spatial dimension, n h (t) is the target quantity of the latent layers, n in (t) is the node number of the input layer, n out (t) is the node number of the output layer, and r1 and r2 are random numbers; and P(t) is the total number of particles.

[0023] In a possible embodiment, the determination of the optimal value of the inter-layer connection weight and bias by using the particle swarm algorithm includes:

[0024] generate a plurality of particles in a space corresponding to the spatial dimension, the total number of the plurality of particles being the particle;

[0025] update the speed and position of each particle according to the following formula in each iteration: p ij (t+1)=p ij (t)+v ij (t+1)

[0026] Where i is the number of particles, j is the dimension of the space, and v ij (t) represents the particle velocity after t iterations, p ij (t) represents the position of the particle after t iterations. Let j be the j-th dimension value of the extremum of the i-th particle. denoted as the j-th dimension value of the overall optimal position, r1, r2, and r3 are random numbers on [0,1]; c1 is the individual practice element, c2 is the social practice element, w is the trend weight, and δ is the random influence element;

[0027] The trend weight is iterated using the following formula:

[0028] Where w(t) is the follow-the-momentum weight after the t-th iteration, w max To preset the maximum value of the trend-following weight, w min To preset the minimum value of the trend-following weight, N c (t) represents the current iteration step, N. m σ(t) represents the preset maximum number of iterations, and σ(t) represents the random perturbation value.

[0029] The individual practice elements and social practice elements are iterated using the following formula:

[0030] Where c1(t) is the value of the individual practice element after t iterations, c max (t) represents the maximum value of the preset practice element, and σ1(t) and σ2(t) are both perturbation elements;

[0031] After each update of the velocity and position of each particle, the target loss function value is calculated based on the current velocity and position of the particle. If the minimum value of the target loss function is not reached, the step of updating the velocity and position of each particle according to the following formula in each iteration is returned until the minimum value of the target loss function is obtained.

[0032] In one possible embodiment, the step of inputting the target environment data into a pre-trained target swimming penetration neural network to obtain the error prediction result output by the target error prediction model based on the target environment data includes:

[0033] The target-moving, penetrating neural network outputs the error prediction result according to the following function: y(k,t)=g 1(ω3(u(k,t)))+σ(k,t) u(k,t)=g 2 (ω1u c (k,t)+ω2x(k-1,t))+σ(k,t) u c (k,t)=u(k-1,t)

[0034] Where y(k,t) is the output of the target mobile penetration neural network, x(k-1,t) is the target environment data, u(k,t) is the output of the latent layer, and u c (k,t) represents the output from the latent layer to the carrier layer, ω1 and ω2 are the inter-layer connection weights, k is the iteration number, and g 1 () is the activation function of the neurons in the output layer, g 2 () is the neuron activation function of the latent layer, and σ(k,t) is the perturbation function.

[0035] In one possible embodiment, the method further includes:

[0036] The target mobile penetration neural network is evaluated using a preset validation set according to preset statistical indicators. These preset statistical indicators include the absolute error of the stochastic mean field technique (Ea) and the mean relative error (Er). The absolute error of the stochastic mean field technique is calculated using the following formula:

[0037] Among them, c i To pre-determine the standard coefficients, y i For the error labels corresponding to the environmental data, e i The prediction error value is given by n, which is the amount of environmental data input into the target swimming penetration neural network.

[0038] The average relative error is calculated using the following formula:

[0039] If the preset statistical index is lower than the preset index threshold, it is determined that the target swimming through the neural network has completed training.

[0040] According to another aspect of the present invention, an online prediction device for metering errors of electronic current transformers is provided, the device comprising:

[0041] The acquisition module is used to acquire target environmental data of the environment in which the target electronic current transformer is located. The target environmental data includes temperature data, humidity data, magnetic field data, and corrosion data.

[0042] The input module is used to input the target environment data into a pre-trained target swimming penetration neural network to obtain the error prediction results output by the target error prediction model based on the target environment data. The error prediction results include ratio error prediction results and phase error prediction results.

[0043] The target-moving, penetrating neural network is pre-trained through the following steps:

[0044] Obtain an environmental dataset, including each piece of environmental data and the error label of the electronic current transformer corresponding to each piece of environmental data;

[0045] Each piece of environmental data in the environmental dataset is input into an initial floating-through neural network to obtain the prediction error output by the initial floating-through neural network; wherein, the initial floating-through neural network includes an input layer, a latent layer, a carrier layer and an output layer, and the carrier layer is used to receive the output of the latent layer and store the output of the latent layer in the previous iteration;

[0046] The parameters in the initial floating-through neural network are trained based on the difference between the prediction error and the corresponding error label of the environmental data until the difference converges, thus obtaining the target floating-through neural network; wherein, the parameters include the inter-layer connection weights between each network layer and the bias in each network layer.

[0047] According to another aspect of the present invention, an electronic device is provided, comprising:

[0048] Processor; and

[0049] Stored program memory,

[0050] The program includes instructions that, when executed by the processor, cause the processor to perform any of the above-described online prediction methods for metering errors of electronic current transformers.

[0051] According to another aspect of the present invention, a non-transient computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute any of the above-described online prediction methods for metering errors of electronic current transformers.

[0052] The present invention provides one or more technical solutions that, by acquiring environmental data of the environment in which the electronic current transformer is located and using a neural network to automatically predict the error data of the electronic current transformer, achieve online automatic prediction of the electronic current transformer error, improve the efficiency of electronic current transformer error determination, and thus improve the calibration efficiency of the electronic current transformer. Furthermore, the target moving-through neural network includes a carrier layer, which stores the output of the hidden layer in the previous iteration. Compared with traditional neural networks, this carrier layer structure can effectively increase the overall and local stability of the moving-through neural network, thereby improving the error prediction accuracy of the electronic current transformer. Attached Figure Description

[0053] Further details, features, and advantages of the invention are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0054] Figure 1 is a schematic flowchart of an online prediction method for metering errors of electronic current transformers provided in an embodiment of the present invention;

[0055] Figure 2 is a flowchart illustrating the target moving through-neural network training process in the online prediction method for metering error of electronic current transformers provided in an embodiment of the present invention.

[0056] Figure 3 is a schematic diagram of the architecture of a network platform for implementing the online prediction method for metering errors of electronic current transformers provided in the embodiments of the present invention.

[0057] Figure 4 is a schematic diagram of a target-moving, penetrating neural network provided in an embodiment of the present invention;

[0058] Figure 5 is a schematic diagram of an online prediction device for metering error of electronic current transformer provided in an embodiment of the present invention;

[0059] Figure 6 shows a structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present invention. Detailed Implementation

[0060] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0061] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0062] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0063] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0064] The names of the messages or information exchanged between the multiple devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0065] An electronic current transformer is a device consisting of one or more voltage or current sensors connected to a transmission system and a secondary converter. It transmits a quantity proportional to the measured quantity to measuring instruments, meters, and relay protection or control devices. Due to environmental influences, electronic current transformers may exhibit measurement errors. Since power systems cannot be offline for extended periods, electronic current transformers need to be calibrated online by measuring and recalibrating these errors.

[0066] To improve the calibration efficiency of electronic current transformers, this invention provides an online prediction method and apparatus for electronic current transformer measurement errors. The online prediction method for electronic current transformer measurement errors provided by this invention can be applied to any electronic device equipped with online prediction functionality for electronic current transformer measurement errors, such as a computer, server, or mobile terminal, etc. The following description of the invention is based on the accompanying drawings:

[0067] Figure 1 is a schematic flowchart of an online prediction method for metering errors of electronic current transformers provided in an embodiment of the present invention, which may include the following steps:

[0068] S101. Obtain target environmental data of the environment where the target electronic current transformer is located. The target environmental data includes temperature data, humidity data, magnetic field data, and corrosion data.

[0069] S102. Input the target environment data into a pre-trained target swimming penetration neural network to obtain the error prediction result output by the target error prediction model based on the target environment data. The error prediction result includes the ratio error prediction result and the phase error prediction result.

[0070] In one possible embodiment, as shown in Figure 2, the above-mentioned target-traversing neural network can be pre-trained through the following steps:

[0071] S201. Obtain an environmental dataset, including each piece of environmental data and the error label of the electronic current transformer corresponding to each piece of environmental data.

[0072] S202. Input each piece of environmental data in the environmental dataset into the initial floating-through neural network and obtain the prediction error output by the initial floating-through neural network; wherein, the initial floating-through neural network includes an input layer, a latent layer, a carrier layer and an output layer, and the carrier layer is used to receive the output of the latent layer and store the output of the latent layer in the previous iteration;

[0073] S203. The parameters in the initial floating-through neural network are trained based on the difference between the prediction error and the error label of the corresponding environmental data until the difference converges, thereby obtaining the target floating-through neural network; wherein, the parameters include the inter-layer connection weights between each network layer and the bias in each network layer.

[0074] By applying the embodiments of this invention, environmental data of the environment in which the electronic current transformer is located is acquired, and error data of the electronic current transformer is obtained through automatic prediction using a neural network. This achieves online automatic prediction of the electronic current transformer error, improving the efficiency of error determination and thus enhancing the calibration efficiency of the electronic current transformer. Furthermore, the target moving-through neural network includes a carrier layer, which stores the output of the hidden layer in the previous iteration. Compared to traditional neural networks, this carrier layer structure effectively increases the overall and local stability of the moving-through neural network, thereby improving the accuracy of error prediction for the electronic current transformer.

[0075] In this invention, the target electronic current transformer can be any electronic current transformer. In one possible embodiment, a time counter can be set when each electronic current transformer is activated. This time counter is used to monitor the usage time of the electronic current transformer, and electronic current transformers whose usage time exceeds a preset time threshold can be used as target electronic current transformers for error prediction. In another possible embodiment, error prediction can also be performed on each electronic current transformer in the power system according to a preset time period. The aforementioned time threshold and time period can be set according to the actual application scenario, and this invention does not impose specific limitations on them.

[0076] When predicting errors in a target electronic current transformer, the target environmental data of the environment in which the target electronic current transformer is located can be obtained first. This target environmental data may include temperature data, humidity data, magnetic field data, and corrosion data, and may also include data on vibration, electric field load, frequency, harmonics, and communication and safety aspects.

[0077] The aforementioned target environmental data can be acquired through various sensors. For example, temperature data can be acquired through a temperature sensor, which includes the ambient temperature and the corresponding acquisition timestamp; humidity data can also be acquired through a humidity sensor, which includes the ambient humidity and the corresponding acquisition timestamp. The aforementioned magnetic field data can include magnetic field information generated by various electrical devices in the environment where the target electronic current transformer is located, as well as information about the Earth's magnetic field, including magnetic field strength and direction. The aforementioned corrosion data can include various factors that may cause corrosion to the electronic current transformer, such as temperature, humidity, pollutant concentration, and wind speed. For example, the aforementioned corrosion data can be acquired through a corrosion sensor that can measure parameters such as corrosion potential and current density.

[0078] The aforementioned sensors can be installed inside the target electronic current transformer or at a location close to it, such as 1 meter away, to obtain target environmental data about the environment in which the target electronic current transformer is located as accurately as possible. In one possible embodiment, the target environmental data may include environmental data collected within a preset time period, which can be set according to the actual application scenario, such as 1 day, 3 days, etc.

[0079] After obtaining the target environment data, the target environment data can be input into a pre-trained target mobile penetration neural network. In one possible embodiment, after the target mobile penetration neural network is pre-trained, it can be loaded into an edge computing device. The edge computing device refers to an electronic device located near the target electronic current transformer contained in a pre-set big data intelligent monitoring platform. By using the edge computing device to perform error prediction on the target electronic current transformer, the data transmission path can be shortened, thereby improving the error prediction efficiency.

[0080] As shown in Figure 3, Figure 3 is a schematic diagram of the network monitoring platform used to implement the online prediction method for metering error of electronic current transformers provided in the embodiments of the present invention. Specifically, the error data and corrosion monitoring data of electronic current transformers that are transmitted through the neural network are collected and calculated by the data processing device, and the error comparison monitoring is carried out in real time at the edge of the big data intelligent monitoring platform.

[0081] The aforementioned target electronic current transformer is pre-trained through steps S201-S203. In step S201, the environmental dataset can consist of historically collected environmental data of the environments in which each electronic current transformer is located, along with the corresponding error data of the electronic current transformer. This environmental data may include temperature, humidity, magnetic field, and corrosion data. Since the electronic current transformer typically measures alternating current, the error data of the aforementioned electronic current transformer may include ratio error and phase error. The ratio error may include the error between the output of the electronic current transformer and the amplitude of the original current.

[0082] In one possible implementation, the correspondence between error and environmental data can be constructed based on timestamps. For example, the error of an electronic current transformer can be expressed as having four input variables and two output variables at time t:

[0083] Where δ(t) and γ(t) represent the ratio error and phase error at time t, respectively; x1(t), x2(t), x3(t), and x4(t) represent the temperature data, humidity data, magnetic field data, and corrosion data at time t, respectively; and ε1(t) and ε2(t) represent the disturbance error caused by other environmental data, the specific value of which can be determined based on the specific values ​​of other environmental data.

[0084] From the above formula, we can obtain:

[0085] Among them, F1(t)=[δ1,δ2,…,δ n ] T F(t)2=[γ1,γ2,…,γn ] T ,X(t)=[x1(1…n),x2(1…n),x3(1…n),x4(1…n)] T B1 and B2 are the transformation matrices between the ratio error and phase error to be determined and the environmental data. The above X(t), F1(t), and F(t)2 constitute the environmental dataset.

[0086] In one possible embodiment, the above-mentioned environmental dataset can be obtained through the following steps:

[0087] S211. Obtain historically collected operating status data of electronic current transformers and corresponding environmental data, wherein the operating status data includes the ratio difference and phase difference of the electronic current transformers.

[0088] S212. Filter the operating status data of the electronic current transformer according to the preset filtering rules to obtain the filtered operating status data and the corresponding environmental data, forming an environmental dataset. The error value is the operating status data that occurs more than a preset number of times, and the outlier value is the operating status data whose difference from other operating status data is greater than a preset state difference threshold.

[0089] Due to network conditions, the operating status data of electronic current transformers (i.e., the aforementioned error situation) experiences storage delays during storage and transmission, resulting in multiple repetitions during data storage and leading to erroneous values ​​in the stored operating status data. Therefore, this type of data can be filtered when constructing the environmental dataset. Specifically, if the same operating status data appears consecutively more than a preset threshold, it can be determined as erroneous data and thus deleted along with the corresponding environmental data. In other words, if operating status data shows continuously changing order but the ratio difference and related phase differences remain unchanged, this data can be identified as erroneous and removed from the storage system during data profiling.

[0090] Outlier values ​​are operating state data whose difference from other operating state data is greater than a preset state difference threshold. For example, operating states that deviate significantly from the normal range due to problems with the electronic current transformer itself or network transmission problems are irrelevant to the neural network's learning of the correlation between the operating state of the electronic current transformer and environmental data. On the contrary, they may lead to overfitting of the model. Therefore, these operating states that deviate significantly from the normal range can be removed.

[0091] By using the above technical solution, data can be effectively filtered. Abnormal values ​​that affect the accuracy of the measurement error of the electronic current transformer that is affected by the floating penetration neural network, as well as data that deviates far from the data position reflecting the error characteristics but is not abnormal, are removed from the measurement data. This allows for the accurate analysis of the error characteristics of the electronic current transformer.

[0092] After obtaining the environmental dataset, each piece of environmental data can be input into the initial Swim through the neural network (STNN). The initial Swim through the neural network includes an input layer, a latent layer, a carrier layer, and an output layer. The latent layer is used to compress the input of the input layer into the hidden space, thereby extracting the features of the input data. The carrier layer has a time delay and is used to receive the output of the latent layer and store the output of the latent layer in the previous iteration. Then, the output of the hidden layer in the previous iteration is fed back into the current training round, thereby increasing the network's learning of the temporal features of the input data and effectively increasing the overall and local stability of the Swim through the neural network.

[0093] Each of the above network layers contains inter-layer connection weights, which are used to weight the output of the network layer before inputting it into the next layer. Each network layer also includes a bias, which is used to add random perturbation values ​​to the output of the network layer to avoid model overfitting.

[0094] In one possible embodiment, the input-output function relationship of the above-mentioned floating-through neural network can be expressed as: y(k,t)=g 1 (ω3(u(k,t)))+σ(k,t) u(k,t)=g 2 (ω1u c (k,t)+ω2x(k-1,t))+σ(k,t) u c (k,t)=u(k-1,t)

[0095] Where y(k,t) is the output of the target mobile penetration neural network, x(k-1,t) is the target environment data, u(k,t) is the output of the latent layer, and u c (k,t) represents the output from the latent layer to the carrier layer, ω1 and ω2 are the inter-layer connection weights, k is the iteration number, and g 1 () represents the activation function of the neurons in the output layer, specifically it can be... x is the function input, g 2 () represents the activation function of the neurons in the latent layer, specifically it can be... σ(k,t) is the perturbation function.

[0096] There are multiple latent layers and carrier layers mentioned above, and each latent layer corresponds one-to-one with a carrier layer. That is, the output data of a latent layer will be input into a unique carrier layer. Correspondingly, the number of latent layers and carrier layers is the same. As shown in Figure 4, which is a schematic diagram of the structure of the latent layers and carrier layers contained in the initial floating-through neural network, the latent layer outputs data to the carrier layer, and the carrier layer can feed back the previous output data of the latent layer to the latent layer.

[0097] In one possible embodiment, the number of latent layers and the number of supporting layers can be determined by the following steps:

[0098] S221. Based on the number of nodes in the input layer, the number of nodes in the output layer, and a preset number of node perturbations, calculate the range of the number of latent layers, wherein the number of nodes in the input layer is the number of data types input to the input layer, the number of nodes in the output layer is the preset number of data types output by the output layer, and the preset number of node perturbations is a constant within a preset range.

[0099] For example, when the environmental data includes temperature data, humidity data, magnetic field data, and corrosion data, the number of nodes in the input layer is 4. When the output prediction result includes ratio error and phase error, the number of nodes in the output layer is 2. The range of the number of node perturbations can be preset according to the actual application scenario, such as a constant in [3, 10].

[0100] As one possible implementation, this quantity range can be calculated using the following formula:

[0101] Where, n in (t) represents the number of nodes in the input layer, n out (t) represents the number of nodes in the output layer, c h (t) represents the preset number of node disturbances.

[0102] S222. Obtain the candidate number of latent layers based on the range of the number of latent layers, wherein the candidate number is each integer value within the range of the number of latent layers;

[0103] S223. Calculate the error deviation value for each of the potential candidate numbers based on the prediction error and the error label of the corresponding environmental data, and determine the potential candidate number of the most error deviation value among the error deviation values ​​for each of the potential candidate numbers as the target number of the potential layer and the bearing layer.

[0104] For example, each integer value in the above range can be used as a candidate number of latent layers to participate in the output prediction error, and the candidate number of latent layers corresponding to the minimum difference value can be determined as the target number based on the difference between the error output by the neural network and the corresponding error label.

[0105] The difference between the prediction error and the error label can be obtained through a target loss function. As one possible implementation, the target loss function can be:

[0106] Among them, e i (k,t) is the error label, y i (k,t) represents the prediction error output by the floating-point neural network, r i (k,t) is a standardization factor, and n is the amount of environmental data input into the target mobile penetration neural network.

[0107] As mentioned above, the parameters included in a swimming penetration neural network can include inter-layer connection weights and biases within the network layers. In one possible implementation, the optimal values ​​of these inter-layer connection weights and biases can be determined using the Particle Swarm Optimization (PSO) algorithm (also known as Particle swimming penetration optimization (PSPO)). PSO is an optimization method based on swarm intelligence. It simulates the foraging behavior of flocks of birds in nature, treating the solution space of the problem as a multi-dimensional space where each potential solution is abstracted as a "particle." As a possible implementation, the optimal values ​​of inter-layer connection weights and biases within the network layers can be determined through the following steps:

[0108] The inter-layer connection weights, bias particles, and spatial dimension are set, and the optimal values ​​of the inter-layer connection weights and biases are determined using the particle swarm optimization algorithm. The spatial dimension and total number of particles of the inter-layer connection weights and bias particles are determined by the following formula: S N (t)=n h (t)(r1n in (t)+r2n out (t)+n h (t)+1)+r2n out (t)

[0109] Among them, S N (t) represents the dimension of the space, n h (t) represents the number of targets in the latent layer, n in(t) represents the number of nodes in the input layer, n out (t) represents the number of nodes in the output layer, where r1 and r2 are both random numbers; P(t) represents the total number of particles.

[0110] Multiple particles are generated in the space corresponding to the spatial dimension, and the total number of the multiple particles is the number of particles;

[0111] In each iteration, the velocity and position of each particle are updated according to the following formula: p ij (t+1)=p ij (t)+v ij (t+1)

[0112] Where i is the number of particles, j is the dimension of the space, and v ij (t) represents the particle velocity after t iterations, p ij (t) represents the position of the particle after t iterations. Let j be the j-th dimension value of the extremum of the i-th particle. denoted as the j-th dimension value of the overall optimal position, r1, r2, and r3 are random numbers on [0,1]; c1 is the individual practice element, c2 is the social practice element, w is the trend weight, and δ is the random influence element;

[0113] In particle swarm optimization (PSO) algorithms, the trend-following weight reflects the particle's optimization ability. Setting a larger trend-following weight in the early stages of the algorithm can reduce the time it takes for particles to find the optimal position. In the later stages, a smaller trend-following weight is needed to find local optima. For example, the trend-following weight is iterated using the following formula:

[0114] Where w(t) is the follow-the-momentum weight after the t-th iteration, w max To preset the maximum value of the trend-following weight, w min To preset the minimum value of the trend-following weight, N c (t) represents the current iteration step, N. m σ(t) represents the preset maximum number of iterations, and σ(t) represents the random perturbation value. The maximum and minimum values ​​of the trend-following weights, as well as the maximum number of iterations, can be set according to the actual application scenario.

[0115] In the initial stage of the iterative process of the mobile penetrating particle swarm, none of the particles have found the optimal solution. At this time, it is necessary to focus on self-practice to autonomously find the optimal solution through mobile penetration. After the iteration ends, some penetrating particles have found the optimal solution, and at this time, the optimal solution is found based on the extensive information of the mobile particle swarm. Therefore, at time t, the individual practice element c1(t) should decrease little by little, while the social practice element c2(t) should increase little by little.

[0116] The individual practice elements and social practice elements are iterated using the following formula:

[0117] Where c1(t) is the value of the individual practice element after t iterations, and c2(t) is the value of the social practice element after t iterations. max (t) represents the preset maximum value of the practice elements, c min σ(t) represents the minimum value of the preset practice element, and σ1(t) and σ2(t) are both perturbation elements. The initial value of the above individual practice elements can be c. max (t), the initial value of the social practice element can be c. min (t).

[0118] After each update of the velocity and position of each particle, the target loss function value is calculated based on the current velocity and position of the particle. If the minimum value of the target loss function is not reached, the step of updating the velocity and position of each particle according to the following formula in each iteration is returned until the minimum value of the target loss function is obtained.

[0119] Before the iterative calculation begins, an initial population with particles of zero or arbitrarily set velocity is randomly generated in the D-dimensional space corresponding to the aforementioned spatial dimension. For example, with a total of 30 particles, 28 particles can be set to represent inter-layer weights, and 2 particles can represent biases. During the iterative calculation, particles continuously search for and update their optimal positions by sharing position information within the population. Referring to the mutation-based cognition in genetic algorithms, a random variable is added to the velocity of each particle within a certain range during each iteration of the algorithm. This method ensures the randomness of the algorithm during the search process, thereby reducing the probability of low-velocity particles appearing and improving the algorithm's ability to find the overall optimum.

[0120] In one possible embodiment, during the particle swarm optimization algorithm, a large number of low-speed particles may aggregate and get trapped in local optima due to premature particle convergence, resulting in low particle mobility and penetration ability, meaning that the particle optimization process cannot cover more dimensions. For ease of description, these low-speed particles are referred to as low-energy particles in this invention. As one possible implementation, when low-energy particles appear during the particle optimization process, they can be replaced with high-energy particles, thereby improving the particle optimization speed and particle penetration ability.

[0121] For example, low-energy particles can be replaced by the following formula: v ij (t+1)=rand(v min ,v max )δ ij (t) p ij (t+1)=rand(p min (t),pmax (t))+σ ij (t) stI′>I t ,v ij (t) <v e (t)

[0122] Among them, v ij (t+1) represents the updated particle velocity, v min v is the minimum velocity among all particles. max p is the maximum value among all particle velocities. ij (t+1) represents the updated particle position, p min (t) represents the minimum value at each particle position, p max (t) represents the maximum value at each particle position, δ ij (t) and σ ij (t) represents the perturbation element. I' is the counting variable, specifically, when the overall optimal objective function value is... The I′ count is not updated. After the update, I′ is reset to 0. t v is the iteration threshold. e (t) is the velocity threshold, i.e., the penetration effect threshold.

[0123] Through the above training process, the specific values ​​of parameters such as the number of latent layers and carrier layers, the inter-layer connection weights between each network layer, and the biases in the network layers can be obtained. Based on the network results and parameter values, the target-moving, penetrating neural network can be derived.

[0124] In one possible embodiment, the aforementioned environmental dataset can be divided into a training set and a validation set. An initial mobile penetration model is trained using the training set according to the training process described above to obtain a target mobile penetration model. The prediction performance of the target mobile penetration model is then evaluated using the validation set. The proportion of environmental data included in the training and validation sets can be set according to the actual application scenario.

[0125] As one possible implementation, the target mobile penetration neural network can be evaluated using a preset validation set according to preset statistical indicators. These preset statistical indicators include the absolute error of the stochastic mean-field technique, Ea, and the mean relative error, Er. The absolute error of the stochastic mean-field technique is calculated using the following formula:

[0126] Among them, c i To preset the standard coefficients for prediction, y i For the error labels corresponding to the environmental data, e i The prediction error value is given by n, which is the amount of environmental data input into the target swimming penetration neural network.

[0127] The average relative error is calculated using the following formula:

[0128] If the preset statistical index is lower than the preset index threshold, it is determined that the target swimming through the neural network has completed training.

[0129] This invention employs an error prediction algorithm for electronic current transformers using a moving-through neural network to perform data mining tasks. This enables error analysis of online digital intelligent monitoring data for electronic current transformers using the moving-through neural network, obtaining its error change trends. By comparing the intelligent digital monitoring data of phase difference and cross-sectional difference, calibration of the electronic current transformer using the moving-through neural network is achieved. Simultaneously, based on the offline calibration data of electromagnetic current transformers using the moving-through neural network, error correction is used to optimize the system error introduced by the magnetic current transformer using the moving-through neural network, resulting in more accurate and fair metering.

[0130] Based on the same inventive concept, this invention also provides an online prediction device for metering errors of electronic current transformers, as shown in FIG5. The device 500 may include:

[0131] The acquisition module 501 is used to acquire target environmental data of the environment in which the target electronic current transformer is located. The target environmental data includes temperature data, humidity data, magnetic field data, and corrosion data.

[0132] The input module 502 is used to input the target environment data into a pre-trained target swimming penetration neural network to obtain the error prediction result output by the target error prediction model based on the target environment data. The error prediction result includes the ratio error prediction result and the phase error prediction result.

[0133] The target-moving, penetrating neural network is pre-trained through the following steps:

[0134] Obtain an environmental dataset, including each piece of environmental data and the error label of the electronic current transformer corresponding to each piece of environmental data;

[0135] Each piece of environmental data in the environmental dataset is input into an initial floating-through neural network to obtain the prediction error output by the initial floating-through neural network; wherein, the initial floating-through neural network includes an input layer, a latent layer, a carrier layer and an output layer, and the carrier layer is used to receive the output of the latent layer and store the output of the latent layer in the previous iteration;

[0136] The parameters in the initial floating-through neural network are trained based on the difference between the prediction error and the corresponding error label of the environmental data until the difference converges, thus obtaining the target floating-through neural network; wherein, the parameters include the inter-layer connection weights between each network layer and the bias in each network layer.

[0137] An exemplary embodiment of the present invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the electronic device to perform a method according to an embodiment of the present invention.

[0138] An exemplary embodiment of the present invention also provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of the present invention.

[0139] An exemplary embodiment of the present invention also provides a computer program product, including a computer program, wherein, when executed by a computer's processor, the computer program is used to cause the computer to perform a method according to an embodiment of the present invention.

[0140] Referring to Figure 6, a structural block diagram of an electronic device 600 that can serve as a server or client of the present invention is now described, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0141] As shown in Figure 6, the electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 can also store various programs and data required for the operation of the electronic device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0142] Multiple components in electronic device 600 are connected to I / O interface 605, including: input unit 606, output unit 607, storage unit 608, and communication unit 609. Input unit 606 can be any type of device capable of inputting information to electronic device 600. Input unit 606 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 607 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 608 may include, but is not limited to, disks and optical discs. Communication unit 609 allows electronic device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth™ devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0143] The computing unit 601 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above. For example, in some embodiments, the online prediction method for metering errors of any of the above-described electronic current transformers can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 600 via ROM 602 and / or communication unit 609. In some embodiments, the computing unit 601 can be configured by any other suitable means (e.g., by means of firmware) to perform the online prediction method for metering errors of any of the above-described electronic current transformers.

[0144] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0145] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0146] As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0147] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0148] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0149] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

Claims

1. A method for online prediction of metering errors in electronic current transformers, characterized in that, The method includes: Acquire target environmental data of the environment in which the target electronic current transformer is located, including temperature data, humidity data, magnetic field data, and corrosion data; The target environment data is input into a pre-trained target swimming penetration neural network to obtain the error prediction results output by the target error prediction model based on the target environment data. The error prediction results include ratio error prediction results and phase error prediction results. The target-moving, penetrating neural network is pre-trained through the following steps: Obtain an environmental dataset, including each piece of environmental data and the error label of the electronic current transformer corresponding to each piece of environmental data; Each piece of environmental data in the environmental dataset is input into an initial floating-through neural network to obtain the prediction error output by the initial floating-through neural network; wherein, the initial floating-through neural network includes an input layer, a latent layer, a carrier layer and an output layer, and the carrier layer is used to receive the output of the latent layer and store the output of the latent layer in the previous iteration; The parameters in the initial floating-through neural network are trained based on the difference between the prediction error and the corresponding error label of the environmental data until the difference converges, thus obtaining the target floating-through neural network; wherein, the parameters include the inter-layer connection weights between each network layer and the bias in each network layer.

2. The method according to claim 1, characterized in that, The acquisition of the environmental dataset includes: Acquire historically collected operating status data of electronic current transformers and corresponding environmental data, wherein the operating status data includes the ratio difference and phase difference of the electronic current transformers; The operating status data of the electronic current transformer is filtered for error values ​​and outliers according to preset filtering rules to obtain filtered operating status data and corresponding environmental data, forming an environmental dataset. The error values ​​are operating status data that occur consecutively more than a preset threshold number, and the outliers are operating status data whose difference from other operating status data is greater than a preset state difference threshold.

3. The method according to claim 1, characterized in that, The initial floating-through neural network includes multiple latent layers and carrier layers, with each latent layer and carrier layer corresponding one-to-one. Training the parameters in the initial floating-through neural network based on the difference between the prediction error and the corresponding error label of the environmental data includes: Based on the number of nodes in the input layer, the number of nodes in the output layer, and a preset number of node perturbations, the range of the number of latent layers is calculated, wherein the number of nodes in the input layer is the number of data types input to the input layer, the number of nodes in the output layer is the preset number of data types output by the output layer, and the preset number of node perturbations is a constant within a preset range. The number of candidate latent layers is obtained based on the range of the number of latent layers, wherein the number of candidate latent layers is an integer value within the range of the number of latent layers; For each candidate number of latent layers, an error deviation value is calculated based on the prediction error and the error label of the corresponding environmental data. The candidate number of latent layers with the highest error deviation value among the error deviation values ​​is determined as the target number of the latent layer and the bearing layer.

4. The method according to claim 3, characterized in that, The step of training the parameters in the initial floating-through neural network based on the difference between the prediction error and the corresponding environmental data error labels further includes: The inter-layer connection weights, bias particles, and spatial dimension are set, and the optimal values ​​of the inter-layer connection weights and biases are determined using the particle swarm optimization algorithm. The spatial dimension and total number of particles of the inter-layer connection weights and bias particles are determined by the following formula: S N (t)=n h (t)(r1n in (t)+r2n out (t)+n h (t)+1)+r2n out (t) Among them, S N (t) represents the dimension of the space, n h (t) represents the number of targets in the latent layer, n in (t) represents the number of nodes in the input layer, n out (t) represents the number of nodes in the output layer, where r1 and r2 are both random numbers; P(t) represents the total number of particles.

5. The method according to claim 4, characterized in that, The step of using the particle swarm optimization algorithm to determine the optimal values ​​of the inter-layer connection weights and biases includes: Multiple particles are generated in the space corresponding to the spatial dimension, and the total number of the multiple particles is the number of particles; In each iteration, the velocity and position of each particle are updated according to the following formula: p ij (t+1)=p ij (t)+v ij (t+1) Where i is the number of particles, j is the dimension of the space, and v ij (t) represents the particle velocity after t iterations, p ij (t) represents the position of the particle after t iterations. Let j be the j-th dimension value of the extremum of the i-th particle. denoted as the j-th dimension value of the overall optimal position, r1, r2, and r3 are random numbers on [0,1]; c1 is the individual practice element, c2 is the social practice element, w is the trend weight, and δ is the random influence element; The trend weight is iterated using the following formula: Where w(t) is the follow-the-momentum weight after the t-th iteration, w max To preset the maximum value of the trend-following weight, w min To preset the minimum value of the trend-following weight, N c (t) represents the current iteration step, N. m σ(t) represents the preset maximum number of iterations, and σ(t) represents the random perturbation value. The individual practice elements and social practice elements are iterated using the following formula: Where c1(t) is the value of the individual practice element after t iterations, c max (t) represents the maximum value of the preset practice element, and σ1(t) and σ2(t) are both perturbation elements; After each update of the velocity and position of each particle, the target loss function value is calculated based on the current velocity and position of the particle. If the minimum value of the target loss function is not reached, the step of updating the velocity and position of each particle according to the following formula in each iteration is returned until the minimum value of the target loss function is obtained.

6. The method according to claim 5, characterized in that, The step of inputting the target environment data into a pre-trained target mobile penetration neural network to obtain the error prediction result output by the target error prediction model based on the target environment data includes: The target-moving, penetrating neural network outputs the error prediction result according to the following function: y(k,t)=g 1 (ω3(u(k,t)))+σ(k,t) u(k,t)=g 2 (ω1u c (k,t)+ω2x(k-1,t))+σ(k,t) u c (k,t)=u(k-1,t) Where y(k,t) is the output of the target mobile penetration neural network, x(k-1,t) is the target environment data, u(k,t) is the output of the latent layer, and u c (k,t) represents the output from the latent layer to the carrier layer, ω1 and ω2 are the inter-layer connection weights, k is the iteration number, and g 1 () is the activation function of the neurons in the output layer, g 2 () is the neuron activation function of the latent layer, and σ(k,t) is the perturbation function.

7. The method according to claim 5, characterized in that, The method further includes: The target mobile penetration neural network is evaluated using a preset validation set according to preset statistical indicators. These preset statistical indicators include the absolute error of the stochastic mean field technique (Ea) and the mean relative error (Er). The absolute error of the stochastic mean field technique is calculated using the following formula: Among them, c i To pre-determine the standard coefficients, y i For the error labels corresponding to the environmental data, e i The prediction error value is given by n, which represents the amount of environmental data input into the target swimming penetration neural network. The average relative error is calculated using the following formula: If the preset statistical index is lower than the preset index threshold, it is determined that the target swimming through the neural network has completed training.

8. An online prediction device for metering errors of electronic current transformers, characterized in that, The device includes: The acquisition module is used to acquire target environmental data of the environment in which the target electronic current transformer is located. The target environmental data includes temperature data, humidity data, magnetic field data, and corrosion data. The input module is used to input the target environment data into a pre-trained target swimming penetration neural network to obtain the error prediction results output by the target error prediction model based on the target environment data. The error prediction results include ratio error prediction results and phase error prediction results. The target-moving, penetrating neural network is pre-trained through the following steps: Obtain an environmental dataset, including each piece of environmental data and the error label of the electronic current transformer corresponding to each piece of environmental data; Each piece of environmental data in the environmental dataset is input into an initial floating-through neural network to obtain the prediction error output by the initial floating-through neural network; wherein, the initial floating-through neural network includes an input layer, a latent layer, a carrier layer and an output layer, and the carrier layer is used to receive the output of the latent layer and store the output of the latent layer in the previous iteration; The parameters in the initial floating-through neural network are trained based on the difference between the prediction error and the corresponding error label of the environmental data until the difference converges, thus obtaining the target floating-through neural network; wherein, the parameters include the inter-layer connection weights between each network layer and the bias in each network layer.

9. An electronic device, comprising: processor; as well as Stored program memory, The program includes instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images