Intelligent interpolation method and system for iot device data
By constructing a generative adversarial network for IoT devices, adding an auxiliary encoder, and optimizing training using test metrics, the accuracy and efficiency issues of missing data imputation for IoT devices were solved, achieving efficient data imputation processing.
Patent Information
- Application Number
- CN202210509839.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-11
- Publication Date
- 2026-01-02
- Estimated Expiration
- 2042-05-11
AI Technical Summary
Existing technologies cannot accurately impute missing data from IoT devices, and current imputation methods based on generative adversarial networks are time-consuming and lack accuracy.
An initial generative adversarial network (GAN) is constructed, and an auxiliary encoder is added. The GAN is then trained to generate IoT devices, including an auxiliary encoder, a generator, and a discriminator. The training process is optimized using test metrics to improve interpolation accuracy.
By using an improved generative adversarial network structure, the encoder can quickly find the input value after training, avoiding optimization time, improving imputation performance, and achieving accurate imputation of missing data.
Smart Images
Figure CN115270908B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, in particular to an intelligent interpolation method and system for Internet of Things device data. BACKGROUND
[0002] With the continuous development of Internet of Things technology, sensing devices are used to obtain information used by Internet of Things technology. Due to the characteristics of sensing devices collecting data, sensing technology has become an important part of many intelligent applications such as smart cities, intelligent driving, intelligent industries, and smart grids. In some fields, it is necessary to process sensor data, for example: in the power grid field, by processing sensor data, the safe and efficient operation of power equipment is ensured.
[0003] Data missing is an important problem of intelligent application of Internet of Things devices. When sensors collect data, interference may occur due to changes in the environment, causing data to be missing. At one time, the missing data needs to be interpolated for normal use.
[0004] In order to solve the problem of missing data of Internet of Things devices, most of the interpolation methods at the present stage cannot utilize global effective information, and the interpolation results are often unsatisfactory. With the development of deep learning, an interpolation method based on generative adversarial network is proposed. However, the data interpolation method based on generative adversarial network at the present stage adopts the structure of the original generative adversarial network, and a large amount of time is needed to optimize the input value of the network during the training process, which also causes the loss of interpolation accuracy. And at the present stage, the training of the network cannot find a good evaluation index to measure the good and bad of the network training, and it is impossible to accurately interpolate the missing data. SUMMARY
[0005] (I) Technical problems to be solved
[0006] In view of the deficiencies of the prior art, the present application provides an intelligent interpolation method and system for Internet of Things device data, which solves the technical problem that the prior art cannot accurately interpolate missing data.
[0007] (II) Technical solutions
[0008] In order to achieve the above purpose, the present application is realized by the following technical solutions:
[0009] The intelligent interpolation method for Internet of Things device data provided by the present application solves the technical problems, which is executed by a computer and includes the following steps:
[0010] Obtain an Internet of Things device data sample set and construct an initial generative adversarial network; the data sample set includes missing data and complete data;
[0011] train the initial generative adversarial network based on the missing data and the complete data to obtain an internet-of-things device generative adversarial network, the internet-of-things device generative adversarial network comprising an auxiliary encoder, a generator and a discriminator;
[0012] perform an imputation on the missing data of the internet-of-things device based on the internet-of-things device generative adversarial network.
[0013] Preferably, the training of the initial generative adversarial network based on the missing data and the complete data comprises:
[0014] divide the missing data into training data and test data based on a preset ratio;
[0015] perform an iterative training of the initial generative adversarial network for a preset number of rounds based on a preset loss function with the training data as input and the complete data as expected output, and test the initial generative adversarial network after the training based on the test data.
[0016] Preferably, the testing of the initial generative adversarial network after the training based on the test data comprises:
[0017] input the test data into the initial generative adversarial network after the training to obtain a test index, the test index comprising: a mean absolute error, a mean relative error, a root mean square error and a determination coefficient;
[0018] obtain a test result according to the test index;
[0019] determine the internet-of-things device generative adversarial network according to the test result.
[0020] Preferably, the method for obtaining the mean absolute error, the mean relative error and the root mean square error comprises:
[0021]
[0022]
[0023]
[0024] wherein,
[0025] MAE represents the mean absolute error, RMSE represents the mean relative error, and MAPE represents the root mean square error;
[0026] y i is an actual value corresponding to a missing position in the test data, is an imputation value of the missing position after the initial generative adversarial network after the training performs an imputation on the test data, and N is a total number of missing values.
[0027] Preferably, the method for obtaining the determination coefficient comprises:
[0028]
[0029] wherein,
[0030] R 2 represents the determination coefficient;
[0031] y i is an actual value corresponding to a missing position in test data, is an imputation value of the missing position after the initial generative adversarial network trained is used to impute the test data, and N is the total number of missing values.
[0032] Preferably, the method for obtaining the test result according to the test index comprises:
[0033] F = a * MAE + b * RMSE + c * MAPE + d * (1-R 2 )
[0034] wherein,
[0035] F represents the test result, a represents the weight of the average absolute error, b represents the weight of the average relative error, c represents the weight of the root mean square error, and d represents the weight of the determination coefficient;
[0036] The method for determining the initial generative adversarial network of the Internet of Things device according to the test result comprises:
[0037] determining the test result of the initial generative adversarial network after each round of iteration training;
[0038] determining the initial generative adversarial network corresponding to the round of iteration training of the smallest test result in all test results as the initial generative adversarial network of the Internet of Things device.
[0039] Preferably, the method for imputing the missing data of the Internet of Things device based on the initial generative adversarial network of the Internet of Things device comprises:
[0040] obtaining a random vector and a matrix mask according to the missing data of the Internet of Things device;
[0041] inputting the missing data of the Internet of Things device, the random vector and the matrix mask into the auxiliary encoder to obtain encoded data;
[0042] inputting the encoded data into the generator to make the generator impute the encoded data to obtain complete data corresponding to the missing data of the Internet of Things device.
[0043] The application provides an intelligent interpolation system for Internet of Things equipment data, which comprises the following modules.
[0044] A preparation module is configured to obtain an Internet of Things equipment data sample set, and construct an initial generative adversarial network; the data sample set comprises missing data and complete data;
[0045] A training module is configured to train the initial generative adversarial network based on the missing data and the complete data, and obtain an Internet of Things equipment generative adversarial network; the Internet of Things equipment generative adversarial network comprises an auxiliary encoder, a generator and a discriminator;
[0046] A processing module is configured to perform interpolation processing on missing data of the Internet of Things equipment based on the Internet of Things equipment generative adversarial network.
[0047] The application provides a computer readable storage medium, which stores a computer program for multi-round conversation generation based on information enhancement, and the computer program enables a computer to perform the intelligent interpolation method for Internet of Things equipment data.
[0048] The application provides an electronic device, which comprises the following modules.
[0049] One or more processors;
[0050] A memory; and
[0051] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs comprise a program for performing the intelligent interpolation method for Internet of Things equipment data.
[0052] (Three) beneficial effects
[0053] The application provides an intelligent interpolation method and system for Internet of Things equipment data. Compared with the prior art, the following beneficial effects are achieved:
[0054] This invention constructs an initial generative adversarial network (GAN) by acquiring a data sample set from IoT devices. The data sample set includes both missing and complete data. The initial GAN is trained based on the missing and complete data to obtain an IoT device GAN, which includes an auxiliary encoder, a generator, and a discriminator. The missing data from IoT devices is then imputed using the IoT device GAN. Compared to existing technologies, the GAN provided in this invention features a structural improvement by incorporating an auxiliary encoder. The encoder is trained alongside the GAN, allowing it to find the network's input values after training. This avoids spending time optimizing the input values and ensures optimal input values, significantly improving imputation performance and enabling accurate imputation of missing data. Attached Figure Description
[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0056] Figure 1 This is a schematic diagram illustrating a scenario of the intelligent interpolation method for IoT device data provided in an embodiment of the present invention. Detailed Implementation
[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0058] This application provides an intelligent interpolation method and system for IoT device data, which solves the problem that existing technologies cannot accurately interpolate missing data, and achieves accurate interpolation of missing data.
[0059] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows:
[0060] The embodiment of the application acquires an Internet of Things device data sample set, constructs an initial generative adversarial network, the data sample set includes missing data and complete data, trains the initial generative adversarial network based on the missing data and the complete data, obtains an Internet of Things device generative adversarial network, the Internet of Things device generative adversarial network includes an auxiliary encoder, a generator and a discriminator, and performs interpolation processing on missing data of the Internet of Things device based on the Internet of Things device generative adversarial network. Compared with the prior art, the generative adversarial network provided by the application is improved in structure, and the auxiliary encoder is added. The encoder is trained together with the generative adversarial network, and the input value of the network can be found after the encoder is trained, so that time is saved for optimizing the input value, the input value can reach the best, the performance of interpolation is greatly improved, and the missing data is accurately interpolated.
[0061] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in combination with the drawings of the specification and specific embodiments.
[0062] The embodiment of the application provides an intelligent interpolation method for Internet of Things device data, Figure 1 The scene schematic diagram of the intelligent interpolation method for Internet of Things device data provided by the embodiment of the application. The method is executed by a computer, and the method comprises the following steps:
[0063] S1, acquiring an Internet of Things device data sample set, constructing an initial generative adversarial network; the data sample set includes missing data and complete data;
[0064] S2, training the initial generative adversarial network based on the missing data and the complete data, obtaining an Internet of Things device generative adversarial network, the Internet of Things device generative adversarial network including an auxiliary encoder, a generator and a discriminator;
[0065] S3, performing interpolation processing on missing data of the Internet of Things device based on the Internet of Things device generative adversarial network.
[0066] Compared with the prior art, the generative adversarial network is improved in structure, and the auxiliary encoder is added. The encoder is trained together with the generative adversarial network, and the input value of the network can be found after the encoder is trained, so that time is saved for optimizing the input value, the input value can reach the best, the performance of interpolation is greatly improved, and the missing data is accurately interpolated.
[0067] The steps will be analyzed in detail below.
[0068] In step S1, an Internet of Things device data sample set is acquired, and an initial generative adversarial network is constructed; the data sample set includes missing data and complete data.
[0069] In order to generate a high-accuracy generative adversarial network, the embodiments of the present application can obtain real and complete data of various Internet of Things devices and missing data corresponding to the complete data, such as gas content sensors, pressure sensors, temperature sensors, etc. in a transformer. By comparing the missing data and the complete data, an accurate generative adversarial network is trained.
[0070] The missing data and the complete data are in a corresponding relationship. It should be noted that, compared with the complete data, the missing data is complete data in which data at some positions is missing. The complete data X com and the missing data X inc are respectively as follows:
[0071]
[0072]
[0073] As can be seen, the complete data and the missing data are two data of the same size, except that some data at positions in the missing data is missing. Therefore, the missing data can include two cases, missing positions and remaining data. The missing positions are NaN, and the remaining data is the content with real data.
[0074] The remaining data X rem is as follows:
[0075]
[0076] At the same time, an initial generative adversarial network corresponding to the Internet of Things device can be constructed.
[0077] Compared with the prior art, the initial generative adversarial network constructed in the embodiments of the present application has a structural improvement, and an auxiliary encoder is introduced. The auxiliary encoder is essentially an encoder.
[0078] Specifically, the initial generative adversarial network includes an auxiliary encoder, a generator, and a discriminator connected in sequence. The encoder can find the best random noise corresponding to the missing data input to the generator. The generator uses the best random noise input by the encoder to interpolate the missing data to generate a complete data simulation value corresponding to the missing data. The discriminator is used to discriminate between the complete data simulation value generated by the generator and the real complete data value.
[0079] The auxiliary encoder, the generator, and the discriminator can all be regarded as a network form.
[0080] The encoder network comprises two network layers and two fully connected layers, the fully connected layers, the network layers, the network layers and the fully connected layers are sequentially connected; the fully connected layers and the network layers are provided with ReLU activation function layers, and the fully connected layers and the network layers are connected through the ReLU activation function layers. The two network layers each comprise a convolution layer, a normalization layer and a ReLU activation function layer.
[0081] The generator network comprises two network layers and two fully connected layers, the fully connected layers, the network layers, the network layers and the fully connected layers are sequentially connected; the fully connected layers and the network layers are provided with ReLU activation function layers, and the last fully connected layer is provided with a Tanh activation function layer. The two network layers each comprise a deconvolution layer, a normalization layer and a ReLU activation function layer.
[0082] The encoder network comprises four network layers and two fully connected layers, the fully connected layers, the network layers, the network layers, the network layers, the network layers and the fully connected layers are sequentially connected; the fully connected layers and the network layers are provided with ReLU activation function layers, and the last fully connected layer is provided with a Sigmoid activation function layer. The four network layers each comprise a convolution layer, a normalization layer and a ReLU activation function layer.
[0083] In step S2, the initial generative adversarial network is trained based on the missing data and the complete data to obtain an Internet of Things device generative adversarial network, and the Internet of Things device generative adversarial network comprises an auxiliary encoder, a generator and a discriminator.
[0084] Specifically, the initial generative adversarial network can be trained, and the training specifically comprises the following steps.
[0085] S201: The missing data is divided into training data and test data based on a preset ratio.
[0086] In the embodiment of the application, when the initial generative adversarial network is trained, testing is required at the same time to improve the accuracy of the final network.
[0087] S202: The initial generative adversarial network is iteratively trained for a preset number of rounds based on a preset loss function with the training data as input and the complete data as expected output, and the initial generative adversarial network after training is tested based on the test data.
[0088] During the preset number of rounds of iterative training of the initial generative adversarial network, data imputation training can be performed, and the data imputation training process specifically comprises the following steps.
[0089] S2021: A matrix mask and a random vector are obtained according to the missing data and the complete data. The matrix mask is a matrix with the same size and sample data as the random vector, and the random vector is noise.
[0090] Specifically, the matrix mask is obtained according to the missing data and the complete data, and 1 in the mask represents that the data is not missing, and 0 represents that the data is missing. An example of the matrix mask M is shown as follows:
[0091]
[0092] A random vector can also be generated. The random vector is a data obtained by filling the positions corresponding to 0 in the remaining data.
[0093] S2022: inputting the missing data, the matrix mask and the random vector into the initial generative adversarial network to obtain the interpolation data.
[0094] According to the initial generative adversarial network, the missing data can be interpolated to obtain interpolation data, but the effect may be poor, and therefore continuous training is required.
[0095] S2023: inputting the interpolation data and the complete data into the discriminator to perform back propagation based on a second total loss function.
[0096] Specifically, the weight parameters of the initial generative adversarial network can be randomly initialized using a standard Gaussian distribution. Then, the back propagation of the discriminator is performed using the second total loss function. The second total loss function is a pre-generated loss function, and the second total loss function is as follows:
[0097]
[0098] wherein, L D represents the second total loss function of the discriminator;
[0099] p(x rem ) represents the distribution of the remaining data, and p(x com ) represents the distribution of the sensor sample data; represents the uniform sampling distribution between p(x rem ) and p(x com ) straight line, and D(X com ) represents the probability value of the complete data being true; represents the gradient penalty value of the discriminator, and λ represents the gradient penalty parameter.
[0100] S2024: performing back propagation based on a first total loss function. The first total loss function is a pre-generated loss function, and the first total loss function is as follows:
[0101] L E,G =αL E +L G
[0102] wherein,
[0103] L E,G is the total loss function of the auxiliary encoder and generator joint;
[0104] L E is the total loss function of the auxiliary encoder, L G is the total loss function of the generator, and a is the weight parameter of the auxiliary encoder;
[0105] X gen = G (E (Z, X rem , M))
[0106] L E = ||X gen ⊙M-X rem ⊙M||2
[0107] X imp = X rem ⊙M+X gen ⊙ (1-M)
[0108]
[0109] wherein,
[0110] Z represents a random vector, M represents a matrix mask; X gen represents the generated data generated by the generator; X rem represents the remaining data in the missing data, X imp represents the imputed data;
[0111] E [*] represents the expected value of the distribution function, ||X gen ⊙M-X rem ⊙M||2 represents the reconstruction loss of the generated data and the remaining data, p z (z) represents the distribution of z, D (X imp ) represents the probability value that the imputed data is true.
[0112] S2025: Iterative training, performing a preset number of iterations of iterative training. The user can set the number of training rounds, for example, twenty thousand times.
[0113] When training the initial generative adversarial network, testing needs to be performed at the same time as training, including the following steps:
[0114] S2121: After each iteration of the initial generative adversarial network, input the test data into the trained initial generative adversarial network to obtain test indicators, including: mean absolute error, mean relative error, root mean square error and determination coefficient.
[0115] It should be noted that during the training process, after each iteration of training, the test data needs to be input into the network after the round of training, and the training of the network is tested according to the evaluation index. The test indexes mainly include mean absolute error, mean relative error, root mean square error and determination coefficient.
[0116] The acquisition method of each test index is as follows:
[0117]
[0118]
[0119]
[0120] Among them,
[0121] MAE represents the mean absolute error, RMSE represents the mean relative error, and MAPE represents the root mean square error.
[0122] y i is the actual value corresponding to the missing position in the test data, is the imputation value of the missing position after the initial generative adversarial network after training is used to impute the test data, and N is the total number of missing values.
[0123]
[0124] Among them,
[0125] R 2 represents the determination coefficient.
[0126] S2122: Obtain a test result according to the test index.
[0127] MAE and RMSE are the deviation degree of the imputation value from the true value, the closer to 0, the better the imputation effect; MAPE is the proportion of the difference between the imputation value and the actual value to the true value, the closer to 0, the better the effect; R 2 is the fitting degree of the evaluation of the imputation result, the closer to 1, the imputation data can represent the true sample.
[0128] Therefore, the test result in the embodiment of the application is set as:
[0129] F=a*MAE+b*RMSE+c*MAPE+d*(1-R 2 )
[0130] Among them,
[0131] F represents the test result, a represents the weight of the mean absolute error, b represents the weight of the mean relative error, c represents the weight of the root mean square error, and d represents the weight of the determination coefficient.
[0132] The weight of each test index can be set by the user, and can be a=b=c=d=0.25.
[0133] According to the above formula, the test result of the initial generative adversarial network after each iteration training can be determined.
[0134] S2123: Determine the Internet of Things device generative adversarial network according to the test result.
[0135] Specifically, the test result of the initial generative adversarial network after each iteration training can be determined first. Taking 20,000 times as a preset number of rounds for example, 20,000 test results can be obtained, and each test result corresponds to an initial generative adversarial network of a round.
[0136] In some embodiments, the initial generative adversarial network corresponding to the minimum test result in all test results can be determined as the Internet of Things device generative adversarial network.
[0137] The smaller the test result is, the better the accuracy of the network is, so the minimum test result can be determined, and the network of the corresponding round can be determined as the trained network, i.e., the final Internet of Things device generative adversarial network.
[0138] In some embodiments, in order to avoid the influence of network volatility, the Internet of Things device generative adversarial network can also be selected in the network training stable stage.
[0139] Specifically, the stable stage of the network needs to be determined first. A network threshold can be set, which can be 1,000. If the difference between the maximum value and the minimum value in the test results of the networks after 1,000 adjacent training satisfies a preset condition, which can be that the difference is less than the proportion of the minimum value, for example, less than one tenth of the minimum value, it is considered that the networks after 1,000 training are in a stable stage.
[0140] In the stable stage, the network with the minimum test result can be selected as the Internet of Things device generative adversarial network.
[0141] In step S3, the Internet of Things device missing data is processed by the Internet of Things device generative adversarial network.
[0142] Specifically, the following steps are included:
[0143] The random vector and the matrix mask are obtained according to the Internet of Things device missing data. Specifically, the auxiliary encoder can obtain the optimal random vector.
[0144] The Internet of Things device missing data, the random vector and the matrix mask are input into the auxiliary encoder to obtain the encoded data.
[0145] The encoding data is input into the generator to enable the generator to interpolate the encoding data to obtain complete data corresponding to the missing data of the Internet of Things device.
[0146] The embodiment of the present application also provides an intelligent interpolation system for Internet of Things device data.
[0147] The preparation module is configured to obtain an Internet of Things device data sample set and construct an initial generative adversarial network; the data sample set comprises missing data and complete data;
[0148] The training module is configured to train the initial generative adversarial network based on the missing data and the complete data to obtain an Internet of Things device generative adversarial network, wherein the Internet of Things device generative adversarial network comprises an auxiliary encoder, a generator and a discriminator;
[0149] The processing module is configured to perform interpolation processing on missing data of an Internet of Things device based on the Internet of Things device generative adversarial network.
[0150] It can be understood that the above intelligent interpolation system provided by the embodiment of the present application corresponds to the above intelligent interpolation method, and the explanation, examples, beneficial effects and the like of the related content can refer to the corresponding content in the intelligent interpolation method for Internet of Things device data, which will not be repeated here.
[0151] The embodiment of the present application also provides a computer readable storage medium, and at least one instruction is stored on the medium, and at least the one instruction is loaded and executed by a processor to implement the intelligent interpolation method for Internet of Things device data as described above.
[0152] The embodiment of the present application also provides an electronic device, which comprises:
[0153] One or more processors;
[0154] A memory; and
[0155] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the programs comprise instructions for executing the intelligent interpolation method for Internet of Things device data as described above.
[0156] Compared with the prior art, the above-mentioned embodiment of the present application has the following beneficial effects:
[0157] The embodiment of the application acquires an Internet of Things device data sample set, constructs an initial generative adversarial network, the data sample set includes missing data and complete data, trains the initial generative adversarial network based on the missing data and the complete data, obtains an Internet of Things device generative adversarial network, and the Internet of Things device generative adversarial network includes an auxiliary encoder, a generator and a discriminator; and the Internet of Things device generative adversarial network is used for supplementing missing data of the Internet of Things device.
[0158] The auxiliary encoder is added before the generator in the original structure of the generative adversarial network, the encoder is trained together with the generative adversarial network, the input value of the network can be found after the training of the encoder, the time for optimizing the input value is saved, the input value can be optimized to the best, and the performance of the supplement is greatly improved. In addition, the training evaluation index is introduced in the whole network training stage to monitor the state of the network training, help the network training, and accurately supplement the missing data.
[0159] It should be noted that, through the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary general hardware platforms. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in the embodiments or some parts of the embodiments. In the specification provided herein, a large number of specific details are described. However, it can be understood that the embodiments of the application can be practiced without these specific details. In some examples, well-known methods, structures and techniques are not described in detail in order not to obscure the understanding of the specification.
[0160] In this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.
[0161] The above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. An intelligent imputation method of Internet of Things device data, the intelligent imputation method being executed by a computer, characterized in that, The method comprises the following steps: obtaining an Internet of Things device data sample set and constructing an initial generative adversarial network; the data sample set comprises missing data and complete data; training the initial generative adversarial network based on the missing data and the complete data to obtain an Internet of Things device generative adversarial network, wherein the Internet of Things device generative adversarial network comprises an auxiliary encoder, a generator and a discriminator; performing imputation processing on Internet of Things device missing data based on the Internet of Things device generative adversarial network; the training of the initial generative adversarial network based on the missing data and the complete data comprises: dividing the missing data into training data and test data based on a preset ratio; performing preset rounds of iterative training on the initial generative adversarial network based on a preset loss function, taking the training data as input and taking the complete data as expected output, and testing the initial generative adversarial network after training based on the test data; the testing of the initial generative adversarial network after training based on the test data comprises: inputting the test data into the initial generative adversarial network after training to obtain a test index after each round of iterative training, wherein the test index comprises: mean absolute error, mean relative error, root mean square error and determination coefficient; obtaining a test result according to the test index; determining an Internet of Things device generative adversarial network according to the test result; the method for obtaining the mean absolute error, the mean relative error and the root mean square error comprises: wherein, MAE represents the mean absolute error, RMSE represents the mean relative error, and MAPE represents the root mean square error; y i to test the actual value corresponding to the missing position in the data, the interpolation value of the missing position after the initial generative adversarial network trained is used to interpolate the test data, and N is the total number of missing values; the method for obtaining the determination coefficient comprises: wherein, R 2 denotes the determination coefficient; y i to test the actual value corresponding to the missing position in the data, the interpolation value of the missing position after the initial generative adversarial network trained is used to interpolate the test data, and N is the total number of missing values. the method for obtaining the test result according to the test index comprises: F = a * MAE + b * RMSE + c * MAPE + d * (1 - R 2 ) wherein, F represents the test result, a represents the weight of the mean absolute error, b represents the weight of the mean relative error, c represents the weight of the root mean square error, and d represents the weight of the determination coefficient; the method for determining the Internet of Things device generative adversarial network according to the test result comprises: determining the test result of the initial generative adversarial network after each round of iterative training; determining the initial generative adversarial network of the corresponding round of the smallest test result in all test results as the Internet of Things device generative adversarial network.
2. The intelligent imputation method of claim 1, wherein, the method for performing imputation processing on Internet of Things device missing data based on the Internet of Things device generative adversarial network comprises: obtaining a random vector and a matrix mask according to the Internet of Things device missing data; inputting the Internet of Things device missing data, the random vector and the matrix mask into the auxiliary encoder to obtain encoded data; inputting the encoded data into the generator to enable the generator to perform imputation processing on the encoded data to obtain complete data corresponding to the Internet of Things device missing data.
3. An intelligent imputation system of IoT device data, applying the intelligent imputation method of claim 1, characterized in that, The system comprises: a preparation module configured to obtain an Internet of Things device data sample set and construct an initial generative adversarial network; the data sample set comprises missing data and complete data; The training module is configured to train the initial generative adversarial network based on the missing data and the complete data to obtain an Internet of Things device generative adversarial network, the Internet of Things device generative adversarial network comprising an auxiliary encoder, a generator and a discriminator; The processing module is configured to perform an interpolation processing on the missing data of the Internet of Things device based on the Internet of Things device generative adversarial network.
4. A computer-readable storage medium, characterized in that, It stores a computer program for multi-round dialogue generation based on information enhancement, wherein the computer program enables a computer to execute the intelligent interpolation method for Internet of Things device data as claimed in claim 1 or 2.
5. An electronic device, comprising: Comprise: One or more processors; Memory; And One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs comprising instructions for executing the intelligent interpolation method for Internet of Things device data as claimed in claim 1 or 2.