CN-120180365-B - Data prediction processing system based on Internet of things safety

CN120180365BCN 120180365 BCN120180365 BCN 120180365BCN-120180365-B

Abstract

The invention discloses a data prediction processing system based on the safety of the Internet of things, which relates to the technical field of data processing and comprises the following steps of data preprocessing; the method comprises the steps of data stabilization, ARIMA modeling and residual extraction, LSTM residual modeling and prediction result fusion, wherein ARIMA modeling and residual extraction, LSTM residual modeling and prediction result fusion are carried out, ARIMA is responsible for extracting dominant linear rules in time sequences, LSTM is focused on learning residual sequences after ARIMA prediction, double-layer modeling of 'linear fundamental mode + nonlinear correction' is achieved, complex time sequence data processing in safety data of the Internet of things is effectively achieved, alpha t dynamic weight coefficients are added in prediction result fusion, contribution proportion of ARIMA and LSTM prediction results is controlled, beta sensitivity adjusting factors are determined, steepness degree of weight along with residual change is determined, flexible adjustment can be carried out according to actual conditions, and flexibility of a system is improved.

Inventors

Shao Chifeng
Li Duoqin
LI FUFU
WANG QIANQIAN
LI KE

Assignees

安徽理工大学

Dates

Publication Date: 20260508
Application Date: 20250310

Claims (6)

1. A data prediction processing system based on the safety of the Internet of things is characterized by comprising the following steps: The method comprises the steps of firstly, preprocessing data, wherein the data comprises unified multi-source data time stamps, linear interpolation or forward filling is adopted for missing values, and the data comprises operation logs and sensor data of equipment; Step two, data stabilization, which comprises judging data stability through ADF test, and performing differential or logarithmic transformation; step three, ARIMA modeling and residual error extraction; Modeling LSTM residual errors; Fifthly, fusion of prediction results, generation of baseline prediction by ARIMA, prediction of a parameter sequence by LSTM, superposition of the prediction results and the parameter sequence to obtain a result, In the fifth step, fusion is carried out through a superposition formula The superposition formula is: , Wherein For the sensitivity coefficient, the weight is adaptively adjusted along with the absolute value of the residual, The value is between 0 and 1, and is a dynamic weight coefficient; beta parameter adjusting strategy; beta is dynamically adjusted by a PID controller, and the adjusting formula is as follows: Wherein 、、 Representing the proportional, integral and differential coefficients, respectively; Step six, evaluating indexes for quantifying the prediction capacity of the model; the third step comprises the following steps: S31, defining parameters in ARIMA (p, d, q), wherein p is an autoregressive order, d is a differential order, and q is a moving average order; S32, parameter determination, ACF/PACF graph analysis, if ACF tailing, PACF tail cutting and k-order AR (k), if ACF truncates and m-order PACF tails MA (m), minimizing AIC or BIC, , K is the total number of model parameters, L is a likelihood conversion value, and n is a sample size; s33, model training, wherein an ARIMA equation is as follows: l is a hysteresis operator, and the hysteresis operator, ; As a result of the autoregressive coefficients, In order to move the coefficient of the average, ; S34, residual extraction and white noise detection, wherein the residual sequence is that The white noise test uses Ljung-Box test, , For the autocorrelation coefficient of the lag k-order, m is the maximum lag order.
2. The system for predicting and processing data based on Internet of things security according to claim 1, wherein the following formula is adopted in the first step: Wherein, the Interpolation weight, taking the value influence before and after 0.5 balance.
3. The data prediction processing system based on the security of the Internet of things according to claim 1, wherein the second step adopts the following formula: And (3) difference: D is the differential order; Logarithmic transformation: , is a small constant, preventing zero value from overflowing.
4. The data prediction processing system based on the security of the internet of things according to claim 1, wherein the fourth step comprises the following steps: s41, dividing a time window, and { a residual sequence } Conversion to supervised learning format: t is the length of a time window and is a periodic integer multiple; S42, normalization processing, min-Max scaling to [ -1,1] interval, E is the whole data of the residual sequence; S43, carrying out unit internal operation; s44, network topology design, including an input layer, a hidden layer and an output layer.
5. The data prediction processing system based on the security of the Internet of things as set forth in claim 4, wherein the intra-unit operation step includes: S431, forgetting the door, ; S432, an input gate, , ; S433, updating the state of the cells, ; S434, outputting the gate, generating the current hidden state, , Wherein For the weight matrix and the bias term, As a function of the Sigmoid, Is Hadamard product.
6. The system for predicting and processing data based on Internet of things safety according to claim 1, wherein in the sixth step, the indexes of the evaluation index include average absolute error, average absolute percentage error, root mean square error and symmetric average absolute percentage error.

Description

Data prediction processing system based on Internet of things safety Technical Field The invention relates to the technical field of data processing, in particular to a data prediction processing system based on the safety of the Internet of things. Background The data prediction processing system is a full-flow platform integrating data acquisition, preprocessing, algorithm modeling and result application, and aims to mine rules from historical and real-time data so as to form scientific predictions of future trends or events. According to the application publication number CN117574948A, a data prediction processing method and system based on a neural network is described, wherein the method and system are described in the following commonly referred to reference patents, wherein the number of nodes of an input layer and an output layer is determined, the number of layers and the number of nodes of an hidden layer are determined, a BP neural network basic model is built, a particle swarm algorithm is obtained, the BP neural network basic model is optimized according to the particle swarm algorithm, a BP neural network optimization model is generated, a result acquisition value is obtained, a predicted value is obtained according to the BP neural network optimization model, the result acquisition value is compared with the predicted value, and processing is performed according to a comparison result. According to the method, a predicted value can be obtained through a BP neural network optimization model, whether data abnormality exists or not is accurately judged by comparing the predicted value with a result acquisition value, the hardware cost of equipment is not required to be increased, and the content of safe operation of the equipment is not influenced. In the field of internet of things security, the internet of things devices are various, and data sources are wide, including sensors, cameras, intelligent terminals and the like, so that complex time sequence data exist, and the algorithm in the reference patent cannot process the complex time sequence data. In summary, a data prediction processing system based on the security of the internet of things is designed. Disclosure of Invention The invention aims to overcome the defects and provides a data prediction processing system based on the safety of the Internet of things. The invention realizes the above purpose through the following technical scheme: a data prediction processing system based on the security of the Internet of things comprises the following steps: Firstly, preprocessing data, unifying multi-source data time stamps, adopting linear interpolation or forward filling for missing values, wherein clocks of equipment of the Internet of things are possibly asynchronous and need to be aligned to the same time reference (such as UTC) so as to ensure time sequence continuity; Step two, data stabilization, namely judging data stability through ADF test, carrying out differential or logarithmic transformation if necessary, wherein ARIMA requires the data to meet weak stability (mean, variance and covariance do not change with time); step three, ARIMA modeling and residual error extraction; Modeling LSTM residual errors; fifthly, fusion of prediction results, generating baseline prediction by ARIMA, predicting a parameter sequence by LSTM, and overlapping the two to obtain a result; Step six, evaluating indexes for quantifying the prediction capacity of the model; the third step comprises the following steps: S31, defining parameters in ARIMA (p, d, q), wherein p is an autoregressive order (the influence length of a historical value on a current value), d is a differential order (the minimum operation frequency for stabilizing data), and q is a moving average order (the correction frequency of a historical error on the current value); S32, parameter determination, ACF/PACF graph analysis, if ACF tailing, PACF tail cutting and k-order AR (k), if ACF truncates and m-order PACF tailsMA (m), minimizing AIC or BIC,,K is the total number of model parameters, L is a likelihood conversion value, and n is a sample size; s33, model training, wherein ARIMA equation is that L is a hysteresis operator, and the hysteresis operator,; As a result of the autoregressive coefficients,In order to move the coefficient of the average,; S34, residual extraction and white noise detection, wherein the residual sequence is thatThe white noise test uses Ljung-Box test (if p value >0.05, indicating that the residual has no autocorrelation),,For the autocorrelation coefficient of the lag k-order, m is the maximum lag order. Preferably, the following formula is adopted in the first step: Wherein, the Interpolation weights are often taken to balance the effect of the values before and after 0.5. Preferably, the second step adopts the following formula: And (3) difference: - D is the differential order; Logarithmic transformation: , is a small constant, preventing zero value