CN103309235A - Industrial control system active safety method - Google Patents

Abstract

The invention discloses a data prediction method for an industrial control system active safety mechanism. The method mainly comprises the steps of: preprocessing industrial field data, building a model according to space-time sequential data variation rules, and dynamically predicting the data in real time so as to provide safety references. By utilizing the real-time data prediction method based on a least square support vector machine, an input vector is restructured after noise is removed according to input data characteristics, the delay time and embedded dimension parameters of a restructured phase space are determined, and the regularization parameters and kernel function parameters of the model are determined; a regression model is trained, a kernel functional matrix is structured, and an N-dimensional linear equation system is solved so as to obtain a Lagrangian multiplier and a deviant, so that a final decision function is realized; a single-step prediction is realized according to the output of a model prediction system, and a self-adaption related iterative method is adopted to realize multi-step prediction; finally, the real-time, online and multi-step prediction is realized, the prediction efficiency and accuracy are improved, and a reasonable guidance is provided for spot dispatching and controlling personnel.

Description

A kind of industrial control system is the method for anti-danger initiatively

Technical field

The present invention designs the anti-danger technology of industrial control system, and the industrial control system that refers to especially a kind of time-based sequence data is the method for anti-danger initiatively.

Background technology

From industrial accident prevention and control field, the flow of research of its control theory and technology obviously lags behind economy, social requirement to industrial development, in a lot of weak links that exist aspect accident characteristic, prevention and the control theory research.To the research that the anti-danger property of industrial accident system is carried out in theory, used, pay attention to the profound theory and technology problem of solution accident frequent occurrence, to reach the purpose of control industrial accident danger.The Evolution Characteristics of research and exploration industrial accident, pests occurrence rule and effective control technology become China's industry must pay attention to also the as early as possible great safe problem of solution.

The anti-danger of the active of based on data prediction, namely set up the data prediction model by historical data, realize the prediction to the industrial control system image data, thereby can provide to the security risk of industrial control system accurate prediction, accomplish the early-warning and predicting to danger, take precautions against in possible trouble.The anti-danger of raising system if realize trend prediction to industrial control system, thereby can provide to the safety of industrial control system accurate prediction, is to realize the initiatively important place of anti-danger of industrial control system.

Traditional anti-danger security mechanism is passive operation demonstration mode, can't shift to an earlier date the danger of Prediction System state-evaluation.The danger technology of initiatively preventing of temporal based data prediction is verified at any time to the state of equipment, accomplishes the anti-possible trouble of being loyal to.Set up model according to the rule that the space-time sequence data at industrial control system scene changes, dynamic, real-time single step and even the multistep data prediction of carrying out.Precision of prediction is most crucial problem in the time series forecasting research always, but in the very high industrial control field of requirement of real-time, forecasting efficiency be key problem of equal importance.How under the prerequisite that guarantees high precision of prediction, at utmost improving forecasting efficiency and satisfy the industrial control system requirement of real-time, is the initiatively major issue of anti-danger of industrial control system.Need design to realize satisfying that the seasonal effect in time series of industrial control system is general, dynamic, the real-time estimate algorithm, reach the high and high double goal of forecasting efficiency of precision of prediction.

Summary of the invention

In view of this, fundamental purpose of the present invention is to provide the initiatively method of anti-danger of kind of the industrial control system of time-based sequence data, on the basis that guarantees single step, multistep data prediction precision, maximization improves data prediction efficient, the industrial control system requirement of real-time is satisfied in assurance, the application technology means are so that the method on system robustness and generalization ability, also has outstanding performance.The final Industry Control scene of giving provides in real time reliably early warning information.

In order to achieve the above object, be divided into following step in the design of the method:

Step 1, determine the moving window parameter, choose the real time data that industrial control system satisfies moving window quantity, different pieces of information is weighted; , set up the LS-SVM model according to the weighting real time data as training sample;

Step 2, new input sample is carried out pre-service, remove the industry spot noise effect, dimension and time delay the input sample is carried out phase space reconfiguration and normalization according to embedding, so that input vector satisfies the prediction requirement, set up weighting LS-SVM forecast model according to input vector;

Step 3, the input vector of processing is inputed to the training pattern that step 1 is set up as forecast model, obtain prediction of output value;

Step 4, the prediction output valve that will obtain are used the iteration autocorrelation technique as feedback, set up the multi-step prediction that the forecast model feedback realizes data;

Step 5, predicted value and actual value are compared, carry out precision evaluation and counting yield evaluation to predicting the outcome, judge whether to carry out model

Upgrade, need to upgrade and then rebuild the LS-SVM forecast model;

Adopt the variation tendency that method for designing of the present invention can Accurate Prediction Industry Control field data, early warning signal is sent in the warning setting that coupling system is default in advance, for the site operation personnel provides strong support, can effectively prevent the dangerous generation of locking system.

The present invention effectively raises pre-warning time, adopts the predicting means of multi-step prediction can effectively find danger, and the error starting of alarm can be effectively avoided in the adding of adaptive ability, has improved the accuracy of early warning.

The present invention is before setting up model, use the method for vectorial phase space reconfiguration, make the input data be more suitable for the multi-step prediction of forecast analysis, the employing of least square vector machine effectively reduces the complexity of model, enables better to be used in the better industrial control system of real-time.Self-timing of the present invention carries out model modification, guarantees that precision of prediction can not descend, and guarantees the requirement to forecasting efficiency;

Description of drawings

Figure 1 shows that general frame figure of the present invention;

Embodiment

For making the purpose, technical solutions and advantages of the present invention clearer, lift specific embodiment below in conjunction with accompanying drawing, the present invention is done further clear, detailed, complete explanation, and described embodiment only is a part of embodiment of the present invention, but not whole embodiment.

Traditional anti-danger security mechanism is passive operation demonstration mode, can't shift to an earlier date the danger of Prediction System state-evaluation.The danger technology of initiatively preventing of temporal based data prediction is verified at any time to the state of equipment, accomplishes to prevent trouble before it happens.How under the prerequisite that guarantees high precision of prediction, at utmost improve forecasting efficiency and satisfy the industrial control system requirement of real-time.What the embodiment of the invention provided a kind of time-based sequence prediction initiatively prevents the danger technology, as shown in Figure 1, comprising:

Step 1, determine the moving window parameter, choose the real time data that industrial control system satisfies moving window quantity, different pieces of information is weighted; , set up the LS-SVM model according to the weighting real time data as training sample;

At first according to the concrete condition of industrial field device, determine the sliding window parameter of image data, obtain the data of input; Use Bayesian evidence framework to determine the weights of different input data, be used for improving the robustness of forecast model, make up weighted prediction model;

Step 2, new input sample is carried out pre-service, remove the industry spot noise effect, dimension and time delay the input sample is carried out phase space reconfiguration and normalization according to embedding, so that input vector satisfies the prediction requirement, set up weighting LS-SVM forecast model according to input vector;

The input sample is carried out pre-service, use wavelet analysis to remove the industry spot noise; Determine that vectorial reconstruction parameter embeds dimension and time delay.Carry out vectorial phase space reconfiguration; Use and adopt the Bayesian evidence framework method to determine parameter, carry out the LS-SVM modeling according to the reconstruct vector as training sample; Model representation is

$f\left(x\right)=\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{\mathrm{\α}}_i\mathrm{\φ}{\left(x\right)}^T\mathrm{\φ}\left(x_i\right)+b=\underset{i=1}{\overset{l}{\mathrm{\Σ}}}{\mathrm{\α}}_{i}K(x_i,x)+b$

α wherein _iBe the Lagrange multiplier, kernel function is chosen Wavelet Kernel, and the Parameters in Formula formula is as follows:

$\left\{\begin{array}{c}b=\frac{{\stackrel{\→}{1}}^T{A}^{-1}y}{{\stackrel{\→}{1}}^T{A}^{-1}\stackrel{\→}{1}}\\ \mathrm{\α}=A^{-1}(Y-b\stackrel{\→}{1})\end{array}\right.$

Step 3, the input vector of processing is inputed to the training pattern that step 1 is set up as forecast model, obtain prediction of output value;

Step 4, the prediction output valve that will obtain are used the iteration autocorrelation technique as feedback, set up the multi-step prediction that the forecast model feedback realizes data; Adopt the autocorrelation analysis method to determine and the most closely-related time series historical data of predicted value, determine to embed dimension and time delay, the input vector of structure LS-SVM makes historical data as much as possible participate in modeling.The precise decreasing of avoiding sample to replace actual value to consist of by too much predicted value and causing.

Step 5, predicted value and actual value are compared, carry out precision evaluation and counting yield evaluation to predicting the outcome, judge whether to carry out model modification, need to upgrade and then rebuild the LS-SVM forecast model;

The industrial control system of the time-based sequence prediction that example of the present invention provides is initiatively prevented the danger technology, use the existing real time data of industry spot that the contingent future of system is predicted, the system that gives is maximized, initiatively anti-danger pre-warning time and space, eliminate the danger of industrial control system, accomplish to prevent trouble before it happens.Adopt the multiple technologies means, with high forecasting efficiency and precision of prediction, can satisfy the anti-danger requirement of industrial control system.

The above only is preferred embodiment of the present invention, and is in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of doing, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (6)

1. the time series predicting model based on LS-SVM is used with the local single-point of industrial control system and is initiatively prevented danger.It is characterized in that, he may further comprise the steps:

Step 1, determine the moving window parameter, choose the real time data that industrial control system satisfies moving window quantity, different pieces of information is weighted; , set up the LS-SVM model according to the weighting real time data as training sample;

Step 2, new input sample is carried out pre-service, remove the industry spot noise effect, dimension and time delay the input sample is carried out phase space reconfiguration and normalization according to embedding, so that input vector satisfies the prediction requirement, set up weighting LS-SVM forecast model according to input vector;

Step 3, the input vector of processing is inputed to the training pattern that step 1 is set up as forecast model, obtain prediction of output value;

Step 4, the prediction output valve that will obtain are as feedback, and iteration is input vector, carries out the iteration self-adapting multi-step prediction;

Step 5, predicted value and actual value are compared, judge whether to carry out model modification, need to upgrade and then rebuild the LS-SVM forecast model.

2. time series predicting model according to claim 1 is characterized in that, the weighting input data procedures of obtaining in the step 1 is:

Step 11, determine the sliding window parameter of Industry Control collection in worksite data, obtain the data of input;

The weights of step 12, definite different input data, for the robustness that improves forecast model, the structure weighted prediction model.

3. time series predicting model according to claim 2 is characterized in that, definite use Cauchy of weights weighting that distributes in the step 1.

4. time series predicting model according to claim 1 is characterized in that, sets up weighting LS-SVM forecast model in the step 2:

Step 21, the input sample is carried out pre-service, remove the industry spot noise;

Step 22, determine that vectorial reconstruction parameter embeds dimension and time delay.Carry out vectorial phase space reconfiguration;

Step 23, carry out the LS-SVM modeling according to the reconstruct vector as training sample; Model representation is

α wherein _iBe the Lagrange multiplier, kernel function is chosen Wavelet Kernel, and the Parameters in Formula formula is as follows:

5. time series predicting model according to claim 4 is characterized in that, parameter determines to adopt the Bayesian evidence framework method in the step 2.

6. time series predicting model according to claim 1 is characterized in that, step 5 is judged the LS-SVM prediction effect, and the predicted time that uses when model surpasses setting value, upgrades when perhaps model prediction precision is lower than threshold value.