CN105607477A - Industrial control circuit oscillation detection method based on improved local mean value decomposition - Google Patents

Abstract

The invention discloses an industrial control loop oscillation detection method based on improved local mean value decomposition, comprising the following steps: collecting a group of process history data in advance in the control loop to be detected; using a similar waveform matching method to process the process output data End-point continuation; local mean decomposition for improved process output data after end-point continuation; calculation of the regularity index of the autocorrelation function zero crossing point of each decomposed sub-signal; judging whether each regularity index exceeds the set threshold, according to all Judgment result Obtain the detection result. Utilizing the method of the present invention, it is possible to quantitatively detect the time-varying and multi-period oscillation behaviors of industrial control loops, to distinguish components such as time-varying oscillations, multiple oscillations, intermittent oscillations, and non-stationary signals, and to obtain the rules of each oscillation component at the same time. The degree and period provide rich data support for the evaluation of oscillation behavior and diagnosis of fault sources.

Description

A Method for Oscillation Detection of Industrial Control Loops Based on Improved Local Mean Decomposition

technical field

The invention relates to the field of performance evaluation in industrial control, in particular to an industrial control loop oscillation detection method based on improved local mean value decomposition.

Background technique

Modern industrial process equipment has the characteristics of large scale, high comprehensiveness, complex operation, many variables, and long-term operation under closed-loop control. The common industrial chemical production process often contains thousands of control loops, and these control loops affect each other due to the coupling relationship. Due to the ubiquity of controller over-tuning, external disturbance and non-linear operation of control valves in industrial control loops, the oscillation behavior of control loops often occurs, which greatly affects the economic efficiency and stability of industrial process equipment.

Preliminary and accurate vibration detection of industrial process equipment can reduce the amount of waste production, reduce the unqualified rate, improve the reliability and safety of industrial process equipment during operation, and reduce manufacturing costs at the same time. Many controllers can maintain good performance in the initial stage of operation, but as time goes by, due to external interference factors or equipment problems, the performance of the controller will gradually decrease or even fail. The specific manifestation is that various oscillation behaviors occur in the control loop process, which may contain multiple oscillations, intermittent oscillations, nonlinear and other components, which threaten the safe and stable operation of industrial processes.

At the same time, due to the frequent changes in equipment loads and working conditions in real-time environments, industrial processes also exhibit non-stationary data characteristics, specifically manifested in the phenomenon of local mean migration of process data. For important control loops, timely discovery of the oscillation characteristics during its operation will help engineers to diagnose and troubleshoot. Therefore, in the process of industrial control system performance evaluation, effective online monitoring means should be designed to timely and accurately detect various oscillation components of non-stationary process data in the control loop, and distinguish their different frequency ranges. And control loop fault diagnosis are of great significance.

Most of the existing industrial control loop oscillation detection technologies are based on analysis methods of stationary process data. Some oscillation detection methods for non-stationary process data have also appeared in the last two decades. According to its main ideas, it can be roughly classified into three types: analysis methods based on time-domain statistics of process data; analysis methods based on autocorrelation function (ACF) of process data; and signal decomposition methods based on process data (including empirical mode decomposition EMD and basis transform decomposition).

The detection method based on time-domain statistics of process data or autocorrelation function domain analysis has three shortcomings in industrial applications: 1. This method needs to have certain prior knowledge of the detection circuit or process, and some parameters are also determined according to experience; 2. Industrial processes with non-stationary and multiple oscillation cycles cannot realize fully automatic non-intervention detection. It is necessary to design targeted filters for data stabilization and separation of oscillation components; 3. Most detection algorithms cannot quantitatively calculate the regularity of oscillation components.

At present, the signal decomposition method based on process data has improved compared with the above detection methods, but its limitations are mainly reflected in: the number of sub-signal layers decomposed by the existing signal decomposition technology is redundant, resulting in many sub-signals lacking actual physical meaning support. It does not have good representativeness, and the fitting degree of these methods to the trend of non-stationary signals is relatively poor, and the computational complexity is relatively high. In addition, methods based on decomposition techniques cannot deal with components such as intermittent oscillations and time-varying oscillations in process data.

In the practical application of the process oscillation detection algorithm, it can effectively detect whether the industrial control loop has oscillation behavior, and quantitatively evaluate the regularity index of the oscillation behavior, and is generally applicable to time-varying oscillations, intermittent oscillations, non-stationary and nonlinear components. Process data is of great practical significance for accurately diagnosing the existence of industrial process oscillations, and is also conducive to the quantitative evaluation of industrial process control performance.

Contents of the invention

The invention provides an industrial control loop oscillation detection method based on improved local mean value decomposition, which can be applied to industrial control loop processes with time-varying oscillations, multi-period oscillations, etc., and the detection method is generally applicable to non-stationary or stationary process data , only need to obtain routine operation data, no knowledge of process mechanism is required.

An industrial control loop oscillation detection method based on improved local mean decomposition, comprising:

Step 1, collect a group of process output signals of the control loop to be detected;

Step 2, using the similar waveform matching method to extend the endpoint of the process output signal;

Step 3, perform improved local mean value decomposition on the extended process output signal to obtain decomposed sub-signals;

Step 4, calculating the regularity of autocorrelation function zero-crossing intervals of each decomposed sub-signal (Regularity of auto-covariance function zero-crossing intervals);

Step 5, judging whether the regularity index of the zero-crossing points of the respective correlation functions exceeds a threshold, and if it exceeds the threshold, there is oscillation in the decomposition sub-signal corresponding to the control loop.

If there are multiple decomposition sub-signals in the collected process output signal that have oscillation behavior, it is judged that the control loop has multi-period oscillation behavior.

The invention can improve the detection accuracy and reliability of behaviors such as time-varying and multi-period oscillations, and has important practical value in improving economic benefits.

The present invention directly adopts the measurable variable of the chemical process as the process output signal, and the process output signal is obtained through on-site real-time collection, that is, continuously collects and updates the process output signal to the monitoring system as time goes by. Firstly, the collected historical data of the process is processed by endpoint extension, and then the decomposition sub-signal set {x ^k } is obtained by using the improved local mean decomposition, and the regularity of the zero-crossing point of the autocorrelation function corresponding to each decomposition sub-signal x ^k is calculated The calculation complexity of index η ^k and η ^k is extremely small, and it can also be performed on large batches of multiple sets of data at the same time.

As preferably, in step 2, take a section of waveform at the left end of the process output signal, look for the left waveform of the waveform with the highest matching degree with this section waveform in the process output signal, and use the left waveform to extend the left end of the process output signal; Take a waveform at the right end of the output signal, find the right waveform of the waveform with the highest matching degree in the process output signal, and use the right waveform to extend the right end of the process output signal.

Preferably, in step 3, when the improved local mean value decomposition is performed, an adaptive window size selection strategy is used for sliding window averaging, and a two-norm judgment form is used as a judgment standard for pure frequency modulation signals.

The adaptive window size selection strategy means that in each iteration process, the window size can be adjusted according to the characteristics of the output signal of the current process. This strategy allows batch processing of longer time series. Preferably, the adaptive window size Calculated as follows:

$\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; x</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; 10</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}\mathrm{<span\; class="notranslate">\; C</span>}}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ;</span>$

Among them, W represents the window size of the moving average, T _i represents the time interval of all continuous extreme points in the process output signal, T _max = max{T _i }, Respectively represent the mean value of T _i , Indicates the standard deviation of T _i , S _ω is the data length of the current process output signal, n is the magnitude of S _ω , and C is a constant. C defaults to 3.

The judgment standard of pure frequency modulation signal in the iterative process refers to, for the modulation signal S _kn (t) after iterating n times, judge it as the condition of pure frequency modulation signal, preferably, in step 3, the judgment of pure frequency modulation signal The local envelope function with criterion s _kn (t) satisfies the following inequalities:

$\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&le;</span>\sqrt{\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; ;</span>$

Among them, s _kn (t) is the modulated signal after n iterations; a _k(n+1) is the local envelope function of s _kn (t), norm{ } means to take the two-norm, length{x(t )} is the data length of the current process output signal x(t), δ is a constant, and the value range is δ=0.001～0.01.

As preferably, in step 4, the step of calculating the regularity index of the autocorrelation function zero crossing point of each decomposed sub-signal is as follows:

Step 4-1, calculate the autocorrelation function of each decomposed sub-signal x ^k , k is the serial number of decomposed sub-signal;

Step 4-2, count the intervals between all consecutive zero-crossing points in the autocorrelation function Calculate intervals separately mean of and standard deviation

Step 4-3, use the following formula to calculate the regularity index η ^k of the autocorrelation function zero crossing point:

${\mathrm{<span\; class="notranslate">\; \&eta;</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; \&times;</span>\frac{{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}}<span\; class="notranslate">\; .</span>$

Preferably, the threshold in step 5 is 1.

The present invention has the beneficial effect compared with prior art:

1. There is no need for external additional signal excitation, and no additional disturbance will be introduced to the control system, and non-invasive detection and diagnosis can be realized.

2. The calculation complexity is low, it is easy to operate, and the algorithm is easy to write, which is beneficial to implement on the existing DCS workstation or the upper computer of the control system.

3. The signal decomposition method adopted realizes the automatic separation of non-stationary components in the process data. Compared with other existing decomposition techniques, the decomposition efficiency is higher and the calculation complexity is lower.

4. The proposed sliding average window size selection strategy is self-adaptive, and the improved iteration termination condition is more suitable for process control loop oscillation detection.

5. It can detect the time-varying and multi-period oscillation behavior of industrial control loops quantitatively, providing rich data support for the evaluation of the performance of the loops to be tested and the diagnosis of fault sources.

6. Fully adopt data-driven method, no prior knowledge of the process is required, no pre-designed filters are required, and no manual intervention is required.

Description of drawings

Fig. 1 is the schematic flow sheet of chemical process in the embodiment;

Fig. 2 is the process output signal of a group of heating furnace outlet temperature control loops collected in the embodiment;

Fig. 3 is the data after the endpoint extension of the process output signal through the similar waveform matching method in the embodiment;

Fig. 4 is the schematic diagram of process output signal after local mean value decomposition in the embodiment;

Fig. 5 is a flow chart of the method of the present invention.

detailed description

Taking the performance evaluation of the main heating furnace in the delayed coking production process of a large domestic petrochemical enterprise as an example, the time-varying oscillation behavior detection method of the chemical process with the viscous characteristic of the control valve is described in detail.

As shown in Figure 1, the petrochemical process heating furnace is an important link in the production process and one of the main energy-consuming units. The stable control of the furnace outlet temperature is of great significance for improving product quality and reducing energy consumption.

The heating furnace obtains heat through the supply of gas, and the amount of gas fluctuates according to the change of upstream oil properties. It is necessary to control the air intake to make the gas fully combust to obtain the maximum heat. At the same time, a certain air margin should be ensured, but too much low-temperature air will bring The heat in the furnace will lead to waste of fuel and loss of economic benefits. Therefore, the temperature at the outlet of the heating furnace is used as the controlled variable, and the opening of the fuel gas gas is used as the operating variable for loop control. At the same time, there are random disturbances in the process.

The gas opening degree regulating valve (control valve) belongs to the actuator of the control loop. After a period of operation, certain nonlinear characteristics appear. Due to the over-setting of the controller and other reasons, the control loop is prone to continuous oscillation behavior. Moreover, external disturbances can also be introduced into the loop through the coupling loop, causing the loop to oscillate at other frequencies.

As shown in Figure 5, an industrial control loop oscillation detection method based on improved local mean decomposition includes:

Step 1, collect a group of process output signals of the control loop to be tested.

The method of collecting the process output signal is to record the process data in the control loop to be tested in each preset sampling interval, and the process data collected in each sampling interval are added to the end of the previously collected process data .

The sampling interval refers to the sampling interval of the performance evaluation system. The process data is continuously updated over time, and new process data is added to the end of the previously collected process data every time a sampling interval elapses. The sampling interval of the performance evaluation system is generally the same as the control period in the industrial control system, and can also be selected as an integer multiple of the control period, which is determined according to the real-time requirements of performance monitoring and industrial sites and the limitation of data storage capacity.

The process output signal collected in this embodiment is the temperature data at the outlet of the heating furnace when the gas regulating valve is viscous and intermittent external disturbances are introduced. The temperature data at the outlet of the heating furnace after centralization is shown in Figure 2. The abscissa in Figure 2 is the sampling point ordinal, and the unit is Sample (1 Sample corresponds to the sampling interval of one data), and the ordinate is the normal data after centralization. The outlet temperature of the heating furnace under working conditions, in °C.

Step 2, using the similar waveform matching method to extend the endpoint of the process output signal.

The left and right endpoint extension processing is performed on the output signal of the original process to eliminate the influence of the input signal on the endpoint effect of the decomposition method. The specific implementation method is as follows:

Step 2-1, starting from the left endpoint x(t ₀ ) of the process output signal, take a part of the original signal waveform w(t ₀ :t ₀ +l)=[x(t ₀ ),x(t ₀ +1),...x(t ₀ +l)], where l is the maximum value that satisfies w(t ₀ :t ₀ +l) contains only one zero-crossing point;

Step 2-2, find the midpoint of w(t ₀ :t ₀ +l) in [ ] represents the nearest integer in the direction of positive infinity;

Step 2-3, search the next sequence x(t) in the right direction with Points with the same value, denoted as x(t ₁ ), take x(t ₁ ) as the middle point to extract a waveform with the same length as w(t ₀ :t ₀ +l), denoted as w(t ₁ -[l/ 2]:t ₁ +[l/2]-1);

Step 2-4, calculating the waveform matching degree m ₁ of w(t ₀ :t ₀ +l) and w(t ₁ -[l/2]:t ₁ +[l/2]-1). Among them, the waveform matching degree can be based on the existing technology "Zhu Xiaojun, Lu Shiqin, Wang Yanfei, etc. Improved LMD algorithm and its application in EEG signal feature extraction [J]. Journal of Taiyuan University of Technology, 2012,43(3):339- 343." Calculated.

Step 2-5, repeat step 2-3 and step 2-4 until the process output signal search ends, so that a series of matching degrees m=[m ₁ ,m ₂ ,...m _n ] can be obtained;

Step 2-6, find the minimum value m _b in m, then the corresponding band w(t _b -[l/2]:t _b +[l/2]-1) is the same as w(t ₀ :t ₀ + l) The most matching section of the waveform, at this time the left end of the signal x(t) can be extended by w(t _b -l-[l/2]:t _b -[l/2]);

In steps 2-7, use the same method to extend the right endpoint of the signal x(t).

In this embodiment, the extended data of the process output signal is shown in FIG. 3 .

Step 3: Perform improved local mean value decomposition on the extended process output signal to obtain decomposed sub-signals.

The present invention improves the local mean value decomposition in the prior art, retains all the mathematical and computational features of the original method, and only modifies the sliding average window size selection strategy and the judgment standard (termination condition) of the pure frequency modulation signal, For the same process data, the improved decomposition method has stronger decomposition ability, fewer sub-signals and faster iteration time than the original method, and is more suitable for analyzing the oscillation behavior of the original signal. The original local mean decomposition can be performed according to the prior art "SmithJS. The local mean decomposition and its application to EEG perception data [J]. Journal of the Royal Society Interface, 2005, 2(5): 443-454."

In the improved local mean decomposition, the selection strategy of the sliding average window size means that in each iteration, the window size can be adjusted according to the characteristics of the current signal, and this strategy allows batch processing of longer time series. Its specific formula is as follows:

$\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; 10</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}\mathrm{<span\; class="notranslate">\; C</span>}}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ;</span>$

Among them, W represents the window size of the moving average, T _i represents the time interval between all consecutive extreme points in the sequence, T _max = max{T _i }, and respectively represent the mean and standard deviation of T _i , S _ω is the length of the current sequence data, n is the magnitude of S _ω , C is a constant, and the default value is 3.

S _ω /(S _ω +10 ⁿ C) is the contraction factor of the formula, which ensures that the increase of W will not exceed Otherwise the window size will be too large. It can also be seen from the formula that the selection of the window size is also related to the currently considered data length S _ω , the smaller S _ω is, the smaller W is. a larger It often shows that T _i changes significantly, and at this time a smaller S _ω must be selected to ensure that the smaller T _i is not ignored. On the contrary, a larger S ω It shows that a larger S _ω can be selected to ensure the normal progress of the moving average. C is a constant, the default value is 3, but when analyzing fast-changing signals, 3≤C≤5 can achieve better results.

In the improved local mean value decomposition, the criterion for judging the pure frequency modulation signal in the iterative process means that for the modulation signal s _kn (t) after n iterations, the condition for judging it as a pure frequency modulation signal is s _kn (t) The local envelope function satisfies the following inequalities:

$\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&le;</span>\sqrt{\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; ;</span>$

Among them, a _k(n+1) is the local envelope function of s _kn (t), norm{ } means to take the second norm, length{x(t)} is the data length of the current sequence x(t), δ is a constant between 0 and 1, and its usual value range is δ=0.001～0.01.

In this embodiment, an improved local mean value decomposition is performed on the process data x obtained by continuation to obtain a decomposition sub-signal sequence set {x ¹ , x ² , x ³ }, as shown in FIG. 4 .

Step 4, calculating the regularity index of the autocorrelation function zero crossing point of each decomposed sub-signal, the specific steps are as follows:

Step 4-1, calculate the autocorrelation function of each decomposed sub-signal x ^k , k is the serial number of decomposed sub-signal;

Step 4-2, count the intervals between all consecutive zero-crossing points in the autocorrelation function Calculate intervals separately mean of and standard deviation

Step 4-3, use the following formula to calculate the regularity index η ^k of the autocorrelation function zero crossing point:

${\mathrm{<span\; class="notranslate">\; \&eta;</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; \&times;</span>\frac{{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}}<span\; class="notranslate">\; .</span>$

The correlation function zero-crossing regularity index η ^k can be calculated according to the prior art "ThornhillNF, HuangB, ZhangH. Detection of multiple oscillations in control loops [J]. Journal of Process Control, 2003, 13 (1): 91-100." The deep implication of the regularity index of the zero-crossing point of the correlation function is that for a standard oscillation signal, all oscillation waveforms contained in it should have the same wavelength span, so η ^k → ∝, but in actual operation, due to the environment and measurement errors Influenced by unfavorable factors such as η ^k > 1 in this method, it can be determined that there is a high possibility of oscillation in the original signal. In addition, the regularity index of the zero-crossing point of the correlation function is also affected by the number of zero-crossing points in the signal. If the number of zero-crossing points in the original signal is too small, the estimation of the mean and standard deviation may have a large deviation. Therefore, this The invention stipulates that the number of full waves in the original process output signal must be greater than or equal to 11.

In this embodiment, x ³ contains zero crossing points less than 11. When calculating the regularity index of the zero crossing point of the correlation function, this item is discarded, and the remaining x ¹ and x ² corresponding to the auto The index of regularity of zero crossing point of correlation function {η ¹ ,η ² }.

The regularity indices of the zero-crossing point of the autocorrelation function of the two decomposed sub-signals are respectively {η ¹ =0.23,η ² =4.70}. Apparently, the index of regularity of the zero-crossing point of the correlation function obtained by decomposing the sub-signal x ² far exceeds the given threshold Ω=1, so it is certain that the decomposed sub-signal x ² is an oscillation component. For the sub-signal x ¹ decomposed at the first layer, its statistical index value is much smaller than a given threshold, and it should be considered as a sub-signal without oscillation, as shown in FIG. 4 .

Step 5, if one of the correlation function zero-crossing regularity indicators η ^k exceeds the threshold Ω, it is judged that the decomposition sub-signal x ^k corresponding to the control loop is oscillating, if there are multiple decomposition sub-signals in the collected process data Oscillation behavior, it is judged that the control loop has multi-period oscillation behavior. The threshold Ω in this embodiment is 1, that is, when η ^k >1, it indicates that there is an oscillation behavior in x ^k .

Using the method of the present invention, on the basis of single oscillation and multi-period oscillation detection, the time-varying oscillation behavior of the industrial control loop can also be quantitatively detected, and the regularity and period of the time-varying oscillation component can be obtained, which is useful for evaluating and evaluating the oscillation behavior. Fault source diagnosis provides rich data support.

Claims (7)

1. An industrial control loop oscillation detection method based on improved local mean value decomposition, characterized in that, comprising: Step 1, collect a group of process output signals of the control loop to be detected; Step 2, using the similar waveform matching method to extend the endpoint of the process output signal; Step 3, perform improved local mean value decomposition on the extended process output signal to obtain decomposed sub-signals; Step 4, calculating the regularity index of the zero-crossing point of the autocorrelation function of each decomposed sub-signal; Step 5, judging whether the regularity index of the zero-crossing points of the respective correlation functions exceeds the threshold, and if the threshold is exceeded, there is oscillation in the decomposition sub-signal corresponding to the control loop. 2. the industrial control loop oscillation detection method based on improved local mean value decomposition as claimed in claim 1, it is characterized in that, in step 2, get a section of waveform at the left end of process output signal, look for matching with this section waveform in process output signal Use the waveform on the left to extend the left end of the process output signal; take a section of waveform from the right end of the process output signal, find the right waveform of the waveform in the process output signal that matches the waveform with the highest degree, use The right waveform extends the right end of the process output signal. 3. the industrial control loop oscillation detection method based on improved local mean value decomposition as claimed in claim 2, it is characterized in that, in step 3, when carrying out improved local mean value decomposition, adopt adaptive window size selection strategy to carry out sliding window average , the two-norm judgment form is used as the judgment standard of the pure frequency modulation signal. 4. the industrial control loop oscillation detection method based on improved local mean value decomposition as claimed in claim 3, is characterized in that, the adaptive window size calculation formula is as follows: $\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; max</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; 10</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}}{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ;</span>$ Among them, W represents the window size of the moving average, T _i represents the time interval of all continuous extreme points in the process output signal, T _max = max{T _i }, Respectively represent the mean value of T _i , Indicates the standard deviation of T _i , S _ω is the data length of the current process output signal, n is the magnitude of S _ω , and C is a constant. 5. the industrial control loop oscillation detection method based on improved local mean value decomposition as claimed in claim 4, it is characterized in that, in step 3, the judgment standard of pure frequency modulation signal is that the local envelope function of s _kn (t) satisfies the following inequality: $\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; \{</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \}</span><span\; class="notranslate">\; \&le;</span>\sqrt{\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \{</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; \&delta;</span>}<span\; class="notranslate">\; ;</span>$ Among them, s _kn (t) is the modulated signal after n iterations; a _k(n+1) is the local envelope function of s _kn (t), norm{ } means to take the two-norm, length{x(t )} is the data length of the current process output signal x(t), δ is a constant, and the value range is δ=0.001～0.01. 6. the industrial control loop oscillation detection method based on improved local mean value decomposition as claimed in claim 5, is characterized in that, in step 4, the step of calculating the autocorrelation function zero-crossing point regularity index of each decomposed sub-signal is as follows: Step 4-1, calculate the autocorrelation function of each decomposed sub-signal x ^k , k is the serial number of decomposed sub-signal; Step 4-2, count the intervals between all consecutive zero-crossing points in the autocorrelation function Calculate intervals separately mean of and standard deviation Step 4-3, use the following formula to calculate the regularity index η ^k of the autocorrelation function zero crossing point: ${\mathrm{<span\; class="notranslate">\; \&eta;</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 3</span>}}<span\; class="notranslate">\; \&times;</span>\frac{{\mathrm{<span\; class="notranslate">\; \&mu;</span>}}_{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}}<span\; class="notranslate">\; .</span>$ 7. The industrial control loop oscillation detection method based on improved local mean value decomposition as claimed in claim 6, wherein the threshold in step 5 is 1.