Adaptive calculation method for parameter threshold value in activated carbon production process

Abstract

The invention relates to a self-adaptive calculation method for parameter thresholds in an activated carbon production process. In order to solve the problem of false alarm / missing alarm caused by a traditional fixed threshold value, process parameter data are collected in real time, an initial threshold value range is set by combining expert experience after data screening, and internal and external parameters of the threshold value are weighted and reserved respectively. Historical sub-samples are randomly extracted by adopting an isolated tree algorithm to construct a plurality of isolated trees, the path length of data points is calculated, and an abnormal score is obtained based on a normalization factor and a harmonic number. And dynamically tracking real-time abnormal score distribution by using a sliding window, adaptively adjusting an abnormal threshold in combination with a historical standard deviation, and automatically optimizing the number of isolated trees through an abnormal proportion. According to the method, the nonlinear relation among parameters is captured through multi-dimensional segmentation, the calculation precision and performance are dynamically balanced, the method can adapt to raw material difference and process fluctuation, the false alarm rate is remarkably reduced, manual intervention is reduced, and the real-time performance and accuracy of temperature, pressure and other multi-index anomaly detection in the activated carbon production process are improved.

Description

Technical Field

[0001] The present invention relates to a method for adaptively calculating parameter thresholds in the production process of activated carbon, belonging to the technical field of industrial production process control. Background Art

[0002] In the production process of activated carbon, the activated carbon production process involves real-time monitoring of hundreds of production indicators (such as temperature, humidity, gas concentration, pressure, etc.). Due to the complex and rapidly changing production environment, traditional anomaly detection methods usually rely on fixed thresholds set manually to determine whether the data exceeds the normal range. However, without adjustment, since many indicators often show large fluctuations, these fixed thresholds cannot cope with data with large fluctuations, which may result in false alarms or missed alarms, thus affecting the judgment accuracy. If manual adjustment is used, due to environmental changes, the thresholds also need to be set frequently. Facing so many and frequent threshold settings, a large amount of manpower will be consumed. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method for adaptively calculating parameter thresholds in the production process of activated carbon to overcome the deficiencies of the prior art.

[0004] The technical solution of the present invention is: a method for adaptively calculating parameter thresholds in the production process of activated carbon, the method comprising:

[0005] S1. Real-time collect the parameter data of the activated carbon processing technology process;

[0006] S2. Conduct a preliminary screening on the collected real-time data to eliminate the outlier points beyond the sensor capability range;

[0007] S3. Set a preliminary threshold range C for each process parameter according to expert experience or historical data;

[0008] S4. Weight the process parameters within the threshold range C, and keep the process parameters outside the threshold range C unchanged;

[0009] S5. Randomly extract multiple sub-sample sets from the historical data, and each sub-sample set is used to construct multiple isolation trees;

[0010] S6. For each data point, calculate its path length in all isolation trees, and calculate the anomaly score through the average value of the path lengths;

[0011] S7. Use a sliding window to obtain the anomaly score distribution of the real-time data stream, set an anomaly score threshold D according to the anomaly score distribution, and when the anomaly score of a certain data point exceeds D, mark it as an anomaly.

[0012] Further, the calculation method of the anomaly score threshold D is as follows:

[0013]

[0014] Among them, D t is the anomaly score threshold at time t, D t-1 is the anomaly score threshold at time t-1, and α is the adjustment coefficient, and σ t is the standard deviation of the anomaly score distribution at time t.

[0015] Furthermore, the construction method of the isolation tree is as follows: by randomly selecting a splitting value at each tree node,

[0016] dividing the data into a left subtree and a right subtree according to this value until the tree reaches a preset height.

[0017] Furthermore, the calculation method of the anomaly score is as follows:

[0018]

[0019] Among them, c(n) is the normalization factor for eliminating the influence of the sample size n, H(n-1) is the harmonic number, h i (x) is the path length of the i-th tree, F is the number of isolation trees, and E(h(x)) is the average value of the path lengths of F isolation trees.

[0020] Furthermore, the number F of the isolation trees is calculated according to the following method,

[0021] Statistically, calculate the proportion H of the anomaly score values exceeding the set threshold E in the sliding window at the previous moment t-1 ;

[0022] Calculate the average value of the proportion of the anomaly score values exceeding the set threshold E before the previous moment,

[0023]

[0024] Update the number F of the isolation trees,

[0025] F t = F t-1 + β * [H t-1 - E(H t-2 )]

[0026] Among them, F t represents the number of isolation trees at time t, F t-1 represents the number of isolation trees at time t-1, and β is the adjustment coefficient of the number of isolation trees.

[0027] The beneficial effects of the present invention are:

[0028] 1) In this invention, this patent calculates the abnormal score distribution threshold in real time through a sliding window, and dynamically adjusts the threshold in combination with the historical standard deviation, which can effectively reduce the false alarm / missing alarm rate caused by natural fluctuations of parameters. At the same time, only an initial setting is required throughout the process. Using the initial threshold and the collected data, the threshold can be adjusted in a timely manner according to the real-time situation without manual adjustment, which can save a lot of manpower;

[0029] 2) This invention constructs isolation trees by randomly sampling sub-samples and calculates the abnormal score in combination with the path length, which can identify the non-linear relationship between multi-dimensional parameters. For example, the co-fluctuation of carbonization temperature and activation temperature in the production of activated carbon can be accurately captured by this method;

[0030] 3) This invention sets the initial threshold range C according to expert experience, retains key parameters by weighting, and dynamically expands the threshold of secondary parameters at the same time, which can adapt to the differences of different raw materials (such as coal, coconut shell) or process scenarios (physical / chemical activation);

[0031] 4) This invention automatically increases or decreases the calculation of isolation trees with large computational consumption by statistically counting the historical abnormal proportion, and realizes the adaptive balance of accuracy and performance according to the environment;

[0032] 5) This invention uses a sliding window mechanism to analyze real-time data streams, which can quickly respond to production fluctuations and avoid delays caused by batch processing of long-term data;

[0033] 6) This invention reduces the dependence on fixed expert rules by dynamically adjusting the threshold and model parameters;

[0034] 7) This invention aims at multi-link parameters (such as ash content ≤ 3%, pressure loss ≤ 2.5 kPa) in the carbonization, activation, crushing, etc. of activated carbon production. Through the multi-dimensional segmentation ability of isolation trees, it can identify complex anomalies (such as the decline in adsorption performance caused by the superposition of high temperature and high humidity), which is superior to the single-index threshold detection method. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is the flowchart of this invention. DETAILED DESCRIPTION OF THE INVENTION

[0036] Example 1: Refer to Figure 1 , an adaptive calculation method for the threshold of activated carbon production process parameters, the method includes:

[0037] S1. Real-time collect the data of activated carbon processing process parameters; for example, deploy temperature sensors, pressure sensors and gas concentration detectors in the activated carbon activation converter (such as the outer ring activation furnace chamber) to collect parameters such as temperature, pressure, and CO2 concentration in real time. Abnormal values beyond the sensor range are removed through a data acquisition module (such as a PLC controller).

[0038] S2. Preliminary screening of the collected real-time data to remove abnormal value points that exceed the sensor capability range;

[0039] S3. Set a preliminary threshold range C for each process parameter based on expert experience or historical data;

[0040] S4, weighting the process parameters within the threshold range C, and keeping the process parameters outside the threshold range C unchanged;

[0041] S5. Randomly extract multiple sub-sample sets from the historical data, and each sub-sample set is used to construct multiple isolated trees; for example, randomly extract sub-sample sets (such as the last 30 batches of data) from the historical data, and each sub-sample set contains 5,000 data points.

[0042] S6. For each data point, calculate its path length in all isolated trees, and calculate the anomaly score by the average of the path lengths;

[0043] S7. Use a sliding window to obtain the anomaly score distribution of the real-time data stream, set the anomaly score threshold D according to the anomaly score distribution, and mark a data point as abnormal when the anomaly score exceeds D.

[0044] Furthermore, the calculation method of the anomaly score threshold D is as follows:

[0045]

[0046] Among them, D t is the anomaly score threshold at time t, D t-1 is the abnormal score threshold at time t-1, α is the adjustment coefficient, σ t is the standard deviation of the anomaly score distribution at time t.

[0047] Furthermore, the isolation tree is constructed by randomly selecting a split value at each tree node and dividing the data into a left subtree and a right subtree according to the value until the tree reaches a preset height.

[0048] Furthermore, the calculation method of the anomaly score is:

[0049]

[0050] Among them, c(n) is the normalization factor to eliminate the influence of sample size n. H(n-1) is the harmonic number, h i (x) is the path length of the ith tree, F is the number of isolated trees, and E(h(x)) is the average path length of the F isolated trees.

[0051] Furthermore, the number F of isolated trees is calculated according to the following method:

[0052] Statistically calculate the proportion H of abnormal score values exceeding the set threshold E in the sliding window at the previous moment. t-1 ;

[0053] Calculate the average value of the proportion of abnormal score values exceeding the set threshold E before the previous moment.

[0054]

[0055] Update the number F of isolation trees.

[0056] F t = F t-1 + β * [H t-1 - E(H t-2 )]

[0057] Among them, F t represents the number of isolation trees at time t, and F t-1 represents the number of isolation trees at time t - 1, and β is the adjustment coefficient of the number of isolation trees.

[0058] The above content is a further detailed description of the present invention in combination with specific preferred embodiments. It cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention belongs, without departing from the concept of the present invention, several simple deductions or substitutions can be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. An adaptive calculation method for parameter thresholds in the production process of activated carbon, characterized in that, The method includes: S1. Collecting process parameter data of the activated carbon processing technology in real time; S2. Conducting a preliminary screening of the collected real-time data to eliminate the outlier points beyond the sensor's capacity range; S3. Setting a preliminary threshold range C for each process parameter according to expert experience or historical data; S4. Weighting the process parameters within the threshold range C, and keeping the process parameters outside the threshold range C unchanged; S5. Randomly extracting multiple sub-sample sets from historical data, and each sub-sample set is used to construct multiple isolation trees; S6. For each data point, calculating its path length in all isolation trees, and calculating the anomaly score through the average value of the path lengths; S7. Using a sliding window to obtain the anomaly score distribution of the real-time data stream, setting an anomaly score threshold D according to the anomaly score distribution, and when the anomaly score of a certain data point exceeds D, marking it as an anomaly.

2. The self-adaptive calculation method for the threshold of the activated carbon production process parameters according to claim 1, characterized in that, The calculation method of the anomaly score threshold D is as follows: Among them, D t is the anomaly score threshold at time t, D t-1 is the anomaly score threshold at time t-1, α is the adjustment coefficient, and σ t is the standard deviation of the anomaly score distribution at time t.

3. The self-adaptive calculation method for the threshold of activated carbon production process parameters according to claim 1, characterized in that, The construction method of the isolation tree is: randomly selecting a splitting value at each tree node, and dividing the data into a left subtree and a right subtree according to this value until the tree reaches a preset height.

4. The self-adaptive calculation method for the threshold of activated carbon production process parameters according to claim 1, characterized in that, The calculation method of the anomaly score is: where c(n) is a normalization factor that eliminates the influence of the sample size n, H(n - 1) is the harmonic number, h i (x) is the path length of the i-th tree, F is the number of isolated trees, and E(h(x)) is the average path length of F isolated trees.

5. The self-adaptive calculation method of the activated carbon production parameter threshold according to claim 4, characterized in that, The number F of the isolation trees is calculated according to the following method, Statistically calculate the proportion H of abnormal score values exceeding the set threshold E in the sliding window at the previous moment t-1 ; Calculating the average value of the proportion of the anomaly score values exceeding the set threshold E before the previous moment, Updating the number F of the isolation trees, F t = F t-1 + β * [H t-1 - E(H t-2 )] Among them, F t represents the number of isolated trees at time t, and F t-1 represents the number of isolated trees at time t - 1, and β is the adjustment coefficient of the number of isolated trees.

Images