Production safety early warning method based on artificial intelligence
By dynamically adjusting the limit threshold and far exceeding the upper limit threshold of refining and chemical equipment, the problem of sudden numerical changes caused by equipment loss affecting the accuracy of the prediction model is solved, and the accuracy and robustness of production safety warnings of refining and chemical enterprises are improved.
Patent Information
- Application Number
- CN202310983320.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-04
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2043-08-04
AI Technical Summary
In the production workshop of a refining and chemical enterprise, sudden changes in numerical values caused by equipment loss participate in the prediction function process, resulting in inaccurate prediction model and inaccurate alarms for data exceeding limits, which affects the risk assessment score.
By obtaining the preliminary fluctuation evaluation value, discrete deviation value and deviation of the loss degree, combining the loss distribution characteristics within the time window, adjusting the limit threshold and far exceeding the upper limit threshold, dynamically adjusting the evaluation points for risk control of process technical parameters, and then conducting early warning .
Improve the accuracy of risk monitoring, reduce data fluctuations caused by operations, enhance the robustness of iterative evaluation, and ensure that the predicted data reflects actual information.
Smart Images

Figure CN117010690B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a production safety early warning method based on artificial intelligence. Background Art
[0002] In the production workshop of a refining enterprise, the setting of parameters is very particular. The instability of parameters will not only lead to poor quality and low efficiency of the refining products produced, but also because the materials it targets are usually materials such as petroleum, which will produce harmful gases if improperly handled, unreasonable handling will also cause certain dangers to employees in the workshop. Therefore, the present invention proposes to evaluate the risks in various places in the production workshop by setting eight major items and 90 risk factors, so as to control the risk situation every day. Among the eight major risk factors, except for the process technology parameter control risk, the other risk factors are directly recorded through manual inspection, and the process technology parameter control risk is mainly to divide the upper and lower limits of risk determination of important monitoring data in the production process, and obtain the risk assessment value of this type of risk factor according to the number and time of exceeding the upper and lower limits. However, since the refining equipment will be continuously damaged during the production process, and the degree of equipment damage corresponds to the size of the risk hidden dangers to a certain extent. Therefore, fixed upper and lower limit thresholds often cannot obtain better monitoring results. Therefore, the present invention monitors the loss data of the equipment, and then dynamically adjusts the upper and lower limits of each specific parameter in the process technology parameter control risk, so as to achieve more accurate risk monitoring. When the loss degree changes greatly, leakage and other events may have occurred. However, in addition to this, manual water injection and drainage operations may also cause sudden changes in the loss degree. If these suddenly changing values are involved in the process of obtaining the prediction function, this will cause the obtained prediction function to be inaccurate, and thus make it impossible to accurately alarm for data exceeding the limit, resulting in the obtained evaluation score being affected. Summary of the invention
[0003] The present invention provides a production safety early warning method based on artificial intelligence to solve the problem that the sudden change of numerical values involved in the process of obtaining the prediction function will lead to inaccurate prediction models. The technical solutions adopted are as follows:
[0004] An embodiment of the present invention provides a production safety early warning method based on artificial intelligence, the method comprising the following steps:
[0005] The safety risk assessment value before regulation is obtained according to the over-limit threshold and far-exceeding upper limit threshold of different parameters before regulation;
[0006] Obtain the loss degree at different time points, and obtain the preliminary fluctuation assessment value at each time point according to the loss degree at different time points; set a time window with each time point as the center point, and obtain the discrete deviation value at each time point according to the loss degree at the time point as the center point and the loss degree of the time points within the time window of the time point as the center point;
[0007] The deviation of each time point in the time window is obtained according to the loss degree of the time point and the difference in loss degrees between adjacent time points in the time window; the discrete distribution characteristics of the time window corresponding to each time point are obtained according to the deviation and loss degree of all time points in the time window corresponding to each time point; the actual single dispatch of the data point corresponding to each time point is obtained according to the discrete deviation value of each time point and the discrete distribution characteristics corresponding to the time point; the final fit of the data point corresponding to each time point is obtained according to the actual single dispatch and the preliminary fluctuation assessment value, and the loss coefficient is obtained according to the final fit to adjust the over-limit threshold and the far-exceeding upper limit threshold to obtain the regulated over-limit threshold and the far-exceeding upper limit threshold;
[0008] The assessment score of each parameter is obtained based on the over-limit threshold and the far-exceeding upper limit threshold after regulation, and the assessment score of each parameter is accumulated to obtain the assessment score of the process technology parameter control risk. The assessment score belonging to the process technology parameter control risk in the safety risk assessment value before regulation is replaced by the assessment score after regulation to obtain the safety risk value after regulation, and an early warning is issued based on the safety risk value.
[0009] Preferably, the method for obtaining the safety risk assessment value before regulation based on the over-limit threshold and far-exceeding upper limit threshold of different parameters before regulation is:
[0010] There are eight types of risks in the production workshop, namely inherent risks, hazard source risks, process technology parameter control risks, on-site operation risks, safety risk inspections, hidden danger investigation quality, LDAR risks and other risks;
[0011] Obtain assessment scores for seven types of risks other than process technology parameter control risk, obtain assessment scores for process technology parameter control risk based on over-limit thresholds and far-exceeding upper limit thresholds of different parameters before regulation, and construct a safety assessment model to obtain a safety risk assessment value before regulation. The safety assessment model refers to adding the assessment values obtained for eight types of risks.
[0012] Preferably, the method for obtaining the preliminary fluctuation assessment value at each time point according to the loss degree at different time points is:
[0013] The loss degrees of all time points constitute a loss curve. The horizontal axis of the loss curve is the time point, and the vertical axis is the loss degree. The points on the loss curve are data points. The loss degree corresponding to each time point is subtracted from the loss degree corresponding to the previous time point, and the difference is used as the preliminary fluctuation assessment value of the time point.
[0014] Preferably, the method for obtaining the discrete deviation value of each time point according to the loss degree of the time point as the center point and the loss degree of the time points within the time window of the time point as the center point is:
[0015] Get the median of the loss degree of all time points in the time window of each time point, find the maximum and minimum values of the loss degree in the time window, and make the difference between the maximum and minimum values of the loss degree to get the limit range of each time window. Get the discrete deviation value of each time point based on the difference between the loss degree of the time point and the median of the loss degree of the corresponding time window and the limit range in the time window.
[0016] Preferably, the method for obtaining the discrete deviation value of each time point according to the difference between the loss degree at the time point and the median of the loss degree in the corresponding time window and the restricted range in the time window is:
[0017]
[0018] In the formula, a t Indicates the degree of loss corresponding to the t-th time point, a mid(t) represents the median loss of the time window at the tth time point, Δa (t) represents the limit range of the time window at the tth time point, norm() represents the linear normalization function, and p t Represents the discrete deviation value at the tth time point.
[0019] Preferably, the method for obtaining the deviation degree of each time point in the time window according to the loss degree of the time point and the difference in loss degrees of adjacent time points in the time window is:
[0020] The time point within the time window corresponding to each time point is recorded as the window time point, and the data points corresponding to the two window time points at both ends of the time window are connected. The connection line is recorded as the distribution direction function, and each window time point obtains a loss pseudo value on the distribution direction function. The deviation of the window time point is obtained according to the difference between the loss degree of the window time point and the loss pseudo value and the difference in loss degree between adjacent window time points.
[0021] Preferably, the method for obtaining the deviation of the window time point according to the difference between the loss degree at the window time point and the loss hypothetical value and the difference between the loss degrees at adjacent window time points is:
[0022] For each window time point in the time window corresponding to the time point, the difference in loss degree between the window time point and the previous window time point is recorded as the first difference, the difference in loss degree between the window time point and the next window time point is recorded as the second difference, and the difference in loss degree between the window time point before the window time point and the window time point after the window time point is recorded as the third difference. The sum of the first difference and the second difference is divided by the third difference to obtain the first ratio. The difference between 1 and the first ratio is normalized and recorded as linearity. The normalized value of the absolute value of the difference between the loss degree of each window time point and the loss hypothetical value is recorded as the difference value. The difference value of each window time point is multiplied by the linearity to obtain the deviation of the window time point.
[0023] Preferably, the method for obtaining the discrete distribution characteristics of the time window corresponding to each time point according to the deviation and loss degree of all time points in the time window corresponding to each time point is:
[0024]
[0025] In the formula, a t,n Indicates the degree of loss at the nth time point in the time window corresponding to the tth time point, represents the mean of the loss degree of all time points in the time window corresponding to the tth time point, q t,n Indicates the deviation of the nth time point in the time window corresponding to the tth time point, N t represents the length of the time window at the tth time point, δ t Represents the discrete distribution characteristics of the time window corresponding to the t-th time point.
[0026] Preferably, the method for obtaining the actual single scheduling of the data point corresponding to each time point according to the discrete deviation value of each time point and the discrete distribution characteristics corresponding to the time point is:
[0027] The product of the discrete deviation value corresponding to each time point and the discrete distribution feature is used as the actual single dispatch of the data point corresponding to the time point.
[0028] Preferably, the method of adjusting the over-limit threshold and the far-exceeding upper limit threshold by obtaining the loss coefficient according to the final fitting degree to obtain the regulated over-limit threshold and the far-exceeding upper limit threshold is:
[0029]
[0030]
[0031] In the formula, Indicates the exceeding threshold value, Indicates that it is far beyond the upper threshold. Indicates the over-limit threshold after regulation, It means that the value after regulation far exceeds the upper limit threshold, and K represents the loss coefficient.
[0032] The beneficial effect of the present invention is that after fitting the collected data by the least squares method, the present invention performs iterative evaluation through the mean square error to find the optimal fitting parameters. However, in actual scenarios, the actual meaning of some data is not really needed, that is, the reference value is not high, so the reference degree of some data when participating in the mean square error is not high. Therefore, the present invention combines the characteristics of the loss degree data, analyzes its local data characteristics, calculates the distribution trend characteristics of multiple consecutive data in the window, and then calculates the deviation degree actually shown by each data. Finally, the reference degree of the data itself at each moment is obtained, and it is used as the reference degree when it actually participates in the fitting evaluation. While enhancing the robustness of the iterative evaluation, the error caused by the data fluctuation caused by the operation is reduced as much as possible, thereby ensuring that the predicted data obtained can reflect the actual information in the scene. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0034] Figure 1 A flowchart of a production safety early warning method based on artificial intelligence provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0036] See also Figure 1 , which shows a flow chart of a production safety early warning method based on artificial intelligence provided by an embodiment of the present invention, the method comprising the following steps:
[0037] Step S001, construct a safety risk assessment value model and obtain the safety risk assessment value before regulation.
[0038] This embodiment is aimed at the production workshop of a refinery. There are eight types of risks in the production workshop, namely inherent risks, hazard source risks, process and technical parameter control risks, on-site operation risks, safety risk inspections, hidden danger investigation quality, LDAR risks and other risks. Among them, the inherent risks, hazard source risks, on-site operation risks, safety risk inspections, hidden danger investigation quality, and LDAR risks need to be obtained through manual inspections. For example, the number of production units in the inherent risks is +0.5 points for each additional set and -1 point for each reduced set. The number of each hazard source in the hazard source risk is +0.5 points, etc. The assessment scores for these types of risks are obtained every day and recorded as D. Other risks include weather data such as temperature. The weather forecast for the day is extracted through the Internet. The score F is obtained based on the different wind speeds, temperatures, humidity and weather conditions of the day. The wind speed level 1 or below is 0 points, and each increase of 1 level of wind speed is 0 points. 1 point is added for strength, 2 points are added for temperature less than -20 degrees Celsius, no points are added for -20 to 30 degrees Celsius, 2 points are added for above 30 degrees Celsius, no points are added for humidity greater than or equal to 90%, and 0.5 points are added for every 5% decrease. The weather conditions include sunny, less than, moderate rain, heavy rain, rainstorm, and thunderstorm, and the weather conditions add 0 points, 0.5 points, 1 point, 1.5 points, 2 points, and 2.5 points respectively; and the process technology parameter control risk is obtained by monitoring multiple parameters with relevant monitoring equipment. For example, if the parameter is toxic and harmful gas, the detection equipment is used to obtain the index of toxic and harmful gas, and one or two thresholds are set for the index, which are the over-limit threshold and far exceeds the upper threshold In this embodiment, the two thresholds of toxic and harmful gases are 5% IDLH and 10% IDLH, respectively. The size of the threshold is artificially defined according to the historical data. When the indicator exceeds the over-limit threshold but does not exceed the far-exceeding upper limit threshold, an alarm is given +0.1 points. If it exceeds the far-exceeding upper limit threshold, an alarm is given +0.2 points. In addition, points are added according to the time of eliminating the alarm after the alarm. If the alarm is eliminated within one minute, no points are added. After that, for every minute that exceeds the limit threshold, an alarm is given +0.1 points for exceeding the limit threshold and +0.2 points for far-exceeding the upper limit threshold. The safety risk assessment model is obtained based on the number of alarms and the time to eliminate the alarm obtained from all different types of indicators and combined with different types of risks:
[0039]
[0040] Where D represents the total assessment score of inherent risk, hazard source risk, on-site operation risk, safety risk inspection, hidden danger investigation quality, and LDAR risk acquisition, F represents the assessment score of other risks, and x i It represents the score obtained by the i-th parameter in the process technology parameter control risk, I represents the number of parameters in the process technology parameter control risk, and Y is the final safety production risk assessment value.
[0041] At this point, the safety risk assessment value before regulation is obtained.
[0042] Step S002, obtaining the degree of loss, obtaining a preliminary evaluation value according to the degree of loss, and obtaining a discrete deviation value of the time point according to the degree of loss at each time point and the degree of loss at surrounding time points.
[0043] In the production workshop, pipelines and equipment will be worn out during the refining process. When pipelines and equipment are worn out, even if the same toxic and harmful gas is produced, the performance will be reduced due to equipment wear, making it impossible to process the gas in time. If there is no alarm at this time, it is very likely to cause safety hazards. Therefore, the over-limit threshold is set according to the degree of wear of different equipment. and far exceeds the upper threshold To regulate.
[0044] The probe is placed on the equipment, pipeline and other locations that need to be monitored for monitoring. Every H hours, the probe is used to collect the loss level at the current moment. In this embodiment, H is 3.
[0045] The commonly used linear regression calculation for the loss degree monitored by the probe is to use the loss degree in each unit time as a data point, fit the functional relationship between it and time through the least squares method, and establish the relationship between the loss degree and time, so as to help technicians make a certain estimate of the loss degree in the future and take timely measures. However, in actual scenarios, the data monitored by the probe will be interfered by some special reasons. In addition to noise data, there are also data interference caused by some normal scenarios, such as normal water injection and drainage, and uneven loss on the probe's own surface, which causes data distortion. In this case, if it is regarded as normal data and given the same fitting reference, the abnormal volatility of some data points will affect the fitting direction and cause deviation, which will lead to inaccurate output results and interfere with the judgment of the monitoring personnel.
[0046] If the probe is not disturbed by external factors, then its loss degree should show a linear trend over time. The loss degree collected at each time point constitutes a loss curve. The horizontal axis of the loss curve is the time point, and the vertical axis is the loss degree. Each point on the curve is a data point. The loss curve is highly volatile. At this time, the difference between the loss degree corresponding to each time point and the loss degree corresponding to the previous time point is obtained as the preliminary fluctuation assessment value. The larger the preliminary fluctuation assessment value, the less it looks like the fluctuation characteristics of normal loss data. The larger the preliminary fluctuation assessment value, the more likely the time point is to be caused by monitoring data fluctuations caused by water injection and drainage, rather than the fluctuation characteristics of the time point itself, so it is less referenceable.
[0047] Water injection and drainage will cause changes in the probe surface environment. When the environment on the probe surface stabilizes, it will return to the normal range. At this time, under these influences, the loss degree collected by the probe shows a monotonic change trend. If the collection time point is after drainage and water injection, the loss degree will have a large defect. At this time, if the fluctuation degree of the loss curve is directly obtained, it is inaccurate, and it is greatly affected by the water injection and drainage behaviors. Since the probe is inductively monitored, the surface medium needs to be stable. Water injection and drainage will cause changes in the environment of the probe, so it gradually presents a monotonic change feature, which corresponds to the reaction time of the probe and the time for the corrosion inhibitor to form a film on the probe surface. After stabilization, it will instantly return to normal. Therefore, a window is set to obtain the fluctuation characteristics caused by the operation through local features.
[0048] Specifically, a time window is set with each time point as the center, and the length of the time window is 2B+1, that is, B time points are selected on the left side of the center point, and B time points are selected on the right side. In this embodiment, B is 8, and combined with the center point to form a time window. The maximum and minimum values of the loss degree in each time window are obtained, and the limit range in each time window is obtained by calculating the difference between the maximum and minimum values of the loss degree. The median value of the loss degree corresponding to all time points in the time window is obtained and recorded as the median loss value. The discrete deviation value of each time point is obtained according to the limit range in the time window and the difference between the loss degree at each time point and the median loss value. The formula is as follows:
[0049]
[0050] In the formula, a t Indicates the degree of loss corresponding to the t-th time point, a mid(t) represents the median loss of the time window at the tth time point, Δa (t) represents the limit range of the time window at the tth time point, norm() represents the linear normalization function, and p t Represents the discrete deviation value at the tth time point.
[0051] It represents the difference between the t-th time point and the median loss of the time window within the time window. Because it is located in the center of the window, if it is close to the median of the window value range, the loss degree change trend in the time window obtained at this time point gradually shows a monotonic change. At the same time, the normalized limit range is used as its stretching degree. That is, after determining the time window range, the wider the limit range, the more its overall discrete distribution tends to be distributed in a monotonic direction. That is, at this time, both features are quantifications of the data distribution characteristics within the window. Since the logical directions of these two features are opposite, the previous feature is subtracted from 1. After changing its logical direction, finally, when p tThe larger the value is, the more the overall distribution of data points in the time window tends to change monotonically. Whether it increases or decreases, it does not conform to the up and down fluctuation characteristics of normal loss degree data. Therefore, the data at this moment is more likely to be in the fluctuation characteristics caused by operation.
[0052] At this point, the discrete deviation value at each time point is obtained.
[0053] Step S003, obtain the deviation of the time point according to the difference in loss degree between the time point and its adjacent time points, obtain the discrete distribution characteristics of the time window according to the deviation of the time point, and obtain the final fit of the data point in combination with the discrete deviation value, and obtain the adjusted threshold value according to the final fit.
[0054] Since the size of the window is fixed, while the duration of injection and drainage is different, it is impossible to confirm that the fluctuation characteristics within the window are all caused by one operation. That is to say, even if the fluctuation characteristics of the center point of the time window are caused by operations such as injection and drainage, the degree of loss within the time window is not increasing or decreasing. There are normal time points in the time window, as well as time points with complex fluctuation characteristics caused by operations.
[0055] Based on the above reasons, for each time window, the data points corresponding to the time points at both ends of each time window are connected, and the connecting line is called the distribution direction function.
[0056] For each time point, a function value is obtained on the distribution direction function. The function value is recorded as the loss pseudo value. According to the difference in the loss degree between each time point and the adjacent time points in the time window and the difference between the loss pseudo value and the loss degree of each time point, the deviation degree of each time point in the time window is obtained. The formula is as follows:
[0057]
[0058] In the formula, Δa t (n,n-1) represents the difference in the loss degree between the nth time point and the n-1th time point in the time window corresponding to the tth time point, Δa t (n,n+1) represents the difference in the loss degree between the nth time point and the n+1th time point in the time window corresponding to the tth time point, Δa t (n+1,n-1) represents the difference in the loss degree between the n-1th time point and the n+1th time point in the time window corresponding to the tth time point, b t (n) represents the hypothetical loss value at the nth time point in the time window corresponding to the tth time point, a t (n) represents the loss degree of the nth time point in the time window corresponding to the tth time point, norm() represents the linear normalization function, q t,nIndicates the deviation of the nth time point in the time window corresponding to the tth time point.
[0059] like The closer it is to 1, the closer the three data points corresponding to the time point and the adjacent time points are to a straight line, that is, the more monotonic the feature is. Therefore, after subtracting 1 here, the smaller the value is, the closer it is to 0, indicating that the deviation of the time point is smaller, b t (n)-a t The smaller the value of (n), the more it represents that the data points around it are monotonous, indicating that the deviation of the time point is smaller.
[0060] The discrete distribution characteristics of the time window corresponding to the time point are obtained based on the calculated deviation combined with the standard deviation of the loss degree in the time window. The formula is as follows:
[0061]
[0062] In the formula, a t,n Indicates the degree of loss at the nth time point in the time window corresponding to the tth time point, represents the mean of the loss degree of all time points in the time window corresponding to the tth time point, q t,n Indicates the deviation of the nth time point in the time window corresponding to the tth time point, N t represents the length of the time window at the tth time point, δ t Represents the discrete distribution characteristics of the time window corresponding to the t-th time point.
[0063] Where (1-q t,n ) is the normal standard deviation calculation formula. The larger the standard deviation, the more discrete the overall distribution of the time points, and the weaker the fluctuation of the loss curve. After combining the deviation of the time points with the formula, the time points closer to the monotonic features have higher weights. When the discrete distribution features are larger, the distribution within the time window presents more monotonic features, and the fluctuation features of the time points are weaker. When participating in the subsequent fitting, the reference value of this point is poor.
[0064] The actual single dispatch of the data point corresponding to each time point is obtained according to the discrete deviation value and discrete distribution characteristics obtained at each time point. The formula is as follows:
[0065] m t =δ t ×p t
[0066] In the formula, p t represents the discrete deviation value at the tth time point, δ t Represents the discrete distribution characteristics of the time window corresponding to the t-th time point, m tRepresents the actual single schedule of the data point corresponding to the t-th time point.
[0067] where δ t and p t Both express the distribution characteristics of the data in the window at time t. When these two values are larger, their logical order is the same, and both represent that the distribution in the window is more monotonic. At the same time, their value range is 0 to 1, and these two characteristics are mutually regulated. t When the value is larger, the data points in the corresponding time window are closer to the data fluctuation characteristics caused by operations such as water injection and drainage, that is, they are not normal loss change characteristics, so they have less reference value in the subsequent period.
[0068] The final fit is obtained based on the preliminary fluctuation assessment value obtained at each time point and the actual single dispatch of the corresponding data point. The formula is as follows:
[0069] k t =1-norm(Δa t )×norm(m t )
[0070] In the formula, Δa t represents the preliminary fluctuation assessment value at the tth time point, m t represents the actual single dispatch of the data point corresponding to the t-th time point, norm() represents the linear normalization function, k t Indicates the final fit of the data point corresponding to the tth time point. The larger the final fit, the higher the reference when using the least squares method for fitting.
[0071] The final fitting degree of the data point corresponding to each time point obtained is taken as the confidence degree. The principle of the least squares method is to minimize the distance from each data point to the fitting straight line. When using the least squares method, the confidence degree and the distance to the fitting straight line of all data points are minimized. The fitting straight line obtained at this time is the straight line fitted by the least squares method. This straight line is recorded as the loss degree assessment model. The risk value of the next time unit is obtained according to the loss degree assessment model. The risk value is normalized according to the historical data to obtain the loss coefficient K corresponding to each day. The over-limit threshold is determined according to the loss coefficient. and far exceeds the upper threshold Make adjustments.
[0072] Calculate the over-limit threshold and far exceeds the upper threshold After adjustment, first obtain the adjustable range. Even if the equipment loss is large, its performance will not approach 0, so let use Indicates that the adjustable range is the last formula, let use Indicates that the same control range is the last item, thus obtaining the adjusted threshold value:
[0073]
[0074]
[0075] In the formula, Indicates the exceeding threshold value, Indicates that it is far beyond the upper threshold. Indicates the over-limit threshold after regulation, It means that the value after regulation far exceeds the upper limit threshold, and K represents the loss coefficient.
[0076] At this point, the over-limit threshold after regulation and the far-exceeding upper limit threshold after regulation are obtained.
[0077] Step S004, obtaining the assessment score of the process technology parameter control risk according to the adjusted threshold value, combining the assessment score with the safety assessment model to obtain the adjusted safety risk value, and issuing an early warning.
[0078] For the process technology parameter control risk, the over-limit threshold and the far-exceeding upper limit threshold after adjustment of each parameter are obtained. For each parameter, a parameter value is obtained every minute, and the parameter value is compared with the over-limit threshold and the far-exceeding upper limit threshold after adjustment. When the parameter value exceeds the over-limit threshold but does not exceed the far-exceeding upper limit threshold, an alarm +0.1 point is given; if it exceeds the far-exceeding upper limit threshold, an alarm +0.2 point is given, and points are added based on the time it takes to eliminate the alarm after the alarm. No points are added if the alarm is eliminated within one minute, and thereafter, for every minute that passes, an over-limit threshold alarm +0.1 point is given, and an far-exceeding upper limit threshold alarm +0.2 point is given. In this way, the evaluation score corresponding to each parameter is obtained, and the evaluation scores of all parameters are added up to obtain the evaluation score corresponding to the process technology parameter control risk.
[0079] Thus, the evaluation value after process technology parameter control risk adjustment is obtained, and the evaluation values of the other seven categories are added to the evaluation value of this category to obtain the total score of the production workshop. Enter the date, the level of the production workshop, and the ID of the production workshop to obtain the safety risk value of the production workshop after adjustment on the input date.
[0080] According to the total score classification of the production workshop, if Y<125, it is a low risk and a blue warning is issued; if 125≤Y<150, it is a general risk and a yellow warning is issued; if 150≤Y<175, it is a greater risk and an orange warning is issued; if 175≤Y, it is a high risk and a red warning is issued.
[0081] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. The production safety early warning method based on artificial intelligence is characterized by: The method comprises the following steps: The safety risk assessment value before regulation is obtained according to the over-limit threshold and far-exceeding upper limit threshold of different parameters before regulation; Obtain the loss degree at different time points, and obtain the preliminary fluctuation assessment value at each time point according to the loss degree at different time points; set a time window with each time point as the center point, and obtain the discrete deviation value at each time point according to the loss degree at the time point as the center point and the loss degree of the time points within the time window of the time point as the center point; The deviation of each time point in the time window is obtained according to the loss degree of the time point and the difference in loss degrees between adjacent time points in the time window; the discrete distribution characteristics of the time window corresponding to each time point are obtained according to the deviation and loss degree of all time points in the time window corresponding to each time point; the actual single dispatch of the data point corresponding to each time point is obtained according to the discrete deviation value of each time point and the discrete distribution characteristics corresponding to the time point; the final fit of the data point corresponding to each time point is obtained according to the actual single dispatch and the preliminary fluctuation assessment value, and the loss coefficient is obtained according to the final fit to adjust the over-limit threshold and the far-exceeding upper limit threshold to obtain the regulated over-limit threshold and the far-exceeding upper limit threshold; The evaluation score of each parameter is obtained according to the over-limit threshold and far-exceeding upper limit threshold after regulation, and the evaluation score of each parameter is accumulated to obtain the evaluation score of the process technology parameter control risk. The evaluation score belonging to the process technology parameter control risk in the safety risk assessment value before regulation is replaced with the evaluation score after regulation to obtain the safety risk value after regulation, and an early warning is issued according to the safety risk value; The method for obtaining the safety risk assessment value before regulation based on the over-limit threshold and far-exceeding upper limit threshold of different parameters before regulation is: There are eight types of risks in the production workshop, namely inherent risks, hazard source risks, process technology parameter control risks, on-site operation risks, safety risk inspections, hidden danger investigation quality, LDAR risks and other risks; The method for obtaining the preliminary fluctuation assessment value at each time point according to the loss degree at different time points is: The loss degree of all time points constitutes a loss curve, the horizontal axis of the loss curve is the time point, the vertical axis is the loss degree, and the points on the loss curve are data points. The loss degree corresponding to each time point is subtracted from the loss degree corresponding to the previous time point, and the difference is used as the preliminary fluctuation assessment value of the time point; The method for obtaining the discrete deviation value of each time point according to the loss degree of the time point as the center point and the loss degree of the time point within the time window of the time point as the center point is: Obtain the median of the loss degree of all time points in the time window of each time point, find the maximum and minimum values of the loss degree in the time window, make the difference between the maximum and minimum values of the loss degree to obtain the limit range of each time window, and obtain the discrete deviation value of each time point according to the difference between the loss degree of the time point and the median of the loss degree of the corresponding time window and the limit range in the time window; The method for obtaining the deviation degree of each time point in the time window according to the loss degree of the time point and the difference in loss degrees of adjacent time points in the time window is: The time point in the time window corresponding to each time point is recorded as the window time point, and the data points corresponding to the two window time points at both ends of the time window are connected, and the connection line is recorded as the distribution direction function. A loss pseudo value is obtained for each window time point on the distribution direction function. The deviation of the window time point is obtained according to the difference between the loss degree of the window time point and the loss pseudo value and the difference in the loss degree of adjacent window time points; The method for obtaining the discrete distribution characteristics of the time window corresponding to each time point according to the deviation and loss degree of all time points in the time window corresponding to each time point is: In the formula, a t,n Indicates the degree of loss at the nth time point in the time window corresponding to the tth time point, represents the mean of the loss degree of all time points in the time window corresponding to the tth time point, q t,n Indicates the deviation of the nth time point in the time window corresponding to the tth time point, N t represents the length of the time window at the tth time point, δ t Represents the discrete distribution characteristics of the time window corresponding to the t-th time point; The method for obtaining the actual single dispatch of the data point corresponding to each time point according to the discrete deviation value of each time point and the discrete distribution characteristics corresponding to the time point is: The product of the discrete deviation value corresponding to each time point and the discrete distribution feature is used as the actual single dispatch of the data point corresponding to the time point; The method of obtaining the loss coefficient according to the final fitting degree and adjusting the over-limit threshold and the far-exceeding upper limit threshold to obtain the regulated over-limit threshold and the far-exceeding upper limit threshold is: In the formula, Indicates the exceeding threshold value, Indicates that it is far beyond the upper threshold. Indicates the over-limit threshold after regulation, It means that the value after regulation far exceeds the upper limit threshold, and K represents the loss coefficient.
2. The production safety early warning method based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining the discrete deviation value of each time point according to the difference between the loss degree of the time point and the median value of the loss degree of the corresponding time window and the restricted range within the time window is: In the formula, a t Indicates the degree of loss corresponding to the t-th time point, a mid(t) represents the median loss of the time window at the tth time point, Δa (t) represents the limit range of the time window at the tth time point, norm() represents the linear normalization function, and p t Represents the discrete deviation value at the tth time point.
3. The production safety early warning method based on artificial intelligence according to claim 1 is characterized in that: The method for obtaining the deviation of the window time point according to the difference between the loss degree of the window time point and the loss hypothetical value and the difference between the loss degrees of adjacent window time points is: For each window time point in the time window corresponding to the time point, the difference in loss degree between the window time point and the previous window time point is recorded as the first difference, the difference in loss degree between the window time point and the next window time point is recorded as the second difference, and the difference in loss degree between the window time point before the window time point and the window time point after the window time point is recorded as the third difference. The sum of the first difference and the second difference is divided by the third difference to obtain the first ratio. The difference between 1 and the first ratio is normalized and recorded as linearity. The normalized value of the absolute value of the difference between the loss degree of each window time point and the loss hypothetical value is recorded as the difference value. The difference value of each window time point is multiplied by the linearity to obtain the deviation of the window time point.
