Refining device safety production state influence factor mining method and system

Abstract

The embodiment of the invention provides a refining device safety production state influence factor mining method and system, and belongs to the technical field of refining devices. The method comprises the steps that power consumption parameter information of a target enterprise in a preset period is collected, and the comprehensive load degree of the target enterprise is calculated based on the power consumption parameter information; collecting candidate index data influencing the safety production state of the refining device, and preprocessing the candidate index data; performing feature extraction on the preprocessed candidate index data to obtain a plurality of candidate feature vectors; and based on an improved entropy grey correlation algorithm, carrying out correlation analysis on the comprehensive load degree and each candidate feature vector, sorting each candidate feature vector based on the correlation degree, and outputting a sorted candidate feature vector set as the safety production state influence factor of the refining device under the target enterprise. According to the scheme, the problems of low data monitoring efficiency and poor objectivity in the existing scheme are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of refinery equipment, and in particular to a method for mining influencing factors of safe production status of refinery equipment and a system for mining influencing factors of safe production status of refinery equipment. Background Art

[0002] Refining and chemical units are the core of the petrochemical industry, and their operating stability is directly related to the overall stable operation of the entire petrochemical enterprise. Based on this, ensuring the safe and stable operation of refining and chemical units has always been an important goal of petrochemical enterprises' production and operation. The production status of refining and chemical units is jointly affected by factors such as the operating years of the unit, the number of risk hazards, process fluctuations, and operation inspection and maintenance. In order to support the decision-making in the safe production process of refining and chemical enterprises, a relatively accurate theoretical data basis is required, and reasoning is required for the production status of refining and chemical units, so that during the operation of refining and chemical units, the operating status of refining and chemical units can be judged and predicted in real time based on the collected operating parameters. On the other hand, it is also possible to judge whether the refining and chemical units are in the best operating state based on the real-time operating parameters of the refining and chemical units, so as to optimize the control of the refining and chemical units. These requirements all require accurate identification of the factors affecting the safe production status of refining and chemical units, so as to carry out targeted data monitoring and predict the operating status with the monitoring data. However, when the existing schemes monitor the operating status of refining and chemical units, a large amount of operating information is often required to be coupled, and there is no corresponding objective identification standard, which cannot realize the identification of key influencing factors, and there are problems of low data monitoring efficiency and poor objectivity. Summary of the invention

[0003] The purpose of the embodiments of the present invention is to provide a method and system for mining factors affecting the safe production status of a refinery, so as to at least solve the problems of low data monitoring efficiency and poor objectivity in existing solutions.

[0004] In order to achieve the above-mentioned objectives, the first aspect of the present invention provides a method for mining factors affecting the safe production status of a refining and chemical plant, which collects power consumption parameter information of a target enterprise within a preset period, and calculates the comprehensive load degree of the target enterprise based on the power consumption parameter information; collects candidate indicator data affecting the safe production status of the refining and chemical plant, and preprocesses the candidate indicator data; performs feature extraction on the preprocessed candidate indicator data to obtain multiple candidate feature vectors; based on an improved entropy grey correlation algorithm, performs a correlation analysis between the comprehensive load degree and each candidate feature vector, sorts each candidate feature vector based on the correlation degree, and outputs a sorted candidate feature vector set as a factor affecting the safe production status of the refining and chemical plant under the target enterprise.

[0005] Optionally, the power consumption parameter information includes: power consumption information of the target enterprise within a preset period, production volume of the target enterprise within a preset period, maximum inherent load of the industry, minimum inherent load of the industry, load of the target enterprise, and service life of the refining equipment of the target enterprise.

[0006] Optionally, the calculation of the comprehensive load degree of the target enterprise based on the power consumption parameter information includes: calculating a first load index value based on the power consumption information of the target enterprise within a preset period and a power load index calculation rule; calculating a second load index value based on the maximum production volume of the target enterprise and the productivity load index calculation rule; calculating a third load index value based on the maximum inherent load of the industry, the minimum inherent load of the industry, the load of the target enterprise, the service life of the target enterprise's refining equipment and the inherent load calculation rule; performing a weighted operation based on the first load index value, the second load index value and the third load index value to obtain the comprehensive load degree of the target enterprise.

[0007] Optionally, the electric load index calculation rule is:

[0008]

[0009] Among them, y 1 is the first load index value; x 1 is the power consumption information of the target enterprise within the preset period; P is the average power consumption information of the target enterprise within each historical preset period; the productivity load index calculation rule is:

[0010]

[0011] Among them, y 2 is the second load index value; x 2 is the production volume of the target enterprise within a preset period; Q is the maximum production volume of the target enterprise within a preset period; the inherent load calculation rule is:

[0012]

[0013] Among them, y 3 is the third load index value; r max The maximum inherent load in the industry; min The minimum inherent load in the industry; i is the load of the target enterprise; α is the correction parameter set based on the service life of the target enterprise's refining and chemical equipment.

[0014] Optionally, the calculation rule for obtaining the comprehensive load degree of the target enterprise by performing weighted calculation based on the first load index value, the second load index value and the third load index value is:

[0015] Y=w 1 y 1 +w 2 y 2 +w 3 y 3

[0016] Among them, Y is the comprehensive load degree of the target enterprise; w 1 is the preset weight of the first load index value; 2 is the preset weight of the second load index value; 3 is the preset weight of the third load index value.

[0017] Optionally, the preprocessing of the candidate indicator data includes: null value processing, duplicate value processing and data skewness correction.

[0018] Optionally, the null value processing includes: one or more processing rules of fixed value filling, mean filling, median filling, mode filling, random interpolation and adjacent interpolation.

[0019] Optionally, the correction rule for data skewness correction includes: based on a preset skewness calculation rule, calculating the skewness coefficient of the data set composed of the candidate indicator data group in a preset designated data column; comparing the skewness coefficient with a preset skewness coefficient threshold, and when the skewness coefficient is greater than the preset skewness coefficient threshold, performing data balancing based on undersampling and / or oversampling; wherein, the preset skewness calculation rule is:

[0020]

[0021] Among them, n is the number of candidate indicator data; x i is the i-th candidate indicator data; μ is the mean of the preset specified data column; σ is the standard deviation of the preset specified data column.

[0022] Optionally, the feature extraction of the candidate indicator data after preprocessing to obtain multiple candidate feature vectors includes: classifying the candidate indicator data after preprocessing based on multiple preset categories to obtain multiple category data sets; wherein the multiple preset categories include: unordered data, character string data, high cardinality data and continuous numerical data; for each category data set, using a corresponding encoding method to encode the data to obtain multiple initial feature vectors; wherein the encoding method corresponding to the unordered data is a one-hot encoding algorithm; the encoding method for the character string data is a Target Encoding algorithm; the encoding method for the high cardinality data is a K-Flod Target Encoding algorithm; the encoding method for the continuous numerical data is an empirical algorithm; and feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fused features as candidate feature vectors.

[0023] Optionally, the feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fused features, including: identifying the feature meaning of each initial feature vector; fusing the initial feature vectors with the same feature meaning to obtain a fused feature; traversing all initial feature vectors to obtain multiple fused features.

[0024] Optionally, the feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fused features, including: classifying the initial feature vectors based on similarity, and initializing the population under the GA-FCM genetic algorithm based on the classified initial feature vectors; performing multi-generation inheritance based on the initialized population, terminating the inheritance when a preset termination condition is triggered, and outputting the individual with maximum fitness; performing feature normalization processing based on the individual with maximum fitness to obtain the normalized feature vectors after each classification as the fused feature.

[0025] Optionally, the improved entropy grey correlation algorithm is used to perform correlation analysis between the comprehensive load degree and each candidate feature vector, including: taking the comprehensive load degree as a reference sequence and each candidate feature vector as a comparison sequence; standardizing the reference sequence and calculating the absolute difference between the comparison sequence and the standardized reference sequence; calculating the grayscale relationship coefficient between each candidate feature vector and the standardized reference sequence based on the absolute difference; and determining the grayscale relationship level of each candidate feature vector based on the grayscale relationship coefficient.

[0026] The second aspect of the present invention provides a system for mining factors affecting the safe production status of a refinery, the system comprising: a collection unit, for collecting power consumption parameter information of a target enterprise within a preset period, and calculating the comprehensive load level of the target enterprise based on the power consumption parameter information; a processing unit, for collecting candidate indicator data affecting the safe production status of the refinery, and preprocessing the candidate indicator data; a feature extraction unit, for extracting features from the preprocessed candidate indicator data to obtain a plurality of candidate feature vectors; an analysis unit, for performing a correlation analysis between the comprehensive load level and each candidate feature vector based on an improved entropy grey correlation algorithm, sorting each candidate feature vector based on the correlation level, and outputting a sorted candidate feature vector set as a factor affecting the safe production status of a refinery under the target enterprise.

[0027] A third aspect of the present invention provides a computer-readable storage medium having instructions stored thereon, which, when executed on a computer, enables the computer to execute the above-mentioned method for mining factors affecting the safe production status of a refinery.

[0028] A fourth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for mining factors affecting the safe production status of a refining device when executing the computer program.

[0029] Through the above technical solution, the solution of the present invention uses the power consumption parameter information that the enterprise is concerned about as a monitoring indicator for measuring the operating status of the enterprise, so as to facilitate the synchronous monitoring of the parameters that affect the safe operation and optimization of the refining and chemical equipment. And based on the improved entropy gray correlation algorithm, the candidate influencing indicator data is analyzed for correlation to identify the influencing factors with a higher degree of influence, so as to facilitate the subsequent key monitoring based on these key influencing factors, reduce the number of indicator monitoring, and improve the efficiency of status monitoring.

[0030] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0032] Figure 1 It is a flowchart of the steps of a method for mining factors affecting the safe production status of a refinery provided by an embodiment of the present invention;

[0033] Figure 2 It is a system structure diagram of a system for mining factors affecting safe production status of a refinery provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the present invention, and is not used to limit the present invention.

[0035] Refining and chemical units are the core of the petrochemical industry, and their operating stability is directly related to the overall stable operation of the entire petrochemical enterprise. Based on this, ensuring the safe and stable operation of refining and chemical units has always been an important goal of petrochemical enterprises' production and operation. The production status of refining and chemical units is jointly affected by factors such as the operating years of the unit, the number of risk hazards, process fluctuations, and operation inspection and maintenance. In order to support the decision-making in the safe production process of refining and chemical enterprises, a relatively accurate theoretical data basis is required, and reasoning is required for the production status of refining and chemical units, so that during the operation of refining and chemical units, the operating status of refining and chemical units can be judged and predicted in real time based on the collected operating parameters. On the other hand, it is also possible to judge whether the refining and chemical units are in the best operating state based on the real-time operating parameters of the refining and chemical units, so as to optimize the control of the refining and chemical units. These requirements all require accurate identification of the factors affecting the safe production status of refining and chemical units, so as to carry out targeted data monitoring and predict the operating status with the monitoring data. However, when the existing schemes monitor the operating status of refining and chemical units, a large amount of operating information is often required to be coupled, and there is no corresponding objective identification standard, which cannot realize the identification of key influencing factors, and there are problems of low data monitoring efficiency and poor objectivity.

[0036] In view of the problems of low data monitoring efficiency and poor objectivity in existing solutions, the present invention proposes a method for mining factors affecting the safe production status of refining and chemical equipment. The present invention uses the power consumption parameter information that the enterprise is concerned about as a monitoring indicator for measuring the operating status of the enterprise, so as to facilitate the simultaneous monitoring of parameters affecting the safe operation and optimization of the refining and chemical equipment. Based on the improved entropy grey correlation algorithm, the candidate influencing indicator data is analyzed for correlation to identify factors with a higher degree of influence, so as to facilitate subsequent key monitoring based on these key influencing factors, reduce the number of indicator monitoring, and improve the efficiency of status monitoring.

[0037] Figure 1 1 is a flow chart of a method for mining factors affecting the safe production status of a refinery provided by an embodiment of the present invention. Figure 1 As shown, an embodiment of the present invention provides a method for mining factors affecting the safe production status of a refinery, the method comprising:

[0038] Step S10: Collect power consumption parameter information of the target enterprise within a preset period, and calculate the comprehensive load level of the target enterprise based on the power consumption parameter information.

[0039] Specifically, the scheme of the present invention not only aims to identify the key factors affecting the safe and stable operation of the refining and chemical equipment, but also needs to identify the key factors affecting the sensitive indicators of the refining and chemical enterprise experience, and needs to ensure that these two indicators are met at the same time. Based on this, the scheme of the present invention uses the peripheral indicators that the enterprise cares about as related reference indicators. These peripheral indicators include power consumption information, output information, etc. By collecting these business-related information, it can be ensured that the key parameters identified subsequently meet both the safe operation and optimized operation indicators.

[0040] Preferably, the power consumption parameter information includes: power consumption information of the target enterprise within a preset period, production volume of the target enterprise within a preset period, maximum inherent load of the industry, minimum inherent load of the industry, load of the target enterprise, and service life of the refining equipment of the target enterprise.

[0041] Preferably, the calculation of the comprehensive load degree of the target enterprise based on the power consumption parameter information includes: calculating a first load index value based on the power consumption information of the target enterprise within a preset period and a power load index calculation rule; calculating a second load index value based on the maximum production volume of the target enterprise and a productivity load index calculation rule; calculating a third load index value based on the maximum inherent load of the industry, the minimum inherent load of the industry, the load of the target enterprise, the service life of the target enterprise's refining equipment and the inherent load calculation rule; performing a weighted operation based on the first load index value, the second load index value and the third load index value to obtain the comprehensive load degree of the target enterprise.

[0042] In the embodiment of the present invention, if a subsequent correlation analysis is performed for each peripheral indicator, the criticality of some factors is easily magnified, and the indicator identification process needs to be repeated, resulting in increased delay. The solution of the present invention integrates multiple peripheral indicators into an overall indicator, that is, to obtain a comprehensive load degree, so as to facilitate subsequent identification of key influencing factors based on the comprehensive load degree.

[0043] Preferably, the electric load index calculation rule is:

[0044]

[0045] Among them, y 1 is the first load index value; x 1 is the power consumption information of the target enterprise within the preset period; P is the average power consumption information of the target enterprise within each historical preset period; the productivity load index calculation rule is:

[0046]

[0047] Among them, y 2 is the second load index value; x 2is the production volume of the target enterprise within a preset period; Q is the maximum production volume of the target enterprise within a preset period;

[0048] The inherent load calculation rule is:

[0049]

[0050] Among them, y 3 is the third load index value; r max The maximum inherent load in the industry; min The minimum inherent load in the industry; i is the load of the target enterprise; α is the correction parameter set based on the service life of the target enterprise's refining and chemical equipment.

[0051] Embodiment 1:

[0052] The comprehensive load level of the enterprise can be obtained by weighting the average monthly electricity consumption P of the enterprise's equipment during the enterprise cycle (generally stipulated as one quarter), the maximum production capacity Q of the enterprise's equipment, the maximum and minimum inherent load capacities of all enterprises, and the aging correction coefficient of the enterprise's equipment (when the enterprise's equipment is ≥20 years old).

[0053] Preferably, the calculation rule for obtaining the comprehensive load degree of the target enterprise by performing weighted calculation based on the first load index value, the second load index value and the third load index value is:

[0054] Y=w 1 y 1 +w 2 y 2 +w 3 y 3

[0055] Among them, Y is the comprehensive load degree of the target enterprise; w 1 is the preset weight of the first load index value; 2 is the preset weight of the second load index value; 3 is the preset weight of the third load index value. Here we select the mean weight It can also be set manually based on expert experience, or based on algorithms such as AHP and entropy weight method.

[0056] Step S20: collecting candidate indicator data that affect the safe production status of the refinery, and preprocessing the candidate indicator data.

[0057] Specifically, after obtaining the evaluation index for subsequent key factor determination, the corresponding key factor identification is required. These key factors are hidden in all operating parameters, and the scheme of the present invention needs to identify these key factors from all operating parameters. Based on this, the scheme of the present invention needs to classify all parameters for the first time, subdivide all parameters into multiple data items, and use these data items as candidate indicator data. Subsequently, the correlation degree between these candidate indicator data and the comprehensive load degree is analyzed to determine the influence of each candidate indicator on the comprehensive load degree, thereby identifying the criticality of each candidate indicator data.

[0058] In one possible implementation, the candidate indicator data include device attributes (device ID, device type, hazardous process classification, hazardous source code), device failure records, device self-control records), process parameter change trends (process risk level, dual protection risk level, number of device failures, unplanned shutdown indicator downtime, failure type, failure duration), real-time alarms (alarm type, alarm timestamp, total number of alarms), on-site operation conditions (investigation tasks, hidden danger reminder records, work tickets, special work tickets), risk events (real-time risk value, initial risk value, device risk level, timestamp, risk event analysis records, risk event analysis completion time), hidden danger investigation conditions (device inspection points, task investigation status).

[0059] Specifically, before obtaining the candidate indicator data, because all operating parameters are collected, these operating parameters contain a large amount of redundant data or missing data. If such data is directly used for subsequent correlation analysis, there will definitely be a situation where the recognition result does not conform to the actual situation. Based on this, the scheme of the present invention needs to pre-process these data. Preferably, the pre-processing of the candidate indicator data includes: null value processing, duplicate value processing and data skewness correction.

[0060] Embodiment 2:

[0061] The empty value processing includes one or more processing rules of fixed value filling, mean filling, median filling, mode filling, random interpolation and adjacent interpolation. According to the data content in the data table, the scheme of the present invention deletes the devices with empty enterprise load index, deletes the characteristic factors with missing rate of other characteristic data exceeding 50%, and replaces the remaining numerical vacancy values ​​with the mode filling method with the mode of all devices of this characteristic for the remaining vacancy rate less than 10%. The adjacent interpolation and random interpolation methods are used to fill the remaining data with the vacancy rate of 50%-10%.

[0062] Preferably, duplicate value processing is performed, and duplicate data is searched and deleted according to an SQL statement.

[0063] Specifically, the correction rules for data skewness correction include:

[0064] Based on the preset skewness calculation rules, the skewness coefficient of the data set composed of the candidate indicator data group in the preset specified data column is calculated;

[0065] The skewness coefficient is compared with a preset skewness coefficient threshold. When the skewness coefficient is greater than the preset skewness coefficient threshold, data balancing is performed based on undersampling and / or oversampling. The preset skewness calculation rule is:

[0066]

[0067] Among them, n is the number of candidate indicator data; x i is the i-th candidate indicator data; μ is the mean of the preset specified data column; σ is the standard deviation of the preset specified data column.

[0068] Embodiment three:

[0069] Calculate the data skewness coefficient and correct it. Since the data in the data set is biased, such as the low-level data in the warning level data is significantly more than the high-level data. The training effect of the subsequent model will be affected because such high-level data will be ignored. Therefore, it is necessary to detect the characteristic distribution of the data set. Calculate the absolute value of the skewness coefficient of the data set in the specified data column to feedback the degree of data set deviation. The formula for calculating data deviation is as follows:

[0070]

[0071] Among them, n is the number of candidate indicator data; x i is the i-th candidate indicator data; μ is the mean of the preset specified data column; σ is the standard deviation of the preset specified data column. If the full data set is specified, the default value is 0. If the skewness coefficient is greater than 0.2, the specified feature data is randomly deleted (undersampling) and more samples are randomly added from the minority class (oversampling) according to the data bias to balance the data.

[0072] Step S30: extracting features from the candidate indicator data after preprocessing to obtain a plurality of candidate feature vectors.

[0073] Specifically, the candidate indicator data after preprocessing is classified based on multiple preset categories to obtain multiple category data sets; wherein the preset multiple categories include: unordered data, character string data, high cardinality data and continuous numerical data; for each category data set, a corresponding encoding method is used to encode the data to obtain multiple initial feature vectors; wherein the encoding method corresponding to the unordered data is a one-hot encoding algorithm; the encoding method for the character string data is a Target Encoding algorithm; the encoding method for the high cardinality data is a K-Flod TargetEncoding algorithm; the encoding method for the continuous numerical data is an empirical algorithm; feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fused features as candidate feature vectors.

[0074] Embodiment 4:

[0075] For data with no intrinsic order and less than 4 categories, such as the current warning status data (red, orange, yellow, and blue), the OneHotEncoding method is used for encoding and assignment. The Target Encoding method is used to encode string category data that contains a large proportion, such as hazard source codes, emergency event types, and device type names, and smoothing is added to improve the appearance of unknown categories and long-tail phenomena (the phenomenon that most categories have only a small amount of data), and finally standardized. The specific principle is to replace the classification features with the posterior probability of the corresponding target value. For high-cardinality category feature data such as device id, hazard source code, etc., the K-Flod Target Encoding method is used for encoding and assignment. Continuous numerical data such as the total number of alarms (total_alarm_num) are segmented and defined according to expert experience to enhance the model's learning of the number of device alarms.

[0076] Specifically, the feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fused features, including: identifying the feature meaning of each initial feature vector; fusing the initial feature vectors with the same feature meaning to obtain fused features; traversing all initial feature vectors to obtain multiple fused features.

[0077] In an embodiment of the present invention, the most common alarm type of the device is calculated based on the number of device alarms, alarm type and alarm duration, the emergency event type with a larger proportion is calculated based on the emergency event records and timestamps, and the average monthly number of operations of the device is calculated based on the job ticket records and timestamps.

[0078] In another possible implementation, the feature fusion based on the correlation characteristics of each initial feature vector to obtain multiple fusion features includes: classifying each initial feature vector based on similarity, and initializing the population under the GA-FCM genetic algorithm based on each classified initial feature vector; performing multi-generation inheritance based on the initialized population, ending the inheritance when a preset termination condition is triggered, and outputting the individual with the maximum fitness; performing feature normalization processing based on the individual with the maximum fitness, and obtaining the normalized feature vectors after each classification as the fusion feature. Specifically, the following steps are included:

[0079] 1) Input cluster sample X, cluster number c and other related parameters.

[0080] 2) Initialize the normal population and related parameters.

[0081] 3) Initialize high-quality population, low-quality population and related parameters.

[0082] 4) In step 1) and step 2), the three types of populations evolve according to their respective genetic strategies: the normal population performs general genetic operations; the high-quality population is selected according to the fitness ratio, and performs consistent crossover and induced mutation; the low-quality population is selected according to the fitness ratio, and performs consistent crossover and large mutation operations.

[0083] 5) After every K generations, the three populations are merged, and the last μ+λ individuals in descending order of fitness are eliminated. The remaining n individuals are used as the normal population for the next evolution, and individuals from the high-quality and low-quality populations are selected from them;

[0084] 6) If the termination condition is met, the operation ends and the individual with the maximum fitness is output as the optimal solution for this calculation, otherwise go to step 2).

[0085] 7) The cluster center is refined and feature normalization is performed to compress the feature data to between 0 and 1.

[0086] Step S40: Based on the improved entropy grey correlation algorithm, the correlation analysis between the comprehensive load degree and each candidate feature vector is performed, each candidate feature vector is sorted based on the correlation degree, and the sorted candidate feature vector set is output as the influencing factor of the safe production status of the refining and chemical unit of the target enterprise.

[0087] Specifically, the comprehensive load degree is used as a reference sequence, and each candidate feature vector is used as a comparison sequence; the reference sequence is standardized, and the absolute difference between the comparison sequence and the standardized reference sequence is calculated; the grayscale relationship coefficient between each candidate feature vector and the standardized reference sequence is calculated based on the absolute difference; and the grayscale relationship level of each candidate feature vector is determined based on the grayscale relationship coefficient.

[0088] Embodiment five:

[0089] First, determine the reference sequence and comparison sequence: This step includes collecting data and determining the indicators to be used for analysis. First, collect data, determine the indicators, determine the original evaluation matrix, and determine the reference sequence as X′ 0 =(x′ 01 ,x′ 02 ,…,x′ 0n ) to get the formula:

[0090]

[0091] Further, the reference sequence is standardized, and the absolute difference between the comparison sequence and the standardized reference sequence is calculated. This step includes calculating the absolute difference between the reference sequence and the comparison sequence, and the calculation rule is:

[0092] ∣X i -X 0 ∣=∣X ik -X 0k ∣.

[0093] Furthermore, the grayscale relationship coefficient is calculated. This step includes calculating the grayscale relationship coefficient between each indicator and the reference sequence:

[0094]

[0095] Wherein, η is the resolution factor, and its value is between (0, 1), and η is usually set to 0.5.

[0096] Furthermore, the gray entropy correlation is calculated, and the evaluation index to be analyzed is compared with the evaluation object to obtain the original correlation. The gray relationship level of each comparison sequence is calculated. The calculation rule is:

[0097] p=(P 01 , P 02 , …P 0i , ...P 0m ) T

[0098]

[0099]

[0100] Through the grey entropy correlation p 0i Sort the influencing factors from large to small. Among them, the factors with high correlation have a greater impact on the reliability index.

[0101] In an embodiment of the present invention, the scheme of the present invention sorts the candidate indicator data from large to small based on the degree of correlation, indicating that the greater the impact on the comprehensive load index, the higher the corresponding ranking. By presetting the number, the N most influential influencing indicators can be determined, and these indicators can be monitored subsequently.

[0102] Figure 2 1 is a system structure diagram of a system for mining factors affecting the safe production status of a refinery provided by an embodiment of the present invention. Figure 2 As shown, an embodiment of the present invention provides a system for mining factors affecting the safe production status of a refinery, and the system includes:

[0103] The collecting unit is used to collect the power consumption parameter information of the target enterprise within a preset period, and calculate the comprehensive load degree of the target enterprise based on the power consumption parameter information.

[0104] Specifically, the scheme of the present invention not only aims to identify the key factors affecting the safe and stable operation of the refining and chemical equipment, but also needs to identify the key factors affecting the sensitive indicators of the refining and chemical enterprise experience, and needs to ensure that these two indicators are met at the same time. Based on this, the scheme of the present invention uses the peripheral indicators that the enterprise cares about as related reference indicators. These peripheral indicators include power consumption information, output information, etc. By collecting these business-related information, it can be ensured that the key parameters identified subsequently meet both the safe operation and optimized operation indicators.

[0105] Preferably, the power consumption parameter information includes: power consumption information of the target enterprise within a preset period, production volume of the target enterprise within a preset period, maximum inherent load of the industry, minimum inherent load of the industry, load of the target enterprise, and service life of the refining equipment of the target enterprise.

[0106] Preferably, the calculation of the comprehensive load degree of the target enterprise based on the power consumption parameter information includes: calculating a first load index value based on the power consumption information of the target enterprise within a preset period and a power load index calculation rule; calculating a second load index value based on the maximum production volume of the target enterprise and a productivity load index calculation rule; calculating a third load index value based on the maximum inherent load of the industry, the minimum inherent load of the industry, the load of the target enterprise, the service life of the target enterprise's refining equipment and the inherent load calculation rule; performing a weighted operation based on the first load index value, the second load index value and the third load index value to obtain the comprehensive load degree of the target enterprise.

[0107] In the embodiment of the present invention, if a subsequent correlation analysis is performed for each peripheral indicator, the criticality of some factors is easily magnified, and the indicator identification process needs to be repeated, resulting in increased delay. The solution of the present invention integrates multiple peripheral indicators into an overall indicator, that is, to obtain a comprehensive load degree, so as to facilitate subsequent identification of key influencing factors based on the comprehensive load degree.

[0108] Preferably, the electric load index calculation rule is:

[0109]

[0110] Among them, y 1 is the first load index value; x 1 is the power consumption information of the target enterprise within the preset period; P is the average power consumption information of the target enterprise within each historical preset period; the productivity load index calculation rule is:

[0111]

[0112] Among them, y 2 is the second load index value; x 2 is the production volume of the target enterprise within a preset period; Q is the maximum production volume of the target enterprise within a preset period;

[0113] The inherent load calculation rule is:

[0114]

[0115] Among them, y 3 is the third load index value; r max The maximum inherent load in the industry; min The minimum inherent load in the industry; i is the load of the target enterprise; α is the correction parameter set based on the service life of the target enterprise's refining and chemical equipment.

[0116] The processing unit is used to collect candidate indicator data that affect the safe production status of the refining device and pre-process the candidate indicator data.

[0117] Specifically, after obtaining the evaluation indicators for subsequent key factor determination, it is necessary to identify the corresponding key factors. These key factors are hidden in all operating parameters, and the scheme of the present invention needs to identify these key factors from all operating parameters. Based on this, the scheme of the present invention needs to classify all parameters for the first time, subdivide all parameters into multiple data items, and use these data items as candidate indicator data. Subsequently, the correlation degree between these candidate indicator data and the comprehensive load degree is analyzed to determine the influence of each candidate indicator on the comprehensive load degree, thereby identifying the criticality of each candidate indicator data.

[0118] In one possible implementation, the candidate indicator data include device attributes (device ID, device type, hazardous process classification, hazardous source code), device failure records, device self-control records), process parameter change trends (process risk level, dual protection risk level, number of device failures, unplanned shutdown indicator downtime, failure type, failure duration), real-time alarms (alarm type, alarm timestamp, total number of alarms), on-site operation conditions (investigation tasks, hidden danger reminder records, work tickets, special work tickets), risk events (real-time risk value, initial risk value, device risk level, timestamp, risk event analysis records, risk event analysis completion time), hidden danger investigation conditions (device inspection points, task investigation status).

[0119] Specifically, before obtaining the candidate indicator data, because all operating parameters are collected, these operating parameters contain a large amount of redundant data or missing data. If such data is directly used for subsequent correlation analysis, there will definitely be a situation where the recognition result does not conform to the actual situation. Based on this, the scheme of the present invention needs to pre-process these data. Preferably, the pre-processing of the candidate indicator data includes: null value processing, duplicate value processing and data skewness correction.

[0120] Specifically, the correction rules for data skewness correction include:

[0121] Based on the preset skewness calculation rules, the skewness coefficient of the data set composed of the candidate indicator data group in the preset specified data column is calculated;

[0122] The skewness coefficient is compared with a preset skewness coefficient threshold. When the skewness coefficient is greater than the preset skewness coefficient threshold, data balancing is performed based on undersampling and / or oversampling. The preset skewness calculation rule is:

[0123]

[0124] Among them, n is the number of candidate indicator data; x i is the i-th candidate indicator data; μ is the mean of the preset specified data column; σ is the standard deviation of the preset specified data column.

[0125] The feature extraction unit is used to extract features from the candidate indicator data after preprocessing to obtain multiple candidate feature vectors.

[0126] Specifically, the candidate indicator data after preprocessing is classified based on multiple preset categories to obtain multiple category data sets; wherein the preset multiple categories include: unordered data, character string data, high cardinality data and continuous numerical data; for each category data set, a corresponding encoding method is used to encode the data to obtain multiple initial feature vectors; wherein the encoding method corresponding to the unordered data is a one-hot encoding algorithm; the encoding method for the character string data is a Target Encoding algorithm; the encoding method for the high cardinality data is a K-Flod TargetEncoding algorithm; the encoding method for the continuous numerical data is an empirical algorithm; feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fused features as candidate feature vectors.

[0127] The analysis unit is used to perform correlation analysis between the comprehensive load degree and each candidate feature vector based on the improved entropy grey correlation algorithm, sort each candidate feature vector based on the correlation degree, and output the sorted candidate feature vector set as an influencing factor of the safe production status of the refining and chemical unit of the target enterprise.

[0128] Specifically, the feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fused features, including: identifying the feature meaning of each initial feature vector; fusing the initial feature vectors with the same feature meaning to obtain fused features; traversing all initial feature vectors to obtain multiple fused features.

[0129] In an embodiment of the present invention, the most common alarm type of the device is calculated based on the number of device alarms, alarm type and alarm duration, the emergency event type with a larger proportion is calculated based on the emergency event records and timestamps, and the average monthly number of operations of the device is calculated based on the job ticket records and timestamps.

[0130] In another possible implementation, the feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fusion features, including: classifying the initial feature vectors based on similarity, and initializing the population under the GA-FCM genetic algorithm based on the classified initial feature vectors; performing multiple generations of inheritance based on the initialized population, terminating the inheritance when a preset termination condition is triggered, and outputting the individual with the maximum fitness; performing feature normalization processing based on the individual with the maximum fitness to obtain the normalized feature vectors after each classification as the fusion feature.

[0131] Specifically, the comprehensive load degree is used as a reference sequence, and each candidate feature vector is used as a comparison sequence; the reference sequence is standardized, and the absolute difference between the comparison sequence and the standardized reference sequence is calculated; the grayscale relationship coefficient between each candidate feature vector and the standardized reference sequence is calculated based on the absolute difference; and the grayscale relationship level of each candidate feature vector is determined based on the grayscale relationship coefficient.

[0132] The embodiment of the present invention further provides a computer-readable storage medium, on which instructions are stored, which, when executed on a computer, enable the computer to execute the above-mentioned method for mining factors affecting the safe production status of a refinery.

[0133] Those skilled in the art will understand that all or part of the steps in the method for implementing the above-mentioned embodiments can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions for making a single-chip microcomputer, a chip or a processor (processor) perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0134] The optional embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above embodiments. Within the technical concept of the embodiments of the present invention, the technical scheme of the embodiments of the present invention can be subjected to a variety of simple modifications, and these simple modifications all belong to the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present invention will not further describe various possible combinations.

[0135] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the embodiments of the present invention, they should also be regarded as the contents disclosed in the embodiments of the present invention.

Claims

1. A method for mining factors affecting the safe production status of a refinery, characterized in that: The method comprises: Collecting power consumption parameter information of the target enterprise within a preset period, and calculating the comprehensive load degree of the target enterprise based on the power consumption parameter information; Collecting candidate indicator data that affect the safe production status of the refining and chemical unit, and preprocessing the candidate indicator data; Perform feature extraction on the candidate indicator data after preprocessing to obtain multiple candidate feature vectors; Based on the improved entropy grey correlation algorithm, the correlation analysis between the comprehensive load degree and each candidate feature vector is performed, each candidate feature vector is sorted based on the correlation degree, and the sorted candidate feature vector set is output as the influencing factor of the safe production status of the refining and chemical unit of the target enterprise.

2. The method according to claim 1, characterized in that The power consumption parameter information includes: The target enterprise's electricity consumption information within the preset period, the target enterprise's production volume within the preset period, the industry's maximum inherent load, the industry's minimum inherent load, the target enterprise's load, and the service life of the target enterprise's refining and chemical equipment.

3. The method according to claim 2, characterized in that The calculating the comprehensive load level of the target enterprise based on the power consumption parameter information includes: Calculate and obtain a first load index value based on the power consumption information of the target enterprise within a preset period and a power load index calculation rule; Calculate and obtain a second load index value based on the maximum production volume and productivity load index calculation rule of the target enterprise; A third load index value is obtained by calculation based on the maximum inherent load of the industry, the minimum inherent load of the industry, the load of the target enterprise, the service life of the refining and chemical equipment of the target enterprise and the inherent load calculation rule; A weighted operation is performed based on the first load index value, the second load index value, and the third load index value to obtain a comprehensive load degree of the target enterprise.

4. The method according to claim 3, characterized in that The electric load index calculation rule is: Among them, y1 is the first load index value; x1 is the electricity consumption information of the target enterprise within the preset period; P The average electricity consumption information of the target enterprise in each historical preset period; The productivity load index calculation rule is: Wherein, y2 is the second load index value; x2 is the production volume of the target enterprise within the preset period; Q is the maximum production volume of the target enterprise in a preset period; The inherent load calculation rule is: Among them, y3 is the third load index value; r max It is the industry’s largest inherent load; r min It is the industry’s smallest inherent load; r i The load of the target enterprise; α is a correction parameter set based on the service life of the target enterprise's refining and chemical equipment.

5. The method according to claim 4, characterized in that The calculation rule for obtaining the comprehensive load degree of the target enterprise by performing weighted calculation based on the first load index value, the second load index value and the third load index value is: Y=w1y1+w2y2+w3y3 Among them, Y is the comprehensive load degree of the target enterprise; w1 is the preset weight of the first load index value; w2 is the preset weight of the second load index value; w3 is the preset weight of the third load index value.

6. The method according to claim 1, characterized in that The preprocessing of the candidate indicator data includes: Null value processing, duplicate value processing and data skewness correction.

7. The method according to claim 6, characterized in that The null value processing includes: One or more of the following processing rules: fixed value filling, mean filling, median filling, mode filling, random imputation, and neighbor imputation.

8. The method according to claim 6, characterized in that The correction rules for the data skewness correction include: Based on the preset skewness calculation rules, the skewness coefficient of the data set composed of the candidate indicator data group in the preset specified data column is calculated; Comparing the skewness coefficient with a preset skewness coefficient threshold, and when the skewness coefficient is greater than the preset skewness coefficient threshold, performing data balancing based on undersampling and / or oversampling; wherein, The preset skewness calculation rule is: Among them, n is the number of candidate indicator data; x i is the i-th candidate indicator data; μ is the mean of the preset data column; σ is the standard deviation of the preset data column.

9. The method according to claim 1, characterized in that: The feature extraction is performed on the candidate indicator data after preprocessing to obtain multiple candidate feature vectors, including: Based on multiple preset categories, the candidate indicator data after preprocessing is classified and processed to obtain multiple category data sets; wherein, The preset multiple categories include: Unordered data, string data, high cardinality data, and continuous numerical data; For each category data set, the corresponding encoding method is used to encode the data to obtain multiple initial feature vectors; among them, The encoding method corresponding to the disordered data is a one-hot encoding algorithm; The encoding method of the string data is Target Encoding algorithm; The encoding method of the high cardinality data is K-Flod Target Encoding algorithm; The encoding method of the continuous numerical data is an empirical algorithm; Based on the correlation characteristics of each initial feature vector, feature fusion is performed to obtain multiple fused features as candidate feature vectors.

10. The method according to claim 9, characterized in that The feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fusion features, including: Identify the characteristic meaning of each initial characteristic vector; The initial feature vectors with the same feature meaning are fused to obtain fused features; Traverse all initial feature vectors to obtain multiple fusion features.

11. The method according to claim 9, characterized in that The feature fusion is performed based on the correlation characteristics of each initial feature vector to obtain multiple fusion features, including: Classifying the initial feature vectors based on similarity, and initializing the population under the GA-FCM genetic algorithm based on the classified initial feature vectors; Execute multiple generations of inheritance based on the initialized population, end the inheritance when the preset termination condition is triggered, and output the individual with the maximum fitness; Based on the individual with the maximum fitness, feature normalization processing is performed to obtain normalized feature vectors after each classification as fusion features.

12. The method according to claim 1, characterized in that The improved entropy grey correlation algorithm is used to analyze the correlation between the comprehensive load level and each candidate feature vector, including: Taking the comprehensive load degree as a reference sequence and taking each candidate feature vector as a comparison sequence; Standardizing the reference sequence and calculating the absolute difference between the comparison sequence and the standardized reference sequence; Calculate the grayscale relationship coefficient between each candidate feature vector and the reference sequence after the standardization process based on the absolute difference; The grayscale relationship level of each candidate feature vector is determined based on the grayscale relationship coefficient.

13. A system for mining factors affecting the safe production status of a refinery, characterized in that: The system comprises: A collection unit, used to collect power consumption parameter information of a target enterprise within a preset period, and calculate a comprehensive load degree of the target enterprise based on the power consumption parameter information; A processing unit, used to collect candidate indicator data affecting the safe production status of the refining device, and pre-process the candidate indicator data; A feature extraction unit is used to extract features from the candidate indicator data after preprocessing to obtain multiple candidate feature vectors; The analysis unit is used to perform correlation analysis between the comprehensive load degree and each candidate feature vector based on the improved entropy grey correlation algorithm, sort each candidate feature vector based on the correlation degree, and output the sorted candidate feature vector set as an influencing factor of the safe production status of the refining and chemical unit of the target enterprise.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the method for analyzing the components of the collection and annotation stations based on reverse fitting as described in any one of claims 1 to 12.

15. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for mining factors affecting the safe production status of a refinery device as described in any one of claims 1 to 12 is implemented.

Images