An accident explosion traceability method based on a Bayesian classifier

Abstract

An accident explosion source tracing method based on a Bayesian classifier, which belongs to the field of accident explosion damage. The present invention solves the problems of poor real-time performance and low accuracy of existing accident explosion source tracing methods. The method of the present invention first analyzes and constructs the types of accident explosion sources, and then determines the explosion sources for historical accident explosion events, thereby constructing a labeled data set; for the accident explosion event to be traced, a set of similar events is selected from the data set, and explosion source tracing is carried out according to the set of similar events, so as to more accurately identify and track the source and damage impact of the accident explosion event. Compared with the traditional manual explosion source tracing method, the method of the present invention can greatly improve the efficiency of explosion source tracing, and does not require the establishment of a complex physical model, has lower requirements for data, simplifies the explosion source tracing process while improving the accuracy of source tracing, and ensures the feasibility and real-time performance of explosion source tracing. The present invention can be applied to tracing the source of accident explosions.

Description

Technical Field

[0001] The present invention belongs to the field of accidental explosion damage, and particularly relates to a method for tracing the source of accidental explosion. Background Art

[0002] Generally speaking, explosion source tracing refers to collecting, integrating and analyzing various accidental explosion investigation information, reasoning about the type of explosion source and the mechanism of explosion occurrence, so as to provide guidance for further disaster relief deployment.

[0003] Traditional explosion source tracing research methods include on-site investigation method, physical evidence identification method and simulation reconstruction method. First, use the on-site investigation method to visually observe, photograph, sample the site, etc., to obtain information such as explosive residues and explosion effects; then, use the physical evidence identification method to analyze and identify the collected physical evidence to determine the type and properties of explosive substances. Commonly used physical evidence identification methods include chemical composition analysis, structural analysis, etc.; finally, adopt the simulation reconstruction method to use computer simulation and other means based on on-site data and physical evidence analysis results to simulate and reconstruct the process of the explosion event and infer the actual explosion mechanism. The on-site investigation method and the physical evidence identification method have the problems of long time cycle for obtaining data and high labor cost. The simulation reconstruction method needs to establish a physical model with complex parameters to simulate various physical processes when an accidental explosion occurs. However, the environment at the explosion site is very complex when an accidental explosion actually occurs. It is unrealistic to directly apply the physical model to accurately predict the type of explosion source. Moreover, disaster relief after the explosion requires quickly determining the explosion source, which determines that the above methods are not practical for quickly tracing the explosion source.

[0004] In summary, due to the long time cycle required for the existing accidental explosion source tracing methods to obtain data, the real-time performance of accidental explosion source tracing is poor. Moreover, it is difficult to accurately simulate the accidental explosion process using a physical model, so the accuracy rate of accidental explosion source tracing is low. Therefore, it is very necessary to propose a new accidental explosion source tracing method. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of poor real-time performance and low accuracy rate of the existing accidental explosion source tracing methods. Using the existing accidental explosion damage events as prior knowledge and taking the phenomena observed at the explosion site and the situation of rapid investigation as conditions, a Bayesian classifier is used to predict the type of explosion source.

[0006] The technical solution adopted by the present invention to solve the above technical problems is:

[0007] An accidental explosion source tracing method based on a Bayesian classifier, the method specifically includes the following steps:

[0008] Step 1: Classify the explosion sources, and a total of N types of explosion sources are determined;

[0009] Step 2: According to the type of explosion source determined in Step 1, add the corresponding explosion source type label to the historical accidental explosion events with known information to obtain the labeled historical accidental explosion event dataset S1;

[0010] Step 3: The user inputs the on-site description information of the accidental explosion event to be traced.

[0011] Step 4: According to the on-site description information input by the user, calculate the correlation scores between the accidental explosion event to be traced and each accidental explosion event in the historical accidental explosion event dataset S1 respectively, and then sort the obtained correlation scores in descending order;

[0012] Select the accidental explosion events corresponding to the top m correlation scores in the historical accidental explosion event dataset S1, and use all the selected accidental explosion events to form the similar event set S2;

[0013] Step 5: Respectively divide the damage information of the similar event set S2 and the accidental explosion event to be traced into damage degree levels to realize the discretization of the damage information;

[0014] For any accidental explosion event in the similar event set S2, encode the discretized damage information of the accidental explosion event to obtain the feature vector of the accidental explosion event; encode the discretized damage information of the accidental explosion event to be traced to obtain the feature vector of the accidental explosion event to be traced;

[0015] Step 6: Use the accidental explosion event feature vectors obtained in Step 5 and the Bayesian algorithm to calculate the probabilities related to each explosion source category in the similar event set S2 for the accidental explosion event to be traced;

[0016] Then calculate the probabilities of the occurrence of various explosion sources under the on-site description information given by the user, and take the explosion source type with the highest probability as the final accidental explosion tracing result.

[0017] Furthermore, in Step 1, the value of N is 11.

[0018] Furthermore, the N types of explosion sources are respectively:

[0019] Type 1: High-pressure steam;

[0020] Type 2: Natural gas and coal gas;

[0021] Type 3: Chemical raw material gas;

[0022] Type 4: TNT;

[0023] Type 5: Ammonium nitrate;

[0024] Type 6: Black powder;

[0025] Class 7: Nitrocellulose;

[0026] Class 8: Nitroglycerin;

[0027] Class 9: Petroleum gasoline;

[0028] Class 10: Alcohol;

[0029] Class 11: Others.

[0030] Furthermore, the specific process of the second step is as follows:

[0031] Step 2-1: Investigate the information of historical accidental explosion events and the types of accidental explosion sources. The information of historical accidental explosion events includes time, location, explosion source information, detonation information, and damage information;

[0032] The explosion source information is the information of the object where the explosion occurs. The explosion source information includes the physical properties of the object, the chemical properties of the object, the quantity of the object, the container storing the object, and the specific location of the object;

[0033] The detonation information is the information describing the way of explosion of the explosion source in the accidental explosion event;

[0034] The damage information is the information describing various damage situations in the accidental explosion process. The damage information includes the depth of the explosion crater, the flying distance of fragments, the distance of the shock wave, the height of the flame, and the TNT explosion equivalent;

[0035] Step 2-2: Determine the explosion source type label of the historical accidental explosion event

[0036] According to the information of the historical accidental explosion event and the type of accidental explosion source investigated in Step 2-1, after adding the explosion source type label to the historical accidental explosion event, the historical accidental explosion event dataset S1 is obtained.

[0037] Furthermore, the on-site description information of the accidental explosion event to be traced includes explosion crater information, flame height information, TNT explosion equivalent information, shock wave information, and fragment information;

[0038] The explosion crater information is the depth of the explosion crater;

[0039] The flame height information is the height of the flame;

[0040] The shock wave information is the distance of the shock wave;

[0041] The fragment information is the flying distance of the fragments;

[0042] The TNT explosion equivalent information is the number of tons of TNT explosion equivalent to the power generated by the explosion.

[0043] Furthermore, the specific process of Step 4 is as follows:

[0044] Step 4-1: Obtain a keyword set Q = {q1, q2,..., q i , q i+1 ,..., q n} according to the on-site description information of the accidental explosion event to be traced;

[0045] Among them, q1 is the first keyword in the set Q, q2 is the second keyword in the set Q, q i is the i-th keyword in the set Q, q i+1 is the (i + 1)-th keyword in the set Q, q n is the n-th keyword in the set Q, and n is the total number of keywords;

[0046] Step 4-2: Define the calculation method of the relevance score as follows:

[0047]

[0048] Among them, D is a document composed of the explosion source information, detonation information, and damage information of an accidental explosion event in the historical accidental explosion event dataset S1, |D| is the number of accidental explosion events in the historical accidental explosion event dataset S1, avgdl is the average length of the documents corresponding to each accidental explosion event in the historical accidental explosion event dataset S1, b and k1 are adjustable parameters, f(q i , D) represents the number of times q i appears in the document D, IDF(q i ) represents the inverse document frequency of q i in the document set, and BMscore(D, Q) is the relevance score between the accidental explosion event to be traced and the accidental explosion event corresponding to the document D;

[0049] Step 4-3: Calculate the relevance scores between the accidental explosion event to be traced and each accidental explosion event in the historical accidental explosion event dataset S1 respectively, and then sort the obtained relevance scores in descending order;

[0050] Select the accidental explosion events corresponding to the top m relevance scores in the historical accidental explosion event dataset S1, and use all the selected accidental explosion events to form a similar event set S2.

[0051] Furthermore, the calculation method of IDF(q i ) is:

[0052] IDF(q i ) = log(|D| + 1)(N i+1) (2)

[0053] Among them, N i represents the number of documents containing q i in the document set composed of all the documents corresponding to the accidental explosion events in the historical accidental explosion event dataset S1.

[0054] Furthermore, the value range of the parameter b is [0, 1], and the value range of the parameter k1 is [1.2, 2].

[0055] Furthermore, the specific process of the fifth step is as follows:

[0056] Step 5-1. Discretization of damage information

[0057] 1) Pit depth

[0058] The pit depth in the range of [0, 0.5) m is taken as the first level, the pit depth in the range of [0.5, 1) m is taken as the second level, the pit depth in the range of [1, 3) m is taken as the third level, the pit depth in the range of [3, 5) m is taken as the fourth level, and the pit depth greater than or equal to 5 m is taken as the fifth level;

[0059] 2) Fragment flight distance

[0060] The fragment flight distance in the range of [0, 10) m is taken as the first level, the fragment flight distance in the range of [10, 30) m is taken as the second level, the fragment flight distance in the range of [30, 50) m is taken as the third level, the fragment flight distance in the range of [50, 100) m is taken as the fourth level, and the fragment flight distance greater than or equal to 100 m is taken as the fifth level;

[0061] 3) Shock wave distance

[0062] The shock wave distance in the range of [0, 10) m is taken as the first level, the shock wave distance in the range of [10, 50) m is taken as the second level, the shock wave distance in the range of [50, 100) m is taken as the third level, and the shock wave distance greater than or equal to 100 m is taken as the fourth level;

[0063] 4) Flame height

[0064] The flame height in the range of [0, 10) m is taken as the first level, the flame height in the range of [10, 30) m is taken as the second level, the flame height in the range of [30, 50) m is taken as the third level, the flame height in the range of [50, 100) m is taken as the fourth level, and the flame height greater than or equal to 100 m is taken as the fifth level;

[0065] 5) TNT explosive equivalent

[0066] Taking the TNT explosive equivalent in the range of [0, 0.5) tons as the first level, [0.5, 1) tons as the second level, [1, 2) tons as the third level, [2, 10) tons as the fourth level, [10, 20) tons as the fifth level, [20, 50) tons as the sixth level, [50, 100) tons as the seventh level, [100, 300) tons as the eighth level, [300, 500) tons as the ninth level, and the TNT explosive equivalent greater than or equal to 500 tons as the tenth level;

[0067] Step Five Two: For any accidental explosion event e in the similar event set S2, encode the discretized damage information of the accidental explosion event to obtain the feature vector of the accidental explosion event;

[0068] The feature vector of the accidental explosion event e is specifically {t1, t2, t3, …, t n , Btag}, where t i represents the feature of the damage information, and Btag represents the type of explosion source;

[0069] Similarly, obtain the feature vectors of each accidental explosion event in the similar event set S2; then encode the discretized damage information of the accidental explosion event to be traced to obtain the feature vector of the accidental explosion event to be traced.

[0070] Furthermore, the specific process of the said Step Six is as follows:

[0071] Step Six One: In the similar event set S2, use the Bayesian algorithm to calculate the probability P(Btag i ) of the distribution of each explosion source category and the conditional probability P(t|Btag i ) of the feature vector t of the accidental explosion event to be traced in the similar event set;

[0072] where i = 1, 2, …, N, P(Btag i ) is the probability of the i-th explosion source distribution, and P(t|Btag i ) is the conditional probability of the accidental explosion event to be traced relative to the i-th explosion source;

[0073] Step Six Two: Use the trained Bayesian classifier to calculate the probability P(Btag i |t) that the explosion source type of the accidental explosion event to be traced is Btag i , sort the calculated probabilities in descending order, and take the explosion source type corresponding to the maximum probability as the prediction result of the explosion source type;

[0074]

[0075] Among them, P(t) represents the probability that the explosion source feature vector t appears in the similar case set.

[0076] The beneficial effects of the present invention are as follows:

[0077] The method of the present invention uses existing accidental explosion events as prior knowledge, and conducts explosion source tracing by mining and analyzing the associations and laws between the damage information in the investigation data of the accidental explosion event to be traced and historical explosion events, so as to more accurately identify and track the source and damage impact of accidental explosion events. Compared with traditional manual explosion source tracing methods, the method of the present invention can greatly improve the efficiency of explosion source tracing, does not require the establishment of complex physical models, has lower requirements for data, simplifies the explosion source tracing process while improving the tracing accuracy, and ensures the feasibility and real-time nature of explosion source tracing. Description of the Drawings

[0078] Figure 1 It is a flowchart of an accidental explosion source tracing method based on a Bayesian classifier according to the present invention. Detailed Embodiments

[0079] Detailed Embodiment 1. In combination with Figure 1 This embodiment is described. An accidental explosion source tracing method based on a Bayesian classifier described in this embodiment specifically includes the following steps:

[0080] Step 1: Classify existing explosion sources, and a total of N types of explosion sources are determined;

[0081] Step 2: According to the explosion source types determined in Step 1, add corresponding explosion source type labels to historical accidental explosion events with known information to obtain a labeled historical accidental explosion event data set S1;

[0082] Step 3: The user inputs the on-site description information of the accidental explosion event to be traced;

[0083] Step 4: According to the on-site description information input by the user, calculate the correlation scores between the accidental explosion event to be traced and each accidental explosion event in the historical accidental explosion event data set S1 respectively, and then sort the obtained correlation scores in descending order;

[0084] Select the accidental explosion events corresponding to the top m correlation scores in the historical accidental explosion event data set S1, and use all the selected accidental explosion events to form a similar event set S2;

[0085] Step 5: Classify the damage information of the similar event set S2 and the accident explosion event to be traced back respectively according to the damage degree level, and realize the discretization of the damage information;

[0086] For any accident explosion event in the similar event set S2, encode the discretized damage information of the accident explosion event to obtain the feature vector of the accident explosion event; encode the discretized damage information of the accident explosion event to be traced back to obtain the feature vector of the accident explosion event to be traced back;

[0087] Step 6: Use the feature vector of the accident explosion event obtained in Step 5 and the Bayesian algorithm to calculate the probability of the accident explosion event to be traced back related to each explosion source category in the similar event set S2;

[0088] Then calculate the probability of the occurrence of various explosion sources under the condition of the on-site description information given by the user, and take the explosion source type with the highest probability as the final accident explosion traceability result.

[0089] The present invention mainly aims at predicting the type of accidental explosion (i.e., accidental explosion) source, so as to provide support for rapid disaster relief arrangements and avoid secondary disasters such as secondary explosions. By collecting, analyzing and identifying physical evidence, traces, residues, etc. at the explosion event site, the type of explosion source involved can be inferred, and further in-depth analysis of the problems in its manufacturing, storage and transportation processes can be carried out. Therefore, the method of the present invention can improve the accuracy of explosion traceability and provide evidence for reasoning about the explosion process and confirming the accident liability determination.

[0090] Specific Embodiment 2: The difference between this embodiment and Specific Embodiment 1 is that in Step 1, the value of N is 11.

[0091] Other steps and parameters are the same as those in Specific Embodiment 1.

[0092] Specific Embodiment 3: The difference between this embodiment and Specific Embodiment 1 or 2 is that the N types of explosion sources are respectively:

[0093] Type 1: High-pressure steam;

[0094] Type 2: Natural gas and coal gas;

[0095] Type 3: Chemical raw material gas;

[0096] Type 4: TNT;

[0097] Type 5: Ammonium nitrate;

[0098] Type 6: Black powder;

[0099] Type 7: Nitrocellulose;

[0100] Type 8: Nitroglycerin;

[0101] Class 9: Petroleum gasoline;

[0102] Class 10: Alcohol;

[0103] Class 11: Others.

[0104] Other steps and parameters are the same as those in the first or second specific implementation manner.

[0105] Specific implementation manner four: The difference between this implementation manner and one of the first to third specific implementation manners is that the specific process of step two is as follows:

[0106] Step 2-1: Investigate the information of historical accidental explosion events and the types of accidental explosion sources. The information of historical accidental explosion events includes time, location, explosion source information, detonation information, and damage information;

[0107] The explosion source information is the information of the object where the explosion occurs. The explosion source information includes the physical properties of the object, the chemical properties of the object, the quantity size of the object, the container for storing the object, and the specific location of the object;

[0108] The detonation information is the information describing the explosion occurrence method of the explosion source in the accidental explosion event;

[0109] The damage information is the information describing various damage situations in the accidental explosion process. The damage information includes the depth of the explosion crater, the flight distance of fragments, the distance of the shock wave, the height of the flame, and the TNT explosion equivalent (which refers to the power equivalent to how many tons of TNT explosion generated by the accidental explosion);

[0110] Step 2-2: Determine the explosion source type label of the historical accidental explosion event

[0111] After adding the explosion source type label to the historical accidental explosion event according to the information of the historical accidental explosion event and the type of accidental explosion source investigated in step 2-1, the historical accidental explosion event dataset S1 is obtained.

[0112] Manually mark the explosion source of each accidental explosion. If there are multiple explosion sources, the main explosion source type is used as the classification of the explosion source of this accidental explosion.

[0113] Other steps and parameters are the same as those in one of the first to third specific implementation manners.

[0114] Specific implementation manner five: The difference between this implementation manner and one of the first to fourth specific implementation manners is that the on-site description information of the accidental explosion event to be traced includes explosion crater information, flame height information, TNT explosion equivalent information, shock wave information, and fragment information;

[0115] The explosion crater information is the depth of the explosion crater;

[0116] The flame height information is the height of the flame;

[0117] The shock wave information is the distance of the shock wave;

[0118] The fragment information is the flying distance of the fragments;

[0119] The TNT explosion equivalent information is the number of tons of TNT equivalent to the power generated by the explosion.

[0120] According to the corresponding relationship between the number of tons of TNT and the power generated during a TNT explosion, the power generated by the explosion is obtained in terms of the power generated by the explosion of a certain number of tons of TNT.

[0121] Other steps and parameters are the same as those in any one of the specific embodiments one to four.

[0122] The present invention can trace the explosion using the crater formation information, flame height information, TNT explosion equivalent information, shock wave information, and fragment information, solving the problem of the long time required to obtain explosion information from multiple aspects in the existing methods, improving the efficiency of tracing, and ensuring the real-time nature of tracing.

[0123] Specific Embodiment Six: The difference between this embodiment and any one of the specific embodiments one to five is that the specific process of step four is as follows:

[0124] Step Four One: Obtain the keyword set Q = {q1, q2,..., q i , q i+1 ,..., q n} according to the on-site description information of the accidental explosion event to be traced;

[0125] Among them, q1 is the first keyword in the set Q, q2 is the second keyword in the set Q, q i is the i-th keyword in the set Q, q i+1 is the (i + 1)-th keyword in the set Q, q n is the n-th keyword in the set Q, and n is the total number of keywords;

[0126] Step Four Two: Define the calculation method of the relevance score as follows:

[0127]

[0128] Among them, D is the document composed of the explosion source information, detonation information, and damage information of an accidental explosion event in the historical accidental explosion event dataset S1, |D| is the number of accidental explosion events in the historical accidental explosion event dataset S1, avgdl is the average length of the documents corresponding to the accidental explosion events in the historical accidental explosion event dataset S1, b and k1 are adjustable parameters, f(qi , D) represents the number of occurrences of q i in document D, and IDF(q i ) represents q i inverse document frequency in the document set (which is composed of documents corresponding to each accidental explosion event in the historical accidental explosion event dataset), and BMscore(D, Q) is the correlation score between the accidental explosion event to be traced and the accidental explosion event corresponding to document D;

[0129] Step Four Three: Calculate the correlation scores between the accidental explosion event to be traced and each accidental explosion event in the historical accidental explosion event dataset S1 respectively, and then sort the obtained correlation scores in descending order;

[0130] Select the accidental explosion events corresponding to the top m correlation scores in the historical accidental explosion event dataset S1, and use all the selected accidental explosion events to form a similar event set S2.

[0131] Other steps and parameters are the same as those in one of the specific embodiments one to five.

[0132] Specific Embodiment Seven: Different from one of the specific embodiments one to six, the calculation method of the IDF(q i ) is as follows:

[0133] IDF(q i ) = log(|D| + 1)(N i + 1) (2)

[0134] where N i represents the number of documents containing q i in the document set composed of documents corresponding to all accidental explosion events in the historical accidental explosion event dataset S1.

[0135] Other steps and parameters are the same as those in one of the specific embodiments one to six.

[0136] In this embodiment, the base of log is 10.

[0137] Specific Embodiment Eight: Different from one of the specific embodiments one to seven, the value range of the parameter b is [0, 1], and the value range of the parameter k1 is [1.2, 2].

[0138] Other steps and parameters are the same as those in one of the specific embodiments one to seven.

[0139] The parameter b is used to control the degree of penalty of the document length on the weight. b = 0 means that the document length has no influence on the weight, and b = 1 means that the document length has a complete penalty effect on the weight. The parameter k1 controls the sensitivity of the formula to the word frequency. The smaller k1 is, the smaller the influence of the word frequency tf on the scoring formula.

[0140] Specific Embodiment Nine: The difference between this embodiment and one of Embodiments One to Eight is that the specific process of Step Five is as follows:

[0141] Step 5-1: Discretize the damage information

[0142] 1) Crater depth

[0143] A crater refers to a depression on the ground formed by the explosion throwing out soil and other substances on the ground. The crater depth in the range of [0, 0.5) meters is taken as the first level, the crater depth in the range of [0.5, 1) meters is taken as the second level, the crater depth in the range of [1, 3) meters is taken as the third level, the crater depth in the range of [3, 5) meters is taken as the fourth level, and the crater depth greater than or equal to 5 meters is taken as the fifth level;

[0144] 2) Fragment flight distance

[0145] The fragment flight distance in the range of [0, 10) meters is taken as the first level, the fragment flight distance in the range of [10, 30) meters is taken as the second level, the fragment flight distance in the range of [30, 50) meters is taken as the third level, the fragment flight distance in the range of [50, 100) meters is taken as the fourth level, and the fragment flight distance greater than or equal to 100 meters is taken as the fifth level;

[0146] 3) Shock wave distance

[0147] The shock wave distance in the range of [0, 10) meters is taken as the first level, the shock wave distance in the range of [10, 50) meters is taken as the second level, the shock wave distance in the range of [50, 100) meters is taken as the third level, and the shock wave distance greater than or equal to 100 meters is taken as the fourth level;

[0148] 4) Flame height

[0149] The flame height in the range of [0, 10) meters is taken as the first level, the flame height in the range of [10, 30) meters is taken as the second level, the flame height in the range of [30, 50) meters is taken as the third level, the flame height in the range of [50, 100) meters is taken as the fourth level, and the flame height greater than or equal to 100 meters is taken as the fifth level;

[0150] 5) TNT explosive equivalent

[0151] Take the TNT explosive equivalent in the range of [0, 0.5) tons as the 1st level, [0.5, 1) tons as the 2nd level, [1, 2) tons as the 3rd level, [2, 10) tons as the 4th level, [10, 20) tons as the 5th level, [20, 50) tons as the 6th level, [50, 100) tons as the 7th level, [100, 300) tons as the 8th level, [300, 500) tons as the 9th level, and the TNT explosive equivalent greater than or equal to 500 tons as the 10th level;

[0152] Step Five Two: For any accidental explosion event e in the similar event set S2, encode the discretized damage information of the accidental explosion event to obtain the feature vector of the accidental explosion event;

[0153] The feature vector of the accidental explosion event e is specifically {t1, t2, t3, …, t n , Btag}, where t i represents the feature of the damage information, and Btag represents the type of explosion source;

[0154] Similarly, obtain the feature vectors of each accidental explosion event in the similar event set S2; then encode the discretized damage information of the accidental explosion event to be traced to obtain the feature vector of the accidental explosion event to be traced.

[0155] Other steps and parameters are the same as those in any one of the specific implementation manners one to eight.

[0156] Specific Implementation Manner Ten: The difference between this implementation manner and any one of the specific implementation manners one to nine is that the specific process of the said Step Six is as follows:

[0157] Step Six One: In the similar event set S2, use the Bayesian algorithm to calculate the probability P(Btag i ) of the distribution of each explosion source category and the conditional probability P(t|Btag i ) of the feature vector t of the accidental explosion event to be traced in the similar event set;

[0158] where i = 1, 2, …, N, P(Btag i ) is the probability of the i-th explosion source distribution, and P(t|Btag i ) is the conditional probability of the accidental explosion event to be traced relative to the i-th explosion source;

[0159] Step Six Two: Use the trained Bayesian classifier to calculate the probability P(Btag i ) that the explosion source type of the accidental explosion event to be traced is Btagi |t), sort the calculated probabilities in descending order, and take the explosion source type corresponding to the maximum probability as the prediction result of the explosion source type;

[0160]

[0161] Among them, P(t) represents the probability that the explosion source feature vector t appears in the similar case set.

[0162] P(Btag i |t) is the posterior probability, indicating that a new explosion event has occurred, and knowing the information of the explosion event, judging the probability that its explosion source type is Btag i of.

[0163] The explosion source tracing method proposed by the present invention predicts the current explosion based on the historical similar explosion experience knowledge and rules, uses a Bayesian classifier, and combines the accidental explosion event investigation data set to effectively calculate the prior probability and conditional probability, and predicts the explosion source to complete the explosion source tracing.

[0164] Other steps and parameters are the same as those in any one of the first to ninth specific embodiments.

[0165] Experimental part

[0166] Suppose the explosion description given by the user is: "The depth of the explosion crater is 0.7 meters, the height of the flame is about 10 meters, and the fragments fly out 27 meters." The similar accidental explosion data set queried is shown in Table 1.

[0167] Table 1 Similar accidental explosion event set

[0168]

[0169]

[0170] Discretize the data in Table 1 to obtain the data in Table 2.

[0171] Table 2 Quantized features of similar accidental explosion events

[0172]

[0173] Calculate the probabilities of the three types of explosion source types 5, 6, and 7 respectively. Finally, it is calculated that the score of the 6th type of explosion source is the highest. Therefore, it is predicted that the explosion source type of this explosion is the 6th type. It is proved that the explosion source tracing method based on similarity query + Bayesian (i.e., the method of the present invention) is effective and feasible.

[0174] To further verify the feasibility of the explosion source tracing method based on similarity query + Bayesian, the query conditions given by users when an actual accidental explosion occurs are simulated. Therefore, only the event element information included in the query condition Q given by the user is considered, mainly including the crater depth of the explosion, the flying distance of fragments, the distance of the shock wave, the height of the flame, and the TNT explosion equivalent information. The situation where the user queries a specific case by event name, location, and time is not considered; in addition, since the user's purpose is to predict the type of the explosion source, the information of the explosion source is not included in the constructed query condition Q. The explosion source tracing results of each comparison method are shown in Table 3.

[0175] Table 3 Bayesian-based explosion source tracing prediction results

[0176]

[0177] As can be seen from Table 3: Compared with Method 2 (explosion source tracing based only on the Bayesian algorithm), the accuracy rate of Method 1 (the present invention) has increased by 7.5%. This is because Method 1 pre-selects cases that match the characteristics of the explosion site through conditional query and similarity calculation, effectively eliminating irrelevant data and noise data that do not match the current explosion, and using data similar to the instance to be classified as training data, which can effectively improve the accuracy rate of Bayesian classification; while in Method 2, the entire accidental explosion event data is directly used as the input of the Bayesian algorithm, which will input a lot of irrelevant explosion case information, thereby deteriorating the effect of Bayesian classification.

[0178] Compared with Method 3 (explosion source tracing based only on the SVM algorithm), the accuracy rate of explosion source tracing of Method 1 has increased by 10.0%. This is because the explosion source tracing method proposed in the present invention first screens out cases with similar characteristics from the investigation data according to the explosion site conditions, so the prior probability of historical cases can be used to classify the data, thereby effectively improving the classification accuracy of the explosion source. At the same time, the Bayesian algorithm can better handle noise and missing values, while the SVM is sensitive to noise and missing values and depends on the exact positions of all training data. The SVM needs to map the original data to a high-dimensional space and has poor adaptability to the data. Therefore, the Bayesian algorithm is more suitable for source tracing in the field of accidental explosion damage.

[0179] Compared with Method 4 (explosion source tracing of similarity query + SVM), the accuracy rate of Method 1 has increased by 3.5%. This is because both methods screen out a set of similar cases according to the on-site description of the accidental explosion, which leads to a reduction in the amount of data. There is inevitably a problem of unbalanced explosion source categories in the candidate case set, and the Bayesian classifier can handle the problem of class imbalance by adjusting the prior probability. And compared with the SVM, the Bayesian classifier can reduce the influence of noise data and abnormal data through the prior probability and likelihood function.

[0180] The above numerical examples of the present invention are only for illustrating in detail the calculation model and calculation process of the present invention, rather than limiting the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is impossible to enumerate all the implementation manners here. Any obvious changes or modifications derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. A method for tracing the source of an accidental explosion based on a Bayesian classifier, characterized in that: The method specifically comprises the following steps: Step 1: Classify the explosion sources and determine N types of explosion sources; Step 2: According to the explosion source type determined in step 1, add the corresponding explosion source type label to the historical accidental explosion events with known information, and obtain the labeled historical accidental explosion event data set S1; Step 3: The user inputs the on-site description information of the accidental explosion event to be traced; The on-site description information of the accidental explosion event to be traced includes explosion pit information, flame height information, TNT explosion equivalent information, shock wave information, and fragment information; The explosion pit information is the depth of the explosion pit; The flame height information is the height of the flame; The shock wave information is the distance of the shock wave; The fragment information is the flight distance of the fragment; TNT explosion equivalent information is the power generated by the explosion equivalent to the tons of TNT explosion; Step 4: Calculate the correlation scores between the accidental explosion event to be traced and each accidental explosion event in the historical accidental explosion event data set S1 according to the scene description information input by the user, and then sort the obtained correlation scores in descending order; Select the accidental explosion events corresponding to the top m correlation scores in the historical accidental explosion event data set S1, and use all the selected accidental explosion events to form a similar event set S2; Step 5: Classify the damage degree of the similar event set S2 and the damage information of the accidental explosion event to be traced, so as to discretize the damage information; For any accidental explosion event in the similar event set S2, the discretized damage information of the accidental explosion event is encoded to obtain a feature vector of the accidental explosion event; the discretized damage information of the accidental explosion event to be traced is encoded to obtain a feature vector of the accidental explosion event to be traced; Step 6: Calculate the probability of the accidental explosion event to be traced being associated with each explosion source category in the similar event set S2 using the feature vector of the accidental explosion event obtained in step 5 and the Bayesian algorithm; Then calculate the probability of various explosion sources appearing when the user provides the scene description information, and take the explosion source type with the highest probability as the final accidental explosion source tracing result.

2. The method for tracing the source of an accidental explosion based on a Bayesian classifier according to claim 1 is characterized in that: In the step 1, the value of N is 11.

3. The method for tracing the source of an accidental explosion based on a Bayesian classifier according to claim 2 is characterized in that: The N types of explosion sources are: Category 1: High-pressure steam; Category 2: Natural gas and coal gas; Category 3: Chemical raw material gas; Category 4: TNT; Category 5: ammonium nitrates; Category 6: Black powder; Category 7: nitrocellulose; Category 8: Nitroglycerin; Class 9: Petroleum gasoline; Category 10: Alcohol; Category 11: Others.

4. The method for tracing the source of an accidental explosion based on a Bayesian classifier according to claim 3 is characterized in that: The specific process of step 2 is as follows: Step 21: Investigate the information of historical accidental explosions and the types of accidental explosion sources, wherein the information of historical accidental explosions includes time, location, explosion source information, detonation information and damage information; The explosion source information is the information of the object that exploded, including the physical properties of the object, the chemical properties of the object, the size of the object, the container in which the object is stored, and the specific location of the object; Detonation information is information describing the way the explosion of the explosion source occurred in an accidental explosion event; Damage information is information describing various damage conditions during an accidental explosion. Damage information includes explosion pit depth, fragment flight distance, shock wave distance, flame height, and TNT explosion equivalent. Step 2: Determine the explosion source type label of historical accidental explosion events According to the information of the historical accidental explosion events investigated in step 21 and the types of accidental explosion sources, after adding explosion source type labels to the historical accidental explosion events, a historical accidental explosion event dataset S1 is obtained.

5. The method for tracing the source of an accidental explosion based on a Bayesian classifier according to claim 4 is characterized in that: The specific process of step 4 is as follows: Step 41: Obtain a keyword set Q = {q1, q2, ..., q i ,q i+1 ,...,q n }; Among them, q1 is the first keyword in set Q, q2 is the second keyword in set Q, and q i is the i-th keyword in set Q, q i+1 is the i+1th keyword in the set Q, q n is the nth keyword in set Q, where n is the total number of keywords; Step 4.2: Define the calculation method of the relevance score as follows: Where D is a document consisting of the explosion source information, detonation information, and damage information of an accidental explosion event in the historical accidental explosion event dataset S1, |D| is the number of accidental explosion events in the historical accidental explosion event dataset S1, avgdl is the average length of the documents corresponding to each accidental explosion event in the historical accidental explosion event dataset S1, b and k1 are adjustable parameters, and f(q i ,D) represents q i The number of times it appears in document D, IDF(q i ) indicates q i Inverse document frequency in the document set, BMscore(D,Q) is the correlation score between the accidental explosion event to be traced and the accidental explosion event corresponding to document D; Step 43: Calculate the correlation scores between the accidental explosion event to be traced and each accidental explosion event in the historical accidental explosion event data set S1, and then sort the obtained correlation scores in descending order; The accidental explosion events corresponding to the top m correlation scores in the historical accidental explosion event data set S1 are selected, and all the selected accidental explosion events are used to form a similar event set S2.

6. The method for tracing the source of an accidental explosion based on a Bayesian classifier according to claim 5 is characterized in that: The IDF(q i ) is calculated as: IDF(q i )=log(|D|+1)(N i +1) (2) Among them, N i It represents a document set consisting of documents corresponding to all accidental explosion events in the historical accidental explosion event dataset S1, including q i The number of documents.

7. The method for tracing the source of an accidental explosion based on a Bayesian classifier according to claim 6 is characterized in that: The value range of the parameter b is [0, 1], and the value range of the parameter k1 is [1.2, 2].

8. The method for tracing the source of an accidental explosion based on a Bayesian classifier according to claim 7 is characterized in that: The specific process of step five is as follows: Step 51: Discretization of damage information 1) Pit depth The pit depth of [0, 0.5) meters is regarded as the first level, the pit depth of [0.5, 1) meters is regarded as the second level, the pit depth of [1, 3) meters is regarded as the third level, the pit depth of [3, 5) meters is regarded as the fourth level, and the pit depth greater than or equal to 5 meters is regarded as the fifth level; 2) Fragment flight distance The fragmentation distance of [0,10) meters is regarded as the first level, the fragmentation distance of [10,30) meters is regarded as the second level, the fragmentation distance of [30,50) meters is regarded as the third level, the fragmentation distance of [50,100) meters is regarded as the fourth level, and the fragmentation distance greater than or equal to 100 meters is regarded as the fifth level; 3) Shock wave distance The shock wave distance of [0,10) meters is regarded as the first level, the shock wave distance of [10,50) meters is regarded as the second level, the shock wave distance of [50,100) meters is regarded as the third level, and the shock wave distance greater than or equal to 100 meters is regarded as the fourth level; 4) Flame height The flame height of [0,10) meters is regarded as the first level, the flame height of [10,30) meters is regarded as the second level, the flame height of [30,50) meters is regarded as the third level, the flame height of [50,100) meters is regarded as the fourth level, and the flame height greater than or equal to 100 meters is regarded as the fifth level; 5) TNT explosion equivalent The first level is [0,0.5) tons of TNT, the second level is [0.5,1) tons of TNT, the third level is [1,2) tons of TNT, the fourth level is [2,10) tons of TNT, the fifth level is [10,20) tons of TNT, the sixth level is [20,50) tons of TNT, the seventh level is [50,100) tons of TNT, the eighth level is [100,300) tons of TNT, the ninth level is [300,500) tons of TNT, and the tenth level is greater than or equal to 500 tons of TNT. Step 52: for any accidental explosion event e in the similar event set S2, the discretized damage information of the accidental explosion event is encoded to obtain a feature vector of the accidental explosion event; The characteristic vector of the accidental explosion event e is specifically {t1, t2, t3, …, t n ,Btag}, where t i Indicates the characteristics of damage information, Btag indicates the type of explosion source; Similarly, the feature vectors of each accidental explosion event in the similar event set S2 are obtained respectively; then the discretized damage information of the accidental explosion event to be traced is encoded to obtain the feature vector of the accidental explosion event to be traced.

9. The method for tracing the source of an accidental explosion based on a Bayesian classifier according to claim 8, characterized in that: The specific process of step six is ​​as follows: Step 6. In the similar event set S2, use the Bayesian algorithm to calculate the probability P (Btag i ) and the conditional probability P(t|Btag) of the feature vector t of the accidental explosion event to be traced in the set of similar events i ); Where i = 1, 2, ..., N, P (Btag i ) is the probability of the ith explosion source distribution, P(t|Btag i ) is the conditional probability of the accidental explosion event to be traced relative to the i-th explosion source; Step 6.2: Use the trained Bayesian classifier to calculate the explosion source type of the accidental explosion event to be traced as Btag i The probability P(Btag i |t), sort the calculated probabilities in descending order, and take the explosion source type corresponding to the largest probability as the explosion source type prediction result; Among them, P(t) represents the probability of the explosion source feature vector t appearing in a set of similar cases.

Images