Dangerous cargo risk early warning method and system and computer program

Abstract

The invention discloses a dangerous cargo risk early warning method and system and a computer program. The dangerous cargo risk early warning method comprises the steps of obtaining multi-source monitoring data; analyzing the multi-source monitoring data through a preset rule engine and a trained dangerous cargo risk prediction model to obtain a risk prediction result; and performing early warning according to a risk prediction result. According to the technical scheme provided by the invention, the historical and real-time multi-source monitoring data is acquired, and the rule engine and the machine learning algorithm are used for predicting dangerous cargo risk factors, so that dangerous cargo risk early warning is more real-time and accurate; compared with manual inspection and video monitoring, data can be acquired more frequently, and key information is synchronized in a data center; multiple data sources are covered, and the influence of various factors on a monitoring station can be considered more comprehensively; by means of automatic data collection and intelligent data analysis, manpower resources can be saved, problems and risks can be found more timely, and therefore potential loss and cost are reduced.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a dangerous goods risk early warning method, system and computer program. Background Art

[0002] As the country attaches great importance to production safety, how to quickly and accurately detect risks in the transportation and storage of goods has also received increasing attention. In the transportation and storage links, there are many risk factors, such as unstable stacking of goods, equipment failure, and operator errors.

[0003] At present, the commonly used dangerous goods monitoring methods are based on manual inspection and video monitoring, but these methods have problems such as low efficiency, easy to miss detection, difficult to detect in time, and easy to be affected by human factors. Therefore, in order to more quickly and accurately detect the risks in the terminal transportation link, an intelligent dangerous goods monitoring method is urgently needed. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the current technology, the present invention provides a dangerous goods risk warning method, which obtains historical and real-time multi-source monitoring data, uses a rule engine and machine learning algorithm to predict dangerous goods risk factors, and makes the dangerous goods risk warning more real-time and accurate.

[0005] To achieve the above object, the embodiments of the present invention adopt the following technical solutions: A dangerous goods risk early warning method comprises the following steps: Acquire multi-source monitoring data; Through the preset rule engine and the trained dangerous goods risk prediction model, the multi-source monitoring data is analyzed to obtain the risk prediction results; Issue early warning based on risk prediction results; Wherein, the hazardous cargo risk prediction model is a decision tree model, and the tree leaf nodes of the decision tree store the probability of accident occurrence under the data feature path; The analysis of multi-source monitoring data through a preset rule engine and a trained dangerous goods risk prediction model includes: Through the preset rule engine, the multi-source monitoring data is analyzed to obtain the first risk prediction result; Through the trained dangerous goods risk prediction model, the multi-source monitoring data is analyzed to obtain the probability of accidents and calculate the safety score; The risk prediction result is obtained by combining the first risk prediction result and the safety score.

[0006] According to one aspect of the present invention, the dangerous goods risk warning method further includes: Preprocess multi-source monitoring data.

[0007] According to one aspect of the present invention, the step of analyzing multi-source monitoring data through a preset rule engine to obtain risk prediction results includes: Obtain cargo yard data; Set monitoring data alarm rules based on cargo yard data; The multi-source monitoring data is traversed and matched with the alarm rules one by one. Once data that matches the alarm rules is found, the rule engine will trigger the corresponding alarm mechanism.

[0008] According to one aspect of the present invention, the dangerous goods risk warning method further includes: Obtain historical multi-source monitoring data; Based on historical multi-source monitoring data, a decision tree model is constructed and trained to build a dangerous goods risk prediction model.

[0009] According to one aspect of the present invention, the construction and training of a decision tree model based on historical multi-source monitoring data includes: The decision tree selects the optimal features and thresholds for segmentation based on the features and target variables in the training data set to form a tree structure; the segmentation process is repeated until the stopping condition is met.

[0010] According to one aspect of the present invention, the calculation formula of the safety score is: s=m*p+b Where: s is the safety score; p is the probability of an accident; m is the slope, which is used to describe how quickly the score changes with probability; b is the intercept, which represents the score when p=0.

[0011] According to one aspect of the present invention, the dangerous goods risk warning method further includes: Regularly use the latest multi-source monitoring data and actual risk conditions to train and optimize the hazardous goods risk prediction model.

[0012] According to one aspect of the present invention, the method of analyzing multi-source monitoring data through a preset rule engine and a trained dangerous goods risk prediction model to obtain a risk prediction result includes: According to the preset rule engine, the multi-source monitoring data is analyzed to obtain the first risk prediction result; Analyze the multi-source monitoring data according to the trained dangerous goods risk prediction model to obtain the second risk prediction result; The first risk prediction result and the second risk prediction result are combined to obtain a risk prediction result.

[0013] A dangerous goods risk warning system, based on the dangerous goods risk warning method as described above, comprises: Data acquisition module, used to acquire multi-source monitoring data; The prediction module is used to analyze multi-source monitoring data through a preset rule engine and a trained dangerous goods risk prediction model to obtain risk prediction results; The early warning module is used to issue early warnings based on risk prediction results.

[0014] A computer program which, when executed, implements the steps of the dangerous goods risk warning method as described above.

[0015] Advantages of the present invention: The present invention provides a dangerous goods risk warning method, which obtains historical and real-time multi-source monitoring data, uses a rule engine and a machine learning algorithm to predict dangerous goods risk factors, and makes the dangerous goods risk warning more real-time and accurate; the use of automated data collection and intelligent data analysis can save human resources, and can discover problems and risks more timely, thereby reducing potential losses and costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 This is a flow chart of a dangerous goods risk warning method according to Embodiment 1 of the present invention; Figure 2 This is a flow chart of a dangerous goods risk warning method according to Embodiment 2 of the present invention; Figure 3 This is a flow chart of a dangerous goods risk warning method according to Embodiment 3 of the present invention; Figure 4 This is a flow chart of a dangerous goods risk warning according to Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Embodiment 1

[0020] like Figure 1 As shown, a dangerous goods risk warning method includes the following steps: S1: Obtain multi-source monitoring data.

[0021] In actual applications, a variety of monitoring devices and systems are used to collect relevant data, including business flow data, sensor data, weather forecast data, etc.

[0022] For example, business flow data (returned by the Big Data Center) include: daily port throughput, in tons; the number of cargo ships operating at the port; the deadweight tonnage of each cargo ship; cargo types, including but not limited to ore, coal, grain, etc.; the operating time of each cargo ship, in hours; specific berthing and unberthing times; cargo loading and unloading efficiency, in tons and hours, etc.

[0023] Sensor data includes: ambient temperature, accurate to one decimal place; relative humidity, as a percentage; atmospheric pressure, in Pascals; wind speed, in meters per second; wind direction, accurate to 10° intervals, etc.

[0024] Weather forecast data includes: weather conditions, covering sunny, cloudy, rainy, thunderstorm and other states; expected temperature range, including maximum and minimum temperatures; wind force level, 12 levels from light breeze to strong wind; probability of precipitation, expressed as a percentage; visibility, in kilometers, etc.

[0025] The obtained multi-source monitoring data can be saved in json format.

[0026] Preferably, the dangerous goods risk warning method further includes: Preprocess multi-source monitoring data.

[0027] Preprocess the data to ensure the consistency and reliability of subsequent analysis and application. Data preprocessing mainly includes data cleaning, standardization, normalization, labeling and data feature engineering.

[0028] In practical applications, the collected data is often complicated and needs to be cleaned, standardized, and normalized. Among them, data cleaning mainly includes: removing duplicate data, processing missing values, and identifying and processing outliers.

[0029] Missing value processing: For missing values ​​of important data such as cargo ship operation time and tonnage, if the missing rate is less than 5%, the average value is used to fill in; if it is between 5% and 15%, the median is used to fill in; for fields exceeding 15%, reasonable estimates need to be made based on historical data trends or expert opinions.

[0030] Identification and processing of outliers: Outliers are identified using the 3σ principle. For values ​​that deviate significantly from the normal range, the average of the adjacent values ​​is used for correction.

[0031] Data standardization and normalization: Z-score standardization is applicable to numerical data such as port throughput and tonnage, ensuring that the data is distributed in a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0032] This method needs to mine the patterns of past monitoring data to analyze the current monitoring data and obtain risk prediction results. Therefore, it is necessary to annotate the historical monitoring data and correspond it to the actual dangerous goods risk at that time in order to design the rule engine.

[0033] In actual applications, the collected data is diverse, and not all data can reflect the risk of hazardous goods. Therefore, multi-source monitoring data can be identified to find out the key features that affect the safety of goods. For example, key features include: yard capacity, flammable items, explosive items, operation time, weather conditions, etc.

[0034] Different features contribute differently to risk prediction. Statistical methods can be used to evaluate and rank the contribution of key features to risk prediction. When designing a rule engine, priority or weight can be set based on the contribution to process monitoring data. For example, the contribution ranking of the above key features is: flammable items > explosive items > weather conditions > yard capacity > operation time.

[0035] S2: Analyze multi-source monitoring data through the preset rule engine to obtain risk prediction results.

[0036] Step S2 includes: S21: Acquire cargo yard data.

[0037] In practical applications, the rule engine needs to be designed based on cargo yard data.

[0038] The cargo yard data includes: yard name; yard capacity, generally in square meters; whether large items and flammable items are allowed to be stacked; operation entry time, etc. For example, there are 3 yards, the yard names are: Yard A, Yard B, Yard C, and the yard capacities are: 10,000 square meters, 5,000 square meters, and 15,000 square meters respectively.

[0039] S22: According to the cargo yard data, set monitoring data alarm rules.

[0040] The alarm rules of the rule engine need to be set according to the actual needs of dangerous goods risk warning and the characteristics of monitoring data. Specifically, the alarm rules can be set to trigger an alarm when a key feature in the monitoring data reaches a certain threshold, for example, when flammable or explosive items are found in the goods, the temperature of the monitoring site exceeds the normal temperature, etc.; or when several monitoring data reach a certain threshold, the alarm is triggered; or when the thresholds of several monitoring data are compared and the weights are calculated, and a certain risk score is reached, the alarm is triggered.

[0041] For example, in practical applications, a series of rules can be set: Rules for stacking large items: The yards where stacking is allowed include Yard A and Yard C; the yard where stacking is prohibited is Yard B.

[0042] Rules for flammable items: The permitted storage yard is storage yard C; the prohibited storage yards include storage yard A and storage yard B. When flammable items appear in storage yard A or storage yard B, an alarm is triggered.

[0043] Operation time rules: Yard A is "8:00-18:00"; Yard B is "8:00-19:00"; Yard C is "6:00-20:00".

[0044] Temperature rule: When the temperature in the monitored yard is higher than 60°C, an alarm is triggered.

[0045] S23: Traverse the multi-source monitoring data and match them with the alarm rules one by one. Once data that matches the alarm rules is found, the rule engine will trigger the corresponding alarm mechanism.

[0046] Through a series of rules, non-compliance phenomena are screened out and risks are identified; once suspicious problems that match the rules are found, the rule engine will trigger the corresponding alarm mechanism. For example, when weather data shows that the temperature exceeds 60°C and there are flammable items in the yard, the rule engine will immediately identify this situation and notify the relevant administrator in a timely manner.

[0047] S3: Issue early warning based on risk prediction results.

[0048] When the rule engine finds the risk of hazardous goods based on monitoring data, it will issue an early warning in a timely manner.

[0049] The beneficial effects of this embodiment are: compared with manual inspections and video surveillance, this method can obtain data more frequently, and can synchronize key information in the data center, making dangerous goods risk warnings more real-time and accurate; it covers a variety of data sources, including sensor data, weather forecast data, etc., so that data analysis can more comprehensively consider the impact of various factors on monitoring sites; using a rule engine for risk prediction can timely detect dangerous goods risks and issue warnings; using automated data collection and intelligent data analysis can save human resources, and can detect problems and risks more promptly, thereby reducing potential losses and costs.

[0050] Embodiment 2

[0051] like Figure 2 As shown, a dangerous goods risk warning method includes the following steps: S1: Obtain multi-source monitoring data.

[0052] In actual applications, various monitoring devices and systems are used to collect relevant data, including business flow data, sensor data, weather forecast data, etc. For example, business flow data (returned by the Big Data Center) includes: daily port throughput in tons; the number of cargo ships operating in the port; the deadweight tonnage of each cargo ship; the type of cargo, including but not limited to ore, coal, grain, etc.; the operating time of each cargo ship in hours; the specific berthing and unberthing time; cargo loading and unloading efficiency in tons and hours, etc.

[0053] Sensor data includes: ambient temperature, accurate to one decimal place; relative humidity, as a percentage; atmospheric pressure, in Pascals; wind speed, in meters per second; wind direction, accurate to 10° intervals, etc.

[0054] Weather forecast data includes: weather conditions, covering sunny, cloudy, rainy, thunderstorm and other states; expected temperature range, including maximum and minimum temperatures; wind force level, 12 levels from light breeze to strong wind; probability of precipitation, expressed as a percentage; visibility, in kilometers, etc.

[0055] The obtained multi-source monitoring data can be saved in json format.

[0056] Preferably, the dangerous goods risk warning method further includes: Preprocess multi-source monitoring data.

[0057] Preprocess the data to ensure the consistency and reliability of subsequent analysis and application. Data preprocessing mainly includes data cleaning, standardization, normalization, labeling and data feature engineering.

[0058] In practical applications, the collected data is often complicated and needs to be cleaned, standardized, and normalized. Among them, data cleaning mainly includes: removing duplicate data, processing missing values, and identifying and processing outliers.

[0059] Missing value processing: For missing values ​​of important data such as cargo ship operation time and tonnage, if the missing rate is less than 5%, the average value is used to fill in; if it is between 5% and 15%, the median is used to fill in; for fields exceeding 15%, reasonable estimates need to be made based on historical data trends or expert opinions.

[0060] Identification and processing of outliers: Outliers are identified using the 3σ principle. For values ​​that deviate significantly from the normal range, the average of the adjacent values ​​is used for correction.

[0061] Data standardization and normalization: Z-score standardization is applicable to numerical data such as port throughput and tonnage, ensuring that the data is distributed in a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0062] This method needs to mine the patterns of past monitoring data to analyze the current monitoring data and obtain risk prediction results. Therefore, it is necessary to annotate the historical monitoring data and correspond it to the actual dangerous goods risk at that time in order to design the rule engine.

[0063] In actual applications, the collected data is diverse, and not all data can reflect the risk of hazardous goods. Therefore, multi-source monitoring data can be identified to find out the key features that affect the safety of goods. For example, key features include: yard capacity, flammable items, explosive items, operation time, weather conditions, etc.

[0064] Different features contribute differently to risk prediction. Statistical methods can be used to evaluate and rank the contribution of key features to risk prediction. When designing a rule engine, priority or weight can be set based on the contribution to process monitoring data. For example, the contribution ranking of the above key features is: flammable items > explosive items > weather conditions > yard capacity > operation time.

[0065] S2: Analyze multi-source monitoring data through the trained hazardous goods risk prediction model to obtain risk prediction results.

[0066] The method further comprises: Obtain historical multi-source monitoring data; Based on historical multi-source monitoring data, a decision tree model is constructed and trained to build a dangerous goods risk prediction model.

[0067] The construction and training of a decision tree model based on historical multi-source monitoring data includes: The decision tree selects the optimal features and thresholds for segmentation based on the features and target variables in the training data set to form a tree structure. The segmentation process is repeated until the stopping condition is met. The decision tree leaf nodes store the probability of accidents occurring under the data feature path.

[0068] Specifically, the decision tree construction process is as follows: During the training phase, the decision tree will select the optimal features and thresholds for segmentation based on the features in the training data set and the target variable (here, whether an accident has occurred) to form a tree structure. This process will be repeated until the stopping condition is met, that is, a certain set threshold is met, the depth of the decision tree or the number of samples in the node is lower than a certain threshold, etc.

[0069] In the prediction stage, for a new observation, the model will go down the decision tree according to its feature value until it reaches a leaf node, which contains the category probability distribution calculated based on the training data and stores two types of samples (accidents and no accidents) or proportions. For example, the model will go to a leaf node along the decision path based on the characteristics of the day (throughput, wind speed, transportation efficiency, etc.). According to historical data, the proportion of samples with accidents in all samples is 18.5%, so the model will predict that the probability of accidents on that day is 18.5%.

[0070] Preferably, the method further comprises: Regularly use the latest multi-source monitoring data and actual risk conditions to train and optimize the hazardous goods risk prediction model.

[0071] In practical applications, in order to ensure the effectiveness of the hazardous goods risk prediction model, the latest monitoring data and actual risk conditions are regularly used to train and adjust the model, adjust the splitting criteria of the decision tree, and further improve the recognition accuracy of the model to optimize its performance.

[0072] Step S2 includes: S21: Analyze the multi-source monitoring data through the trained hazardous goods risk prediction model to obtain the accident probability p.

[0073] S22: Obtain a safety score s based on the accident probability p and the predetermined accident classification level.

[0074] Specifically, the accident classification levels are: Low risk: the probability of an accident occurring is less than or equal to 5%; Medium risk: the probability of an accident occurring is between 5% and 20%; High risk: the probability of an accident occurring is between 20% and 40%; Extremely high risk: The probability of an accident occurring p is greater than 40%.

[0075] For example, if the predicted probability p of an accident occurring on that day is 18.5%, the accident level is classified as medium risk.

[0076] The calculation formula for the safety score s is: s=m*p+b; where: m is the slope, which is used to describe the speed at which the score changes with probability; b is the intercept, which represents the score when p=0.

[0077] Low risk (accident probability p is less than or equal to 5%): safety score s is between 0 and 1.00; Medium risk (accident probability p between 5% and 20%): safety score s between 1.01 and 2.99; High risk (accident probability p between 20% and 40%): safety score s between 3.00 and 4.00; Extremely high risk (accident probability p greater than 40%): Safety score s is between 4.01 and 5.00.

[0078] It can be set to issue an early warning when the safety score s exceeds 3 points.

[0079] S3: Issue early warning based on risk prediction results.

[0080] When the rule engine finds the risk of hazardous goods based on monitoring data, it will issue an early warning in a timely manner.

[0081] The beneficial effect of this embodiment is that the method can integrate historical data and real-time data, use machine learning algorithms to predict risk factors, and improve the accuracy and efficiency of analysis.

[0082] Embodiment 3

[0083] like Figure 3 and 4 As shown, a dangerous goods risk warning method includes the following steps: S1: Obtain multi-source monitoring data.

[0084] In actual applications, a variety of monitoring devices and systems are used to collect relevant data, including business flow data, sensor data, weather forecast data, etc.

[0085] For example, business flow data (returned by the Big Data Center) include: daily port throughput, in tons; the number of cargo ships operating at the port; the deadweight tonnage of each cargo ship; cargo types, including but not limited to ore, coal, grain, etc.; the operating time of each cargo ship, in hours; specific berthing and unberthing times; cargo loading and unloading efficiency, in tons and hours, etc.

[0086] Sensor data includes: ambient temperature, accurate to one decimal place; relative humidity, as a percentage; atmospheric pressure, in Pascals; wind speed, in meters per second; wind direction, accurate to 10° intervals, etc.

[0087] Weather forecast data includes: weather conditions, covering sunny, cloudy, rainy, thunderstorm and other states; expected temperature range, including maximum and minimum temperatures; wind force level, 12 levels from light breeze to strong wind; probability of precipitation, expressed as a percentage; visibility, in kilometers, etc.

[0088] The obtained multi-source monitoring data can be saved in json format.

[0089] Preferably, the dangerous goods risk warning method further includes: Preprocess multi-source monitoring data.

[0090] Preprocess the data to ensure the consistency and reliability of subsequent analysis and application. Data preprocessing mainly includes data cleaning, standardization, normalization, labeling and data feature engineering.

[0091] In practical applications, the collected data is often complicated and needs to be cleaned, standardized, and normalized. Among them, data cleaning mainly includes: removing duplicate data, processing missing values, and identifying and processing outliers.

[0092] Missing value processing: For missing values ​​of important data such as cargo ship operation time and tonnage, if the missing rate is less than 5%, the average value is used to fill in; if it is between 5% and 15%, the median is used to fill in; for fields exceeding 15%, reasonable estimates need to be made based on historical data trends or expert opinions.

[0093] Identification and processing of outliers: Outliers are identified using the 3σ principle. For values ​​that deviate significantly from the normal range, the average of the adjacent values ​​is used for correction.

[0094] Data standardization and normalization: Z-score standardization is applicable to numerical data such as port throughput and tonnage, ensuring that the data is distributed in a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0095] This method needs to mine the patterns of past monitoring data to analyze the current monitoring data and obtain risk prediction results. Therefore, it is necessary to annotate the historical monitoring data and correspond it to the actual dangerous goods risk at that time in order to design the rule engine.

[0096] In actual applications, the collected data is diverse, and not all data can reflect the risk of hazardous goods. Therefore, multi-source monitoring data can be identified to find out the key features that affect the safety of goods. For example, key features include: yard capacity, flammable items, explosive items, operation time, weather conditions, etc.

[0097] Different features contribute differently to risk prediction. Statistical methods can be used to evaluate and rank the contribution of key features to risk prediction. When designing a rule engine, priority or weight can be set based on the contribution to process monitoring data. For example, the contribution ranking of the above key features is: flammable items > explosive items > weather conditions > yard capacity > operation time.

[0098] S2: Analyze multi-source monitoring data through the preset rule engine and the trained hazardous goods risk prediction model to obtain risk prediction results.

[0099] Step S2 includes: S21: Analyze the multi-source monitoring data according to the preset rule engine to obtain a first risk prediction result.

[0100] S22: Analyze the multi-source monitoring data according to the trained dangerous goods risk prediction model to obtain a second risk prediction result.

[0101] S23: Combining the first risk prediction result and the second risk prediction result to obtain a risk prediction result.

[0102] In practical applications, the monitoring data can be independently analyzed using the rule engine and the hazardous goods risk prediction model. When either the rule engine or the hazardous goods risk prediction model detects a risk, an early warning will be issued.

[0103] In practical applications, the risk prediction results of the rule engine and the dangerous goods risk prediction model can also be combined to obtain the prediction results, and when risks are found, early warnings can be issued. The comprehensive method can be to weight the prediction results according to the accuracy of the predictions of the two to obtain the final risk prediction score. When the score exceeds a certain threshold, an early warning is issued.

[0104] Step S21 includes: S211: Acquire cargo yard data.

[0105] In practical applications, the rule engine needs to be designed based on cargo yard data.

[0106] The cargo yard data includes: yard name; yard capacity, generally in square meters; whether large items and flammable items are allowed to be stacked; operation entry time, etc. For example, there are 3 yards, the yard names are: Yard A, Yard B, Yard C, and the yard capacities are: 10,000 square meters, 5,000 square meters, and 15,000 square meters respectively.

[0107] S212: Setting monitoring data alarm rules according to cargo yard data.

[0108] The alarm rules of the rule engine need to be set according to the actual needs of dangerous goods risk warning and the characteristics of monitoring data. Specifically, the alarm rules can be set to trigger an alarm when a key feature in the monitoring data reaches a certain threshold, for example, when flammable or explosive items are found in the goods, the temperature of the monitoring site exceeds the normal temperature, etc.; or when several monitoring data reach a certain threshold, the alarm is triggered; or when the thresholds of several monitoring data are compared and the weights are calculated, and a certain risk score is reached, the alarm is triggered.

[0109] For example, in practical applications, a series of rules can be set: Rules for stacking large items: The yards where stacking is allowed include Yard A and Yard C; the yard where stacking is prohibited is Yard B.

[0110] Rules for flammable items: The permitted storage yard is storage yard C; the prohibited storage yards include storage yard A and storage yard B. When flammable items appear in storage yard A or storage yard B, an alarm is triggered.

[0111] Operation time rules: Yard A is "8:00-18:00"; Yard B is "8:00-19:00"; Yard C is "6:00-20:00".

[0112] Temperature rule: When the temperature in the monitored yard is higher than 60°C, an alarm is triggered.

[0113] S213: Traverse the multi-source monitoring data and match them with the alarm rules one by one. Once data that matches the alarm rules is found, the rule engine will trigger the corresponding alarm mechanism.

[0114] Through a series of rules, non-compliance phenomena are screened out and risks are identified; once suspicious problems that match the rules are found, the rule engine will trigger the corresponding alarm mechanism. For example, when weather data shows that the temperature exceeds 60°C and there are flammable items in the yard, the rule engine will immediately identify this situation and notify the relevant administrator in a timely manner.

[0115] The method further comprises: Obtain historical multi-source monitoring data; Based on historical multi-source monitoring data, a decision tree model is constructed and trained to build a dangerous goods risk prediction model.

[0116] The construction and training of a decision tree model based on historical multi-source monitoring data includes: The decision tree selects the optimal features and thresholds for segmentation based on the features and target variables in the training data set to form a tree structure. The segmentation process is repeated until the stopping condition is met. The decision tree leaf nodes store the probability of accidents occurring under the data feature path.

[0117] Specifically, the decision tree construction process is as follows: During the training phase, the decision tree will select the optimal features and thresholds for segmentation based on the features in the training data set and the target variable (here, whether an accident has occurred) to form a tree structure. This process will be repeated until the stopping condition is met, that is, a certain set threshold is met, the depth of the decision tree or the number of samples in the node is lower than a certain threshold, etc.

[0118] In the prediction stage, for a new observation, the model will go down the decision tree according to its feature value until it reaches a leaf node, which contains the category probability distribution calculated based on the training data and stores two types of samples (accidents and no accidents) or proportions. For example, the model will go to a leaf node along the decision path based on the characteristics of the day (throughput, wind speed, transportation efficiency, etc.). According to historical data, the proportion of samples with accidents in all samples is 18.5%, so the model will predict that the probability of accidents on that day is 18.5%.

[0119] Preferably, the method further comprises: Regularly use the latest multi-source monitoring data and actual risk conditions to train and optimize the hazardous goods risk prediction model.

[0120] In practical applications, in order to ensure the effectiveness of the hazardous goods risk prediction model, the latest monitoring data and actual risk conditions are regularly used to train and adjust the model, adjust the splitting criteria of the decision tree, and further improve the recognition accuracy of the model to optimize its performance.

[0121] Step S22 includes: S221: Analyze multi-source monitoring data through the trained hazardous cargo risk prediction model to obtain the accident probability p.

[0122] S222: Obtain a safety score s based on the accident probability p and the predetermined accident classification level.

[0123] Specifically, the accident classification levels are: Low risk: the probability of an accident occurring is less than or equal to 5%; Medium risk: the probability of an accident occurring is between 5% and 20%; High risk: the probability of an accident occurring is between 20% and 40%; Extremely high risk: The probability of an accident occurring p is greater than 40%.

[0124] For example, if the predicted probability p of an accident occurring on that day is 18.5%, the accident level is classified as medium risk.

[0125] The calculation formula for the safety score s is: s=m*p+b; where: m is the slope, which is used to describe the speed at which the score changes with probability; b is the intercept, which represents the score when p=0.

[0126] Low risk (accident probability p is less than or equal to 5%): safety score s is between 0 and 1.00; Medium risk (accident probability p between 5% and 20%): safety score s between 1.01 and 2.99; High risk (accident probability p between 20% and 40%): safety score s between 3.00 and 4.00; Extremely high risk (accident probability p greater than 40%): Safety score s is between 4.01 and 5.00.

[0127] It can be set to issue an early warning when the safety score s exceeds 3 points.

[0128] S3: Issue early warning based on risk prediction results.

[0129] When the rule engine finds the risk of hazardous goods based on monitoring data, it will issue an early warning in a timely manner.

[0130] The beneficial effect of this embodiment is that the method can integrate historical data and real-time data, use rule engines and machine learning algorithms to make comprehensive predictions on risk factors, and greatly improve the accuracy and efficiency of analysis.

[0131] Embodiment 4

[0132] A dangerous goods risk warning system, based on the dangerous goods risk warning method as described in Embodiment 1, 2 or 3, comprises: Data acquisition module, used to acquire multi-source monitoring data; The prediction module is used to analyze multi-source monitoring data through a preset rule engine and / or a trained dangerous goods risk prediction model to obtain risk prediction results; The early warning module is used to issue early warnings based on risk prediction results.

[0133] Preferably, the system further comprises: Rules engine; Model building module, used to build a dangerous goods risk prediction model; Model optimization module, used to optimize the dangerous goods risk prediction model; The risk integration module is used to integrate the risk prediction results of the rule engine and the dangerous goods risk prediction model.

[0134] Embodiment 5

[0135] A computer program, which, when executed, implements the steps of the dangerous goods risk warning method as described in Example 1, 2 or 3.

[0136] Embodiment 6

[0137] A readable storage medium having stored thereon a computer program as described in Example 4, wherein when the computer program is executed, the steps of the dangerous goods risk warning method as described in Example 1, 2 or 3 are implemented.

[0138] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the art within the technical scope disclosed in the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A dangerous goods risk early warning method, characterized in that: The following steps are involved: Acquire multi-source monitoring data; Through the preset rule engine and the trained dangerous goods risk prediction model, the multi-source monitoring data is analyzed to obtain the risk prediction results; Issue early warning based on risk prediction results; Wherein, the hazardous cargo risk prediction model is a decision tree model, and the tree leaf nodes of the decision tree store the probability of accident occurrence under the data feature path; The analysis of multi-source monitoring data through a preset rule engine and a trained dangerous goods risk prediction model includes: Through the preset rule engine, the multi-source monitoring data is analyzed to obtain the first risk prediction result; Through the trained dangerous goods risk prediction model, the multi-source monitoring data is analyzed to obtain the probability of accidents and calculate the safety score; The risk prediction result is obtained by combining the first risk prediction result and the safety score.

2. The dangerous goods risk warning method according to claim 1 is characterized in that: The dangerous goods risk early warning method further includes: Preprocess multi-source monitoring data.

3. The dangerous goods risk warning method according to claim 1 is characterized in that: The risk prediction results obtained by analyzing the multi-source monitoring data through the preset rule engine include: Obtain cargo yard data; Set monitoring data alarm rules based on cargo yard data; The multi-source monitoring data is traversed and matched with the alarm rules one by one. Once data that matches the alarm rules is found, the rule engine will trigger the corresponding alarm mechanism.

4. The dangerous goods risk warning method according to claim 1 is characterized in that: The dangerous goods risk early warning method further includes: Obtain historical multi-source monitoring data; Based on historical multi-source monitoring data, a decision tree model is constructed and trained to build a dangerous goods risk prediction model.

5. The dangerous goods risk warning method according to claim 4 is characterized in that: The construction and training of a decision tree model based on historical multi-source monitoring data includes: The decision tree selects the optimal features and thresholds for segmentation based on the features and target variables in the training data set to form a tree structure; the segmentation process is repeated until the stopping condition is met.

6. The dangerous goods risk warning method according to claim 5 is characterized in that: The calculation formula of the safety score is: s=m*p+b Where: s is the safety score; p is the probability of an accident; m is the slope, which is used to describe how quickly the score changes with probability; b is the intercept, which represents the score when p=0.

7. The dangerous goods risk warning method according to claim 4 is characterized in that: The dangerous goods risk early warning method further includes: Regularly use the latest multi-source monitoring data and actual risk conditions to train and optimize the hazardous goods risk prediction model.

8. The dangerous goods risk warning method according to claim 1 is characterized in that: The risk prediction results obtained by analyzing multi-source monitoring data through the preset rule engine and the trained dangerous goods risk prediction model include: According to the preset rule engine, the multi-source monitoring data is analyzed to obtain the first risk prediction result; Analyze the multi-source monitoring data according to the trained dangerous goods risk prediction model to obtain the second risk prediction result; The first risk prediction result and the second risk prediction result are combined to obtain a risk prediction result.

9. A dangerous goods risk warning system, characterized in that: The dangerous goods risk warning method according to any one of claims 1 to 8 comprises: Data acquisition module, used to acquire multi-source monitoring data; The prediction module is used to analyze multi-source monitoring data through a preset rule engine and a trained dangerous goods risk prediction model to obtain risk prediction results; The early warning module is used to issue early warnings based on risk prediction results.

10. A computer program, characterized in that When the computer program is executed, the steps of the dangerous goods risk warning method as described in any one of claims 1 to 8 are implemented.

Images