Disaster prediction system for work sites

Abstract

To predict the occurrence of disasters and issue warnings to mitigate their occurrence. [Solution] A disaster occurrence prediction system A capable of predicting the occurrence of disasters at work sites such as construction sites and factories comprises at least a derivation unit 10 that derives disaster prediction information and keyword information by machine learning processing that includes past disaster information as training data and text mining processing of past disaster information, and a report generation unit 20 that generates a prediction result report C using at least the disaster prediction information and the keyword information.

Description

Technical Field

[0001] The present invention relates to a system for predicting the occurrence of various disasters such as industrial accidents at work sites such as construction sites and factories.

Background Art

[0002] Patent Document 1 below discloses an apparatus that predicts the occurrence status of disasters related to a construction project by using document information related to the construction project (work procedure manuals, construction process manuals, construction plans, reports, etc.). As shown in FIGS. 12 and 13 of Patent Document 1, by designating a target building, a machine-learned disaster occurrence prediction model performs disaster prediction, and as a prediction result screen, comments indicating the possibility of disaster occurrence in the designated building, disaster types, occurrence times, etc. are output.

[0003] In addition, Non-Patent Document 1 below discloses a machine learning model that predicts the occurrence of disasters in the next month using primary data such as the total number of workers and supervisors, total working hours, and the number of disasters in the previous month at each construction site, and secondary data such as the time-series trend and change rate as features.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Non-Patent Documents

[0005]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] The present invention aims to provide a disaster occurrence prediction system that contributes to the suppression of disaster occurrence in a manner different from the conventional methods described above. [Means for solving the problem]

[0007] The present invention, made to solve the above problems, is a disaster occurrence prediction system capable of predicting the occurrence of disasters at work sites such as construction sites and factories, and is characterized by comprising at least: a derivation unit that derives disaster prediction information and keyword information by machine learning processing that includes past disaster information as training data and text mining processing of past disaster information; and a report generation unit that generates a prediction result report using at least the disaster prediction information and the keyword information. Furthermore, the present invention allows the derivation unit to use future prediction values ​​obtained from an arbitrary future prediction model independent of the disaster occurrence prediction system in deriving the disaster prediction information. Furthermore, the present invention may be configured such that the report generation unit has a function to generate a forecast map field that illustrates the degree of risk of disaster occurrence for each work site at predetermined intervals, based on the disaster prediction information. Furthermore, the present invention may be configured such that the report generation unit has a function to generate an overview field consisting of a summary description of the disaster prediction information based on the disaster prediction information. Furthermore, the present invention may be configured such that the report generation unit has a function to generate a caution field consisting of an introductory text for keywords that require attention, based on the keyword information. [Effects of the Invention]

[0008] According to the present invention, at least one of the effects described below is achieved. (1) In addition to deriving disaster prediction information (such as the number of occurrences and probability of occurrence for each work site), it also derives keyword information related to disaster occurrence, which enables more specific warnings and helps to reduce the occurrence of disasters. (2) By using future prediction values ​​obtained from future prediction models such as weather forecast models to derive disaster prediction information, it becomes possible to provide disaster prediction information with higher accuracy. (3) By including a forecast map section in the prediction results report that illustrates the risk of disaster occurrence for each work site over a predetermined period, it is possible to make the warnings to workers easier to understand in a manner similar to a familiar weather forecast map. [Brief explanation of the drawing]

[0009] [Figure 1] Functional block diagram of the disaster occurrence prediction system according to the present invention. [Figure 2] Functional block diagram of the derivation section according to Example 1. [Figure 3] Figure (1) shows an example of the output of the prediction results report. [Figure 4] Figure (2) shows an example of the output of the prediction results report. [Figure 5] Figure (3) shows an example of the output of the prediction results report. [Figure 6] Figure (4) shows an example of the output of the prediction results report. [Modes for carrying out the invention]

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings. [Examples]

[0011] <1> Overall structure (Figure 1) Figure 1 is a functional block diagram of the disaster occurrence prediction system A according to the present invention. The disaster occurrence prediction system A according to the present invention comprises at least a derivation unit 10 that derives disaster prediction information and keyword information based on the results of machine learning and text mining using document information B, and a report generation unit 20 that generates a prediction result report C using at least the disaster prediction information and keyword information derived by the derivation unit 10. Each part can be realized by arbitrarily combining hardware and software. In addition, a configuration in which each part is an individual device, a configuration in which a plurality of parts are incorporated into one device and integrated, etc. can be adopted. Hereinafter, the details of each part will be described.

[0012] <2> Derivation Unit (Figs. 1, 2) The derivation unit 10 has a function for deriving disaster prediction information and keyword information. In this embodiment, as shown in Fig. 2, the derivation unit 10 is divided into two functional blocks (prediction information derivation unit 11, keyword derivation unit 12), but the present invention is not limited to this configuration. Also, the timing for performing the derivation process may be manual execution according to a user's command, or may be automatically executed regularly on the system side.

[0013] <2.1> Prediction Information Derivation Unit (Fig. 2) The prediction information derivation unit 11 has a function for deriving disaster prediction information. This "disaster prediction information" is information indicating the content of the predicted disaster (disaster occurrence date, disaster occurrence location, etc.).

[0014] <2.1.1> An Example of the Derivation Method The prediction information derivation unit 11 can be configured, for example, using a neural network that has been machine-learned with the accumulated document information B as teacher data (learning dataset). For the document information B used as teacher data for deriving disaster prediction information, the matters described in each item in a disaster report or the like that summarizes the content of past disasters can be used.

[0015] <2.1.2> An Example of Teacher Data For example, teacher data for each disaster occurrence fact can be generated from the following item information included in the summary column 30 described in the disaster report. Disaster occurrence date: The date when the disaster occurred Disaster location: The work site (work base) where the disaster occurred.

[0016] <2.1.3> Example of disaster prediction information output The prediction information derivation unit 11 includes a neural network that is machine-learned using this training data. For example, when a future date and predicted location are used as input data to the neural network, it can derive output data indicating whether or not a disaster occurred (or the number of occurrences) at the predicted location, with the predicted date being the predicted date for the disaster. Furthermore, this information, which links the predicted date and location of a disaster, can be used as disaster prediction information. In addition, the prediction information derivation unit 11 can further aggregate information linking the predicted date and location of a disaster to derive the predicted number of disaster occurrences for each location on a monthly basis as disaster prediction information.

[0017] <2.2> Keyword Derivation Section (Figure 2) The keyword derivation unit 12 has the function of deriving keyword information related to the occurrence of a disaster. This "keyword information" refers to information that indicates words highly relevant to the predicted disaster.

[0018] <2.2.1> An example of a derivation method The keyword derivation unit 12 can output analysis results for each word, such as part-of-speech classification, frequency of occurrence, and co-occurrence relationships (associations of words with similar occurrence patterns in a text), by performing text mining on sentences and natural language texts within the stored document information B.

[0019] <2.2.2> An example of text mining The document information B used in the keyword derivation unit 12 can include text entered in detail fields 40, such as the detailed entry field for disaster details or the free entry field, from disaster reports that summarize the details of past disasters.

[0020] <2.2.3> Example of Keyword Information Output As a result of performing text mining on the text entered in detail field 40, the keyword derivation unit 12 can derive keyword information such as words that frequently appear when a disaster occurs, and combinations of words that are found to have a co-occurrence relationship. The target of text mining may be all accumulated document information B, or it may be document information B created within a specific period. For example, if the forecast period is set to the future (January to March), only disaster reports from past January to March that involved disasters may be targeted for text mining.

[0021] <2.2.4> Example of Keyword Information Grouping In the derivation unit 10, each word derived from the keyword information may be further grouped with an arbitrary name. For example, if the words derived from keyword information include words such as "step," "shelf," "electrical outlet," "air," and "entrance / exit," these words can be grouped together under the name "building-related" because they share a common concept. Furthermore, in the case of words derived from keyword information such as "extrusion," "mold," "casting," and "metal sheet," these words can be grouped together under the name "equipment-related" as they share a common concept.

[0022] <3> Report generation unit (Figure 1) The report generation unit 20 has the function of generating a prediction result report C using at least the disaster prediction information and keyword information derived by the derivation unit 10.

[0023] <3.1> Example of displaying the forecast results report It is preferable that the prediction result report C is generated in a manner that makes it easy for safety management personnel and workers at each work site to understand the predicted frequency and nature of future disasters.

[0024] In this embodiment, the disaster occurrence prediction system A according to the present invention is configured so that users can log in from a web browser installed on their terminal and view the prediction result report C on the web browser. Forecast result report C can include sections such as a forecast chart section 50, a forecast summary section 60, a forecast warning section 70, and a graph section 80. The display images for each column will be explained with reference to Figures 3 to 6.

[0025] <3.2> Forecast Chart Section (Figure 3) The forecast map section 50 is a section that illustrates the degree of risk of disaster occurrence for each work site for each predetermined period, based on the disaster prediction information derived by the derivation unit 10. As shown in Figure 3, the forecast map section 50 includes leader lines from the locations of the work sites (locations A to E) on the map of Japan, and provides risk information 51 with icons representing weather, indicating the probability of disasters occurring each month for the next three months. In this embodiment, an icon with a lightning bolt symbol indicates an extremely high probability of disaster occurrence (80-99%), an icon with a rain symbol indicates a high probability of disaster occurrence (60-79%), and an icon with a cloudy sky symbol indicates a moderately high probability of disaster occurrence (40-59%). There are various methods for determining the probability of a disaster occurring from the disaster prediction information derived by the derivation unit 10, and therefore the present invention does not particularly limit these methods. For example, multiple thresholds can be set for the total number of disaster occurrences in a month for each work site, and the system can be configured to determine which icon to display based on whether or not one of these thresholds has been exceeded.

[0026] <3.3> Forecast Summary Section (Figure 4(a)) The forecast summary section 60 is a section that displays a summary of disaster prediction information for each predetermined period, based on the disaster prediction information derived by the derivation section 10. The report generation unit 20 stores template data for summary texts. Based on the disaster prediction information derived by the derivation unit 10, it selects template data and generates a summary text by appropriately filling in the blanks (such as the name of the base, the probability of disaster occurrence, and the predicted month of disaster occurrence) within the selected template data. It is preferable to have various templates for the summary document, depending on factors such as the monthly number of disaster occurrences at each location and the total number of disaster detections across multiple locations.

[0027] <3.4> Forecast and Warning Section (Figure 4(b)) The forecast warning section 70 is a section that displays introductory text for keywords that require attention, based on the keyword information derived by the derivation section 10. The report generation unit 20 stores template data for introductory texts and can generate introductory texts by appropriately filling in blanks in the template data using keyword information derived by the derivation unit 10 (such as frequently occurring words and combinations of words that are considered to have a co-occurrence relationship). It is preferable to prepare various variations of template data for introductory texts, depending on factors such as frequency of occurrence and the conceptual names used for grouping as described in <2.2.4> above.

[0028] <3.5>Graph section (Figures 5 and 6) The graph section 80 is a section that displays the number of disaster occurrences for each predetermined period, narrowed down by arbitrary conditions using at least the disaster prediction information derived by the derivation section 10, in graph form.

[0029] <3.5.1> Display for each work site Graph 80 in Figure 5 is a graph for each work site (each location), with the horizontal axis representing the time of disaster occurrence and the vertical axis representing the number of disasters. The graph on the left side of Figure 5 shows the number of past disaster occurrences and the number of future disaster occurrences from January 2018 to September 2023, while the graph on the right side of the same figure shows the number of disaster occurrences on a monthly basis during 2023.

[0030] <3.5.2> Display by attribute The graph section 80 shown in Figure 6 displays the timing of the disaster on the horizontal axis and the number of disaster occurrences on the vertical axis, categorized by the attributes of the disaster. The "attributes of the disaster content" can include the names used for grouping as described in <2.2.4> above, or other classification criteria names as appropriate. The graph on the left side of Figure 6 shows the total number of disaster occurrences from January 2018 to September 2023, along with a projection of future disaster occurrences. The graph on the right side of the same figure shows the monthly number of disaster occurrences in 2023.

[0031] <4> summary Thus, the disaster occurrence prediction system A according to this embodiment not only derives disaster prediction information (such as the number of disaster occurrences aggregated for each work site and time period, and the probability of disaster occurrence), but also derives keyword information related to disaster occurrence, making it possible to issue more specific warnings. Furthermore, since the prediction results are displayed in a report format similar to a familiar weather forecast map, it is beneficial for safety management personnel and workers in that the content of the warnings is easy to understand.

[0032] <5> supplementary explanation Furthermore, if appropriate warnings are issued using the disaster occurrence prediction system A according to the present invention, it may lead to a decrease in the number of subsequent disasters. In such cases, a discrepancy may occur between the results predicted by the disaster occurrence prediction system A according to the present invention and the actual number of disasters, resulting in a lower prediction accuracy. However, since the purpose of the disaster occurrence prediction system A according to the present invention is to suppress the occurrence of disasters, this purpose is fully achieved. [Examples]

[0033] The disaster occurrence prediction system A according to the present invention can also be further combined with any future prediction model. This future prediction model can utilize various future prediction models that are independent of (and exist in their own right) the disaster occurrence prediction system A, and which are expected to correlate with the occurrence of disasters (for example, weather forecast models (models that predict future weather), product production volume models (models that predict future production volume), construction volume models (models that predict future construction volume), and long-term worker skills models (models that predict the degree of improvement in future workers' skills)). This embodiment makes it possible to provide disaster prediction information with even higher accuracy. The following describes examples of combinations when using a weather forecast model as a future prediction model.

[0034] <1> Example combination (1) Integration into the predictive information derivation unit (not shown) In this example, the training data (training dataset) for machine learning performed in the prediction information derivation unit 11 is configured to include actual weather information values ​​for each location and date obtained from other databases. When deriving disaster prediction information, future weather forecast values ​​obtained by a weather forecast model that operates independently of the disaster occurrence prediction system A are also used as input data to derive the disaster prediction information.

[0035] <2> Example combination (2) Correction of disaster prediction information (not shown) In this example, in addition to the prediction information derivation unit 11 and keyword derivation unit 12 that constituted the derivation unit 10 in Example 1, a correction coefficient output unit is provided as a new functional block. The correction coefficient output unit can be configured as a neural network that uses past weather information for each location and date, along with correction coefficients obtained by quantifying the presence or absence and number of disasters according to predetermined rules, as training data (training dataset). By inputting future weather forecast values ​​obtained by a weather forecast model independent of the disaster occurrence prediction system A according to the present invention as input data to this correction coefficient output unit, correction coefficients for each location and each date are derived. The disaster prediction information (such as the number of disaster occurrences per month) obtained by the prediction information derivation unit 11 is then multiplied by the appropriate correction coefficient to derive the final disaster prediction information (such as the number of disaster occurrences per month after multiplication). [Explanation of symbols]

[0036] A: Disaster Occurrence Prediction System 10: Derivation part 11: Predictive Information Derivation Unit 12: Keyword Derivation Section 13: Correction coefficient output section 20: Report Generation Department B: Document Information 30:Summary column 40:Details column C: Prediction Results Report 50: Forecast Chart Section 51: Risk Level Information 60: Forecast summary section 70: Forecast and Warning Section 80: Graph section

Claims

1. A disaster occurrence prediction system capable of predicting the occurrence of disasters at work sites such as construction sites and factories, The system includes a derivation unit that derives disaster prediction information and keyword information through machine learning processing that includes past disaster information as training data, and text mining processing of past disaster information. A report generation unit generates a prediction result report using at least the aforementioned disaster prediction information and keyword information. It must have at least the following: The derivation unit includes at least a prediction information derivation unit and a keyword derivation unit, The aforementioned prediction information derivation unit is: The aforementioned machine learning process derives the predicted number of disaster occurrences for each work site over a predetermined period in the future, as disaster prediction information. The keyword derivation unit is, The text mining process is used to derive keyword information that is highly relevant to predicted disasters from combinations of words that appear frequently or have a co-occurrence relationship, which are then used by the prediction information derivation unit. Disaster occurrence prediction system.

2. The derivation of the disaster prediction information by the derivation unit is characterized by using future prediction values ​​obtained from an arbitrary future prediction model that is independent of the disaster occurrence prediction system. The disaster occurrence prediction system according to claim 1.

3. The report generation unit is characterized by having a function to generate an overview section consisting of a summary description of the disaster prediction information based on the disaster prediction information. The disaster occurrence prediction system according to claim 1.

4. The report generation unit is characterized by having a function to generate a caution section consisting of introductory text for keywords that require attention, based on the keyword information. The disaster occurrence prediction system according to claim 1.

Images