TWI737073B - Timing analysis system and method for petition cases - Google Patents

Abstract

This invention provides a timing analysis system and a method for petition cases, which is used to summarize repeated and continuous cases, analyze whether there is regularity in the notification of the cases, and use the time series algorithm model for the regular notifications to predict the number of cases in the future time interval; as for irregular notifications, a neural network algorithm model is used to analyze the probability of case characteristics affecting notifications, the analysis results of regular cases and characteristic cases are presented respectively, and the date and time of the predicted receipt notification are sorted in order to simplify the time series analysis method of petition cases, improve the predictive ability of the number of reported cases, strengthen the summary of the characteristic attributes that affect the petitions, and improve the decision-making assistance benefit of the authority.

Description

Time sequence analysis system and method of petition cases

The present invention relates to a time series analysis system and method for petition cases. In detail, it is a system and method that collects historical petition cases and open data, and intelligently analyzes the time sequence rules of petition cases through statistical analysis and machine learning.

The main purpose of the citizen service lines of the county and city governments is to solve the problems and inconveniences of the citizens to the urban environment and facilities. Services such as a single phone number, website or APP are launched to provide citizens with a variety of non-emergency issues (emergency is 110 , 119 service scope), such as noise, garbage, street lights, parks and road trees and other public hazards or facilities that are threatened, road maintenance, transportation complaints, notifications and consultation services. The intention is to immediately accept reports from the public, list a notification item as a case, and assign it to the competent authority as quickly as possible. There is no need to go through the cumbersome city appearance inspection and notification and other internal signing and submission operations, and the competent authority After accepting the content of the report, the situation will be responded to as soon as possible, and convenient citizen services will be established.

In view of the fact that the government's civic service hotline provides free and convenient consultation, rapid dispatch and petition acceptance process, it often attracts some citizens to repeatedly report, report a large number of reports, and even maliciously falsely report for various reasons, which causes the petitioner or the accused and the undertaking agency to consume manpower. , Money and other resources while Facing great distress. However, taking a laundry shop in a community as an example, the smell of detergent is annoying, and the pungent smell of detergent permeates the community every day. Residents are deeply distressed. Neighbors in the victim community have called the customer service hotline more than 500 complaints within 3 years, and the responsible unit is present. More than 200 times, including part of the time when the inspectors arrived, the laundry has not yet started business, or the limitations of the testing equipment, and the evidence of no pollution was checked, and then the motive of the petitioner was suspected, and the petitioner was even listed as a refusal to communicate In the end, because the authority and responsibility unit replaced the odor pollutant detection tool, it was able to detect that the air pollution exceeded the standard, only to know that the complaint was true and violated the target of the citizen service line. In addition, complaints that are difficult to deal with such as illegal construction, illegal parking, vagrant, road tyrants, vendors, etc., there are many offenders, high re-offending rates, time-consuming processing, and partly involve the personal interests of the petitioner. If you receive repeated notifications from the petitioner, The accepting unit also repeated the visits again and again, fearing it would become a specific harassing tool.

For repeated petition cases that occur in neighboring locations and in the same case, if the notification case can be automatically judged in advance at the first time when the petition case is received, the neighboring location of the notification, the past petition cases and the historical notification time, and the timing rules for the notification of petition cases , Find effective processing methods, select appropriate time intervals and detection techniques, conduct on-site inspections, and accurately find problems; and reply to repeated notifications without new facts or evidence by signing to improve service efficiency and effectively prevent citizens from abuse , In addition to saving manpower for checking, it can actually improve citizen satisfaction.

It can be seen that the above-mentioned conventional methods still have poor system usability, which is not a convenient and easy-to-use design, and needs to be improved urgently.

In view of the various shortcomings derived from the above-mentioned conventional methods, the inventor of this case is eager to improve and innovate, and after years of painstaking research, finally successfully completed the development and completion of a time sequence analysis system and method for this petition case.

The purpose of the present invention is to provide a time series analysis system and method for petition cases, using statistical analysis and machine learning technology to improve the forecasting ability of the notification time interval of petition cases in smart cities, and assist the authority and responsibility unit to analyze the law of petition cases in advance and plan ahead. The dispatching period improves the efficiency of public information extraction and service, and enhances citizen satisfaction.

The second purpose of the present invention is to strengthen the smart city's response to continuous petition cases, classify cases into repeated petition cases or single petition cases in advance, analyze the trend of case notification, summarize continuous petition cases or occasional petition cases, and provide rights and responsibilities The unit adjusts the priority of inspections, and excludes used and ineffective detection technologies in advance to make the operation of citizen services smoother.

Another purpose of the present invention is to analyze in advance the number of notifications and processing results of neighboring locations, the same type of case, and use the data exploration of petition cases to sign the reported cases that have been processed or confirmed, simplifying the regularity or The procedures for handling reported cases that obviously consume agency resources can effectively prevent the abuse of resources.

Another purpose of the present invention is to improve the prediction ability of petition cases in smart cities. By predicting the notification time series of petition cases, it assists government authorities and first-line inspectors to grasp the time of case notification predicted by the algorithm in real time. In order to reduce the public’s demand for dispatching services from a preventive perspective, it will increase the influence of artificial intelligence prediction technology on municipalities, citizens, and society as a whole.

In order to achieve the above-mentioned object of the invention, the present invention provides a time series analysis system for petition cases, including: an external data collection module, which stores the data obtained by interfacing a petition integration system and an open data system in a preprocessing process One data storehouse, and then one report The notification time and location of the sentiment case, search for historical petition cases within the preset time range and location in the data repository, and determine whether the petition case is a repeated and continuous petition case in a nearby location; a petition time series exploration module , Is to receive the repeated and continuous petition case, analyze whether the number of cases in different time units has regularity; if there is regularity, the regular case prediction module will use the time series algorithm to build Model prediction to provide time series analysis results of regular cases; and a feature case prediction module. If there is no regularity, the feature case prediction module uses neural network-like algorithms to model and predict, thereby providing feature cases The timing analysis results.

In order to achieve the above-mentioned purpose of the invention, the present invention also provides a time series analysis method of petition cases, including the collection of external data, the exploration of petition time series, the prediction of regular cases, and the prediction of characteristic cases.

The collection of external data is to receive the latest submissions in real time and obtain external open data at regular intervals. After data extraction, conversion, cleaning and de-identification and other pre-processing, the time and location are reported according to the submissions, and the historical submissions in the preset time range and location range are searched. Case, to determine whether the petition is a single petition case or a repeated petition case in a nearby location. If it is a repeated petition case, calculate whether the notification frequency of the petition case is greater than the preset value. If the notification frequency is greater than the preset value, it will be classified as a continuous petition case; otherwise, it will be classified as an occasional petition case. This method understands the repetitiveness and continuity of past petitions, and then makes predictions about future petition trends.

The exploration of the sequence of reports is to use narrative statistics to describe the outline of the number of cases for repeated and continuous petition cases, and to understand the outline of the change of the number of cases and the degree of concentration and dispersion. Draw a time series diagram of the number of cases in different basic time units such as hour, day, week, month, quarter or year, observe whether the number of cases in each time unit changes regularly, and analyze the time of notification Between rules. If the notification time of the case is regular, the time series algorithm is used to model and forecast; otherwise, the neural network-like algorithm is used to model and forecast. This method classifies cases as whether they have regular time series characteristics, and provides important information for the selection of subsequent problem-solving methods.

The prediction of regular cases is based on the statistical analysis method. The regular cases are modeled based on the time point of historical submissions and the number of cases, using a time series algorithm to model, with date as the main key, and the number of historical cases as input and forecast variables, using time series Analyze and predict the number of cases on several dates in the future. By digging out the changes in the number of cases over time, we can predict the number of reported cases and the trend of changes in the number of cases in the future time interval.

The prediction of characteristic cases is based on the machine learning method. The non-regular cases are based on the characteristics of the case itself and the characteristic attributes of the collected neighboring area disaster, geography, weather, environmental quality, and historical complaints, using a neural network-like algorithm Modeling and assessing the probability of each feature's influence on the occurrence of notified cases. Feature is an option of similar neural network analysis. Important features help predict the impact of the feature on case notification and improve the accuracy of the model. Priority is given to the features with greater impact, so that the features of the notification of impact cases can be grasped more and more accurately. The time intervals in which the features appear are listed in order, and the results of time series forecasts can be further used to assist decision-making and pre-arrange inspections. Working time.

The chronological analysis method of petition cases in the smart city gathers the petitions of citizens collected by the government petition platform, and effectively analyzes and predicts the time interval of future notifications in various trivial, partial and irrelevant messages. From the perspective of solving citizen problems, Provide decision-making information with reference value for the decision-making support center to respond to the service needs of the public. This is the key task of seeking public despair, exploring public opinion and managing citizen relations.

Apply the time sequence analysis method of petition cases to predict the time interval of case notification in advance, assist the government authority to adjust the dispatch strategy, effectively improve the timeliness and accuracy of personnel dispatch and dispatch, and reduce the citizen's dispatch service from the perspective of prevention need.

1‧‧‧Time sequence analysis system for petition cases

11‧‧‧External data collection module

12‧‧‧Chen Qing Time Series Exploration Module

13‧‧‧Regular case prediction module

14‧‧‧Characteristic case prediction module

2‧‧‧Interface platform

21‧‧‧Composition System

22‧‧‧Open Data System

23‧‧‧Big Data Analysis System

3‧‧‧Data Repository

31‧‧‧Case Database

32‧‧‧Disaster Information Database

33‧‧‧Geographic Information Database

34‧‧‧Meteorological Information Database

35‧‧‧Environmental Quality Database

S101~S104, S201~S208, S301~S306, S401~S408, S501~S508‧‧‧Step

Figure 1A is a schematic diagram of the structure of the time sequence analysis system for petition cases of the present invention;

Figure 1B is a schematic flow diagram of the time sequence analysis method of petition cases of the present invention;

Figure 2 is a flow chart of collecting external data of the time series analysis method of petition cases of the present invention;

Figure 3 is the exploratory flow chart of the petition sequence of the petition case sequence analysis method of the present invention;

Figure 4 is the prediction flow chart of regular cases in the time series analysis method of petition cases of the present invention;

Figure 5 is the prediction flow chart of the characteristic case of the time series analysis method of the petition case of the present invention;

Figures 6A to 6D are time series diagrams of the number of cases accumulated by hour, day, week, and month in the first embodiment of the time series analysis system and method of petition cases of the present invention;

Figure 7 is a time series diagram of the number of cases forecast in the next 30 days in the first embodiment of the time series analysis system and method for petition cases of the present invention;

Figures 8A to 8D are time series diagrams of the number of cases accumulated by hour, day, week, and month in the second embodiment of the time series analysis system and method for petition cases of the present invention; and

Figure 9 is a bar graph showing the influence of the characteristics of the second embodiment of the time sequence analysis system and method of the petition case of the present invention on the gain value of the reported case.

Please refer to Figure 1A, which is a schematic diagram of the system architecture of a time series analysis system for petition cases of the present invention. A petition case time series analysis system 1 has the capabilities of collecting external data, petting time series exploration, predicting regular cases, and predicting characteristic cases. The case sequence analysis system 1 includes an external data collection module 11, a case sequence exploration module 12, a regular case prediction module 13, and a characteristic case prediction module 14.

The external data collection module 11, the external interface platform 2, receives the latest data in the submission integration system 21 and the open data system in real time, and undergoes preprocessing such as extraction, conversion, cleaning and de-identification, according to the reporting time and location of the submission case. Search for historical petition cases in the preset time range (e.g. within one year) and location (e.g. within 100 meters) in the data repository 3 to determine whether the petition case is a single petition case or a repeated petition in a nearby location case. If it is a repeated petition case, calculate whether the notification frequency of the petition case is greater than the preset value (for example: 1 time/month), if the notification frequency is greater than the preset value, it will be classified as a continuous petition case; otherwise, it will be classified as an occasional petition case.

The report time series exploration module 12 will use narrative statistics to describe the number of repeated and continuous reports, including calculating the total number of cases and the average, mode, median, maximum, minimum, and quarter points of the daily cases. Position difference, variance and standard deviation, etc., understand the concentration and dispersion trend of the number of daily cases, and plot the time series of the number of cases in different basic time units such as hour, day, week, month, quarter, or year for historical and instant notification cases Figure, observe the time sequence diagram of each time unit, calculate the length of time from one peak to the next peak or from the bottom to the bottom of the valley. Use different basic time units such as hour, day, week, month, quarter or year, or extended time units such as weekends/weeks, morning/afternoon, etc. to calculate whether the number of cases in this time unit changes regularly. Case regularity refers to the fixed time The unit or within a period of time will receive the rule or cycle of notification of cases, if If the number of cases that occurred in the basic or extended time unit exceeds 50% of all cases, the probability that future cases may occur in this time interval is significantly higher than that in other intervals, and this time interval has a recurring rule. If the notification time of the case is regular, the time series algorithm is used to model and forecast; otherwise, the neural network-like algorithm is used to model and forecast.

The regular case prediction module 13 uses statistical analysis methods to target regular cases, based on historical submission time points and the number of cases, using time series algorithm modeling, mining the changes in the number of cases over time, and predicting future cases The number, in accordance with the sorting of the time period of the notification, and then predict the date and time that the notification may be received in the future.

The characteristic case prediction module 14 uses machine learning methods to deal with irregular cases, based on the case type, location, time, description, level, status and other characteristics of real-time cases and historical cases, as well as the disaster situation, geography and similar time of the collected neighboring areas The characteristics of externally open data such as weather, environmental quality, etc., are modeled using a neural network algorithm to evaluate the probability of the impact of each of the aforementioned features on the occurrence of notified cases. Through repeated modeling and evaluation, the characteristics of affecting case notifications can be grasped more and more. To be more accurate, and to prioritize the features that have greater impact on the effectiveness, and then predict the date and time that the notification may be received in the future, and provide data basis based on statistical analysis and machine learning for the smart city authority and responsibility unit, and can be further assisted by prediction Decision-making and pre-arrangement to check the dispatch time.

The interface platform 2 includes a sentiment integration system 21, an open data system 22 and a big data analysis system 23. The sentiment integration system 21 provides citizens who can directly dial the civic service line, and the operator can directly accept the citizen’s telephone statement or enter the sentencing person, event, time, and Information on places, objects, and supporting evidence such as uploaded photos, videos, etc. The scope covers suggestions for urban governance, reports of violations or lack of regulations or The maintenance of public rights, etc., after being notified by the citizens, the reporting system will provide services such as train management and follow-up dispatch of workers. There may be a plurality of open data systems 22 to provide an interface with external systems, and the ability to obtain open data such as city-related disasters, geography, weather, and environmental quality. The big data analysis system 23 provides an interface to an external system. The interface system selects the algorithm to be modeled, the ratio of the training set to the test set, and imports the file data to be predicted. After repeated modeling and evaluation, the best output forecast result.

The data repository 3 connects the external data collection module 11 to the report data obtained by the report integration system 21, stores it in the case database 31, and connects to the data obtained by the open data system 22, for example, it interfaces with the data obtained by the open data system 22. Obtain disaster information such as road trees, signboards, roads, tunnels, bridges, flooding, water conservancy facility disasters, and people’s livelihood infrastructure, and load them into the disaster information database 32; for example, connect the layers and networks obtained by the open geographic information system Geographic information such as grids, businesses, disaster prevention and relief, etc., are loaded into the geographic information database 33; the meteorological observation data such as temperature, rainfall, wind direction, wind speed, altitude, relative humidity and other meteorological observation stations obtained by the meteorological data open system will be loaded. Enter the weather information database 34; will interface with the air quality index value, index status, index display color and environmental quality monitoring of ozone, carbon monoxide, sulfur dioxide, nitrogen monoxide, nitrogen dioxide, nitrogen oxides, etc. obtained by the environmental quality open system The data is loaded into the environmental quality database 35 as a data source related to time series analysis.

Please refer to Figure 1B, which is a schematic flow diagram of the time sequence analysis method of a petition case of the present invention. In step S101, the collection of external data is performed, and then the process proceeds to step S102. In step S102, for repeated and continuous petition cases, an exploration of petition sequence is performed to distinguish regular and irregular petition cases. In step S103, the prediction of regular cases is executed for regular complaints. In step S104, the prediction of characteristic cases is executed for irregular petition cases.

Please refer to Figure 2, which is a schematic diagram of the collection of external data in a time series analysis method of petition cases of the present invention.

In step S201, the latest submission is received in real time and the external open data is obtained regularly to be written into the data storage database. In detail, the external data collection module is connected to the report integration system to receive the latest report data in real time and write it into the case database; it is connected to the open data system regularly to obtain disaster, geography, meteorology, environmental quality and other information, and write them into the disaster situation separately Information database, geographic information database, environmental quality information or environmental quality database.

In step S202, according to the time and location of the submission, the historical submission data within the preset time range and the preset location are queried. In detail, the external data collection module queries the historical submission data in the case database within the preset time range and the preset location based on the time and location of the submission.

In step S203, it is checked whether there are historical petitions of the same case type. If yes, proceed to step S204 to classify the petition case as a repeated petition case; if not, proceed to step S205 to classify the petition case as a single petition case.

。若是，進至步驟 S207，將該重複陳情案件歸類為持續陳情案件；若否，進至步驟S208，將該重複陳情案件歸類為偶發陳情案件。 In step S206, it is calculated whether the notification frequency of the petition case is greater than a preset value, and the calculation formula of the notification frequency is

. If yes, go to step S207 to classify the repeated petition case as a continuous petition case; if not, go to step S208 to classify the repeated petition case as an occasional petition case.

Please refer to Fig. 3, which is a flow chart of the exploratory flow chart of the time sequence of the petition in the time-series analysis method of the petition case of the present invention.

In step S301, for repeated and continuous petition cases, the statistical value of the narrative statistical method is calculated. In detail, the sentiment timing exploration module obtains the external data collection module In, for cases that have been classified as repeated and continuous submissions, use narrative statistical methods to calculate their statistical values, including the average, mode, and median number of cases in different time units (hours/days/weeks/months/quarters/years) , Maximum value, minimum value, interquartile range, variance and standard deviation, etc., to grasp the changes in the number of cases and the trend of concentration and dispersion.

In step S302, according to different basic time units such as hour, day, week, month, quarter or year, draw a time series diagram of the number of cases, that is, arrange the number of cases in the order of the time unit, so as to understand the change in the number of petition cases .

In step S303, the basic time unit or the extended time unit is used to calculate the proportion of the number of cases in each time unit to the total number of cases.

In step S304, it is searched whether there is a time unit in which the proportion of the number of cases is greater than 50%, and the regularity of the change in the number of cases is analyzed accordingly. If there is regularity, proceed to step S305 to use time series algorithm for modeling; if not, that is irregular, proceed to step S306 to use neural network-like algorithm for modeling.

Please refer to Figure 4, which is a flow chart of predicting regular cases in a time series analysis method for petition cases of the present invention.

In step S401, the historical case is divided into N equal parts according to the same time interval. In detail, the regular case prediction module divides the real-time and historical cases into N equal parts at the same time interval based on the regular cases analyzed by the petition time sequence exploration module.

In step S402, the last Nth is divided into the test set, and the first N-1 are divided into the training set.

In step S403, the ARIMA (Autoregressive Integrated Moving Average Model; integrated moving average autoregressive model) algorithm is used to establish a time sequence Column model, where ARIMA (p, d, q) is also called differential integrated moving average autoregressive model, where AR is autoregressive, p is autoregressive term; MA is moving average, q is the number of moving average terms; d is The number of differentiations made when the time series becomes stationary.

In step S404, use RMSE (Root Mean Square Error), MAE (Mean Absolute Error), or MAPE (Mean Absolute Percentage Error) and other prediction model performance evaluation methods to calculate the difference between the predicted value and the actual value, and evaluate the predicted result and the actual result Whether the error value is within the tolerance interval. If it is not within the tolerance interval, proceed to step S405 to adjust the autoregressive term, the number of moving average terms and the number of difference parameters; if it is within the tolerance interval, proceed to step S406, and use the best parameters obtained by ARIMA algorithm modeling , To predict the number of cases in the future time interval.

In step S407, according to the maximum number of cases in the selected time unit, sort the date and time that the notification may be received in the future time interval.

In step S408, the time sequence analysis results of regular cases are presented.

The above is based on regular reports, and the ARIMA model is used to predict the number of cases in the future N time intervals. Taking the number of reported cases in the future as the target of prediction, the data series formed by the number of cases over time is regarded as a random time series formed over time, through the trend and regularity of the number of cases in the time series Analysis of other factors, mining the correlation or dependence relationship between the number of cases in each time interval, using time series algorithm modeling and forecasting, to achieve the purpose of using historical and real-time cases to predict the number of future cases.

Example one:

For example, suppose there are continuous complaints of noise type in a certain place, and the date range of the complaints is from 2018-09-01 to 2019-04-30. In 8 months and 242 days, a total of 136 notifications were received. Report, if there is no notification on that day, add 0 to the number of cases, and its {date of notification, number of cases} are briefly listed as {2018-09-01, 2}, {2018-09-02, 0}, {2018-09 -03, 0}, {2018-09-04, 0}, {2018-09-05, 0}, {2018-09-06, 0}, {2018-09-07, 0}, {2018-09 -08, 0}, {2018-09-09, 1}, {2018-09-10, 0},..., {2019-04-21, 1}, {2019-04-22, 0}, {2019-04-23, 1}, {2019-04-24, 1}, {2019-04-25, 0}, {2019-04-26, 1}, {2019-04-27, 2}, {2019-04-28, 5}, {2019-04-29, 0}, {2019-04-30, 1}.

Using hour, day, week and month as time units, draw a time series chart of the number of cases accumulated by hour, day, week and month as shown in Figures 6A to 6D. Observe the number of accumulated cases at 14 o'clock and 20 o'clock; the 18th The cumulative number of cases is the largest; the cumulative number of cases on weekends, Sundays and Saturdays is significantly more than the number of cases during the week and March.

Using the basic time unit and the extended time unit respectively, calculate the probability of each time unit appearing. Using the week as the time unit, calculate the percentage of each day of the week to get {Monday, 6.62%}, {Tuesday, 5.15%}, {Wednesday , 10.29%}, {Thursday, 6.62%}, {Friday, 4.41%}, {Saturday, 23.53%}, {Sunday, 43.38%}, and then use the week and weekend as the time unit to calculate the midweek and weekend occurrences As a percentage, we get {weekly, 33.09%} and {weekend, 66.91%}. Since the proportion of the number of cases on weekends is more than 50%, the number of cases changes regularly, so we choose to use time series algorithm to model and forecast.

Using the ARIMA model that has adjusted the most suitable parameters to predict the number of cases in the next 30 days, please refer to the case number forecast time series chart in Figure 7, and the result of {date, number of predicted cases} is {2019-05-01,- 1.86}, {2019-05-02, 0.48}, {2019-05-03, 0.25}, {2019-05-04, 0.65}, {2019-05-05, 2.02}, {2019-05-06, 0.17}, {2019-05-07, 0.91}, {2019-05-08, -0.06}, {2019-05-09, 0.95}, {2019-05-10, 0.31}, {2019-05-11 , 0.87}, {2019-05-12, 1.99}, {2019-05-13, -0.08}, {2019-05-14, 0.98}, {2019-05-15, -0.37}, {2019-05 -16, 1.19}, {2019-05-17, 0.15}, {2019-05-18, 0.86}, {2019-05-19, 2.07}, {2019-05-20, -0.15}, {2019- 05-21, 1.06}, {2019-05-22, -0.63}, {2019-05-23, 1.29}, {2019-05-24, -0.05}, {2019-05-25, 0.80}, { 2019-05-26, 2.10}, {2019-05-27, -0.29}, {2019-05-28, 1.15}, {2019-05-29, -0.85}, {2019-05-30, 1.45} .

Select the top 3 projects with the highest number of predicted cases in order, namely {2019-05-26, 2.10}, {2019-05-19, 2.07}, {2019-05-05, 2.02}, and then the time unit is hour The maximum number of cases is at 14:00, and the first 3 dates and times that are suitable for dispatch are predicted to be 14:00 on 2019-05-26, 14:00 on 2019-05-19 or 14 on 2019-05-05. Time.

Please refer to Figure 5, which is a flow chart of predicting characteristic cases of a time series analysis method for petition cases of the present invention.

In step S501, according to the range of time and location, historical cases and characteristics of open data such as disaster conditions, geography, weather, and environmental quality are inquired as predictive variables. In detail, the characteristic case prediction module is based on the non-regular cases analyzed by the report time sequence exploration module, and according to the time and location range from the beginning to the end, the characteristics of historical cases and open data such as disaster conditions, geography, meteorology, and environmental quality are used as forecast variables. .

In step S502, take the last T% as the test set, and the remaining (100-T)% as the training set.

In step S503, an ANN (Artificial Neural Networks) algorithm is used to establish a model for predicting the influence of the variable on the presence or absence of notification.

In step S504, if the time interval is predicted to have a notification and there is actually a notification, or if no notification is predicted and no notification is actually made, the prediction is correct; if the time interval is predicted to have a notification but no notification is actually expected or no notification is predicted but there is a notification actually, then The prediction is wrong. Use the classification matrix to calculate whether the correct rate is greater than the preset value. If yes, go to step S505, adjust the prediction variables, keep the feature variables with better prediction effect, and remove the feature variables with poor prediction effect; if not, go to step S506, use ANN algorithm to model the better multiple ones Predict the variable, predict the probability of the variable’s occurrence, and calculate the gain value of each characteristic variable. The calculation formula for the gain value is

In step S507, according to the order of the gain value of each feature, the date and time matching the feature in the future time interval are sorted.

In step S508, the time sequence analysis result of the characteristic case is presented.

The above uses a neural network model to predict the weights that affect the presence or absence of notifications for irregular submissions. The similar neural network improves the learning effect by changing the weight. The algorithm itself does not have the function of filtering variables, and all variables will calculate their own weights. If the selected features are ineffective or unstable on the prediction results, it is easy to cause the model's learning effect or convergence to be too slow; features with high correlation will also cause the model to fail to converge. Therefore, when using neural network-like modeling, variables must be carefully selected.

Use neural network-like modeling and forecasting, input the predicted variable and its value range, you can get the proportion of the variable to whether there is a reported case, and calculate the gain value of whether there is a reported case. If the gain value is greater than 1, it indicates the probability of occurrence of the predicted variable Higher than the original occurrence probability of the variable, the variable and its value range are regarded as a tendency to occur, and the result is presented as a relative gain value.

Embodiment two:

For example, suppose there are continuous reports of environmental pollution in a certain area. The date range of the reports is from 2018-09-01 to 2019-06-30. In 10 months and 303 days, a total of 2437 notifications have been received. If there are no notifications on that day , The number of cases is added to 0, and its {date of notification, number of cases} is abbreviated as {2018-09-01, 4}, {2018-09-02, 7}, {2018-09-03, 5}, {2018-09-04, 3}, {2018-09-05, 7}, {2018-09-06, 6}, {2018-09-07, 7}, {2018-09-08, 10}, {2018-09-09, 12}, {2018-09-10, 4},..., {2019-06-21, 3}, {2019-06-22, 6}, {2019-06-23 , 12}, {2019-06-24, 9}, {2019-06-25, 10}, {2019-06-26, 10}, {2019-06-27, 7}, {2019-06-28 , 9}, {2019-06-29, 5}, {2019-06-30, 7}.

Using hour, day, week and month as the time unit, draw a time series chart of the number of cases accumulated by hour, day, week and month as shown in Figure 8A to 8D. Observe the number of accumulated cases from 18 to 19 o'clock; each date, There was no significant change in the cumulative number of cases during the week, weekend, and each month.

Use the basic time unit to calculate the probability of each time unit. Using hours as the time unit, calculate the percentage of occurrences per hour in a day to get {0, 0.53%}, {1, 0.33%}, {2, 0.90%}, {3, 0.66%}, {4, 0.00%} , {5, 0.00%}, {6, 0.16%}, {7,0.86%}, {8,1.89%}, {9,1.89%}, {10,1.72%}, {11,3.04%}, {12,5.70%}, {13,8.17 %}, {14,6.98%}, {15,5.87%}, {16,5.87%}, {17,10.34%}, {18,12.97%}, {19,13.46%}, {20,11.57% }, {21, 4.84%}, {22, 1.72%}, {23, 0.53%}, there is no obvious regular phenomenon in the number of hourly case changes, so we choose to use neural network-like algorithm to model and forecast.

Set the main category, sub-category, administrative region, case status, case level, time of case, time of report, time of case closing, month, day, week, hour temperature, rainfall, wind direction, wind speed, altitude, relative humidity, air quality index value Attributes such as, sulfur dioxide, carbon monoxide, ozone, aerosols, fine aerosols, nitrogen dioxide, nitrogen oxides, nitric oxide, merchants, etc. are used as input variables. Among them, category variables will present all options, and continuous variables are divided into levels. Using a neural network model such as a plurality of selected predictive variables, predict the proportion of features and their value ranges to the probability of affecting the notification, and calculate the gain value.

Assuming that the predicted variable and its value is 19, the calculation formula of the gain value is: the reported case gain value = the predicted probability of occurrence at 19:00/the original probability of occurrence at 19:00, where the original probability of occurrence at 19:00 = the case at 19:00 Number/total number of cases. The obtained {predictive variable: range of options or values, whether there is a notification case, the gain value} are {hour: 19, yes, 5.44}, {hour: 18, yes, 5.19}, {hour: 20, yes, 5.03 }, {Hour: 23, None, 1.19}, {Hour: 0, None, 1.19}, {Hour: 5, None, 1.19}, {Month: 2, None, 1.14}, {Day: 3, None, 1.14 }, {week: Wednesday, yes, 1.8}, {ozone: 44.965-87.922, yes, 1.56}, {wind speed: 1.835-3.558, yes, 1.54}, {air quality index status: unhealthy for sensitive groups, none, 1.08 }, {Nitrogen Dioxide: 20.144-39.427, yes, 1.53}, {Air Quality Index State: unhealthy for all ethnic groups, none, 1.07}, {the highest temperature of the day: 13.100-22.144, none, 1.07}, {the highest temperature of the day: 28.133-35.000, yes, 1.4}, {wind speed: 0.000-0.836, none, 1.05 }, {Ozone: 2.600-20.046, none, 1.05}, {Relative humidity: 0.344-0.642, yes, 1.3}, {Sulfur dioxide: 2.947-5.785, none, 1.04}.

For the result presentation, please refer to the bar graph of whether the feature affects the reported case gain value in Figure 9. The middle vertical axis represents the feature variable and its options or value range, the double horizontal axis represents the gain value of the reported case, and the long bar The length represents the degree of gain that the feature affects whether there are reported cases. If the time is 19:00, 18:00, and 20:00, the probability of reporting a case is higher than the probability of not reporting a case, so the bar graph will focus on the side where the case is reported on the right; if the time is 23:00, 0:00 At 5 o'clock, the probability of not reporting a case is higher than the probability of reporting a case, so the bar graph will focus on the side where there is no reporting case on the left.

In order to select different characteristic variables, the top 3 gains calculated by their options or value ranges are {hour, 19}, {week, Wednesday}, {ozone, 44.965~87.922}, and then according to the hour, The week and ozone feature options or value ranges, find the first 3 dates and times suitable for dispatch, in order of 19:00 on 2019-07-03 (Wednesday), 19:00 on 2019-07-01 or 19:00 on 2019-07-02 19 o'clock.

The time sequence analysis system and method of petition cases provided by the present invention have the following advantages when compared with other conventional techniques:

1. The time series analysis system and method of petition cases of the present invention, in which the collection of external data is to collect real-time petitions and historical petitions in neighboring locations, and summarize repeated and continuous petition cases based on the repeatability and continuity of the notification, and understand the trend of the case.

2. The time series analysis system and method of petition cases of the present invention, wherein the exploration of petition time series uses narrative statistics and the method of drawing time series diagrams of the number of cases in different time units to understand the concentration and dispersion trend of the number of cases, and analyze the number of historical cases Whether the changes are regular.

3. The time series analysis system and method of petition cases of the present invention, wherein the time series prediction of regular cases uses a time series algorithm model to predict the number of cases in the future time interval, and predict the date of receiving notifications in the future based on the time interval with the largest number of cases Time shows the results of sequential analysis of regular cases, which is highly practical.

4. The time series analysis system and method of petition cases of the present invention, in which the prediction of characteristic cases uses a neural network-like algorithm model to analyze the probability of the case characteristics affecting the notification, according to the characteristics that affect the occurrence of the case and its gain value, and the ranking meets the conditions The date and time of day show the results of time series analysis of characteristic cases, which effectively assists smart city authorities to obtain sufficient information to prevent them from happening beforehand, so that the safety hazards caused by similar negligence can be minimized.

The above detailed description is a specific description of a possible embodiment of the present invention, but this embodiment is not intended to limit the scope of the patent of the present invention. Any equivalent implementation or modification that does not deviate from the technical spirit of the present invention should be included in In the scope of the patent in this case.

To sum up, this case is not only innovative in terms of technical ideas, but also can improve the above-mentioned multiple functions compared with conventional articles. It should fully meet the requirements of novel and progressive statutory invention patents. An application should be filed in accordance with the law. Please approve this case. Patent applications for inventions, to encourage invention, to the sense of virtue.

1‧‧‧Time sequence analysis system for petition cases

11‧‧‧External data collection module

12‧‧‧Chen Qing Time Series Exploration Module

13‧‧‧Regular case prediction module

14‧‧‧Characteristic case prediction module

2‧‧‧Interface platform

21‧‧‧Composition System

22‧‧‧Open Data System

23‧‧‧Big Data Analysis System

3‧‧‧Data Repository

31‧‧‧Case Database

32‧‧‧Disaster Information Database

33‧‧‧Geographic Information Database

34‧‧‧Meteorological Information Database

35‧‧‧Environmental Quality Database

Claims (10)

A time sequence analysis system for a submission case, including: an external data collection module, which stores the data obtained by interfacing a submission integration system and an open data system in a data repository after preprocessing, and then relies on a submission The notification time and location of the case, search for historical petition cases within the preset time range and location in the data repository, and determine whether the petition case is a repeated and continuous petition case in a nearby location; a petition time series exploration module, Receive the repeated and continuous submission case, analyze whether there is a regularity in the number of cases in different time units; if a regular case prediction module has regularity, the regular case prediction module uses a time series algorithm to model and predict. To provide time series analysis results of regular cases; and a feature case prediction module. If there is no regularity, the feature case prediction module uses neural network algorithm (ANN) to model and predict, thereby providing feature cases According to the time sequence analysis result, the model prediction system using the neural network-like algorithm includes: for the repeated and continuous petition cases with irregularities, query the history in the data repository according to the time and location range. Cases and features in an open data system are used as predictive variables; for the predictive variables, take the last T% as the test set, and the remaining (100-T)% as the training set; use this type of neural network algorithm to establish whether or not the predictive variable is The model for reporting the impact; use the classification matrix to calculate whether the correct rate is greater than the preset value; if it is not greater than the preset value, adjust the prediction variable, retain the feature number with better prediction effect, and remove the feature variable with poor prediction effect; If it is greater than the preset value, use the predicted variables modeled by the ANN algorithm to predict the probability of occurrence of each characteristic variable, and calculate the gain value of each characteristic variable. Among them,

According to the order of the gain value of each characteristic variable, sort the date and time that match the characteristic variable in the future time interval; and provide the time sequence analysis result of the characteristic case. For example, the petition case sequence analysis system described in item 1 of the scope of patent application, wherein the petition sequence exploration module describes the outline of the number of cases and draws a time series diagram of the number of cases for the repeated and continuous petition case, thereby analyzing different times Whether the number of cases in the unit changes regularly. For example, the time-series analysis system for petition cases described in the scope of patent application 1, wherein the regular case prediction module uses the time series algorithm to repeatedly model to predict the number of cases in the future time interval, and then the future time interval The date and time of possible receipt of notifications are sorted in order to provide the timing analysis results of the regular cases. For example, the time series analysis system of petition cases described in the scope of patent application, wherein the characteristic case prediction module uses this type of neural network algorithm to repeatedly model to calculate each characteristic variable and its gain value, thereby providing the Time series analysis results of characteristic cases. For example, the time sequence analysis system for petition cases described in the first item of the scope of patent application is connected to an interface platform, which includes: the petition integration system, which includes citizens’ suggestions on urban governance, reports of violations and deficiencies, and public rights Maintenance, after notification, the reporting integration system will perform the management and follow-up dispatch services; the open data system contains all kinds of open data collected by the government or related fields, and provides an interface for interfacing and collecting data; and A large data analysis system that provides users with the expected modeling algorithm, the ratio of the training set to the test set, and the imported file data. After repeated modeling and evaluation, the prediction results are output. For example, the time sequence analysis system of petition cases described in the first item of the scope of patent application is connected to the data repository. The data repository includes: a case database containing case types, locations, times, descriptions, and descriptions of real-time and historical cases Grade and status data; a disaster information database containing disaster information related to the city; a geographic information database containing geographic information related to the city; a meteorological information database containing historical and predicted meteorological information; And an environmental quality database, which contains historical and predicted environmental quality data. A time sequence analysis method for petition cases, including: collecting external data, receiving a petition case from a petition integration system to determine whether the petition case is a repeated and continuous petition case in a nearby location; and exploring the petition sequence for the repeated and continuous petition Report sentiment cases, analyze whether there is regularity in the number of cases in different time units; and if there is regularity, use time series algorithm to model and forecast to provide the time series analysis results of regular cases; and if there is no regularity, use class The neural network (ANN) algorithm performs modeling and prediction to provide time series analysis results of characteristic cases, wherein the use of neural network-like algorithm to perform modeling and prediction system includes: for the repeated and continuous complaint case, there is no regularity According to the time and location range, query the historical cases in the data repository and the features in an open data system as the prediction variable; for the prediction variable, take the last T% as the test set, and the remaining (100-T)% It is the training set; use this type of neural network algorithm to establish a model for predicting the impact of variables on the presence or absence of notification; use the classification matrix to calculate whether the correct rate is greater than the preset value; if it is not greater than the preset value, adjust the predicted variable and keep it The feature number with better prediction effect is removed from the feature variable with poor prediction effect; if it is greater than the preset value, the prediction variable obtained by the ANN algorithm modeling is used to predict the probability of occurrence of each feature variable, and calculate The gain value of each characteristic variable, among which,

According to the order of the gain value of each characteristic variable, sort the date and time that match the characteristic variable in the future time interval; and provide the time sequence analysis result of the characteristic case. For example, the time sequence analysis method of petition cases as described in item 7 of the scope of patent application, wherein the collection of external data includes: receiving the latest petitions in real time and obtaining external open data at regular intervals, and writing them into the data repository; and reporting according to the petition cases Time and location, query the historical petition data within the preset time range and preset location in the data repository; check whether there is historical petition data of the same case type; if it exists, the petition case is classified as a repeated petition case; If it does not exist, the petition case is classified as a single petition case; calculate whether the notification frequency of the petition case is greater than the preset value; if it is greater than the preset value, the repeated petition case is classified as a continuous petition case; and if not If it is greater than the preset value, the repeated petition case is classified as an occasional petition case. For example, the time sequence analysis method of petition cases described in item 7 of the scope of patent application, wherein the time sequence of exploration petitions includes: for the repeated and continuous petition case, calculating the average, mode, and median of the number of cases in different time units The statistical value of, maximum, minimum, interquartile, variance or standard deviation to grasp the number of repeated and continuous complaints and the trend of concentration or dispersion; plot the number of cases according to different basic time units Sequence diagram, arrange the number of cases in the order of time unit to understand the change in the number of repeated and continuous petition cases; use the basic time unit or extended time unit to calculate the proportion of the number of cases in each time unit to the total number of cases ; Look for whether there is a time unit with a proportion of more than 50% of the number of cases, and analyze whether there is a regularity in the number of repeated and continuous petition cases; if there is regularity, use the time series algorithm for modeling; and if not Regularity is modeled using this type of neural network algorithm. For example, the time series analysis method of petition cases described in item 7 of the scope of patent application, wherein the modeling and forecasting system using a time series algorithm includes: dividing the historical cases in the data repository into N according to the same time interval Divide equally; take the last Nth aliquot and divide it into the test set, and the first N-1 into the training set; use the integrated moving average autoregressive model (ARIMA) algorithm to establish a time series model; use the predictive model performance evaluation method, Evaluate whether the error value between the predicted result and the actual result is within the tolerance interval; if it is not within the tolerance interval, adjust the autoregressive term, the number of moving average terms and the number of difference parameters; If it is within the tolerance interval, use the best parameters modeled by the ARIMA algorithm to predict the number of cases in the future time interval; based on the maximum number of cases in the selected time unit, the future time interval may be accepted The date and time of the notification are listed in order; and the time sequence analysis result of the regular case is provided.

Images