CN117010939A

CN117010939A - Crowd dynamic diagram dynamic adjustment method based on deep learning

Info

Publication number: CN117010939A
Application number: CN202310689256.7A
Authority: CN
Inventors: 黄利鑫; 王喜瑞; 赖旦冉; 吴鹏; 徐亚波; 李旭日
Original assignee: Guangzhou Datastory Information Technology Co ltd
Current assignee: Guangzhou Datastory Information Technology Co ltd
Priority date: 2023-06-12
Filing date: 2023-06-12
Publication date: 2023-11-07
Anticipated expiration: 2043-06-12
Also published as: CN117010939B

Abstract

The application provides a crowd dynamic diagram dynamic adjustment method based on deep learning, which comprises the following steps: crowd data are acquired, and crowd line drawing is achieved; identifying sudden events and transient rules of crowd flow through real-time crowd flow trend prediction; judging whether the crowd movement brings business movement at the same time according to the moving trend of the business crowd moving line graph, and judging whether the movement is transient competition or long-term competition; the effective model transfer is carried out in different areas and among different groups, so that the general applicability of the crowd line graph trend model is ensured; analyzing consumption behaviors, frequency loyalty and life cycle values of the regional group in which the crowd is located after moving, and judging the advertising value of the crowd; combining merchant movement and crowd value, predicting the competition trend of the transferred merchant circles and the change of the advertisement input-output ratio; and dynamically adjusting the advertisement strategy according to the real-time change of the crowd dynamic diagram, and outputting advertisement recommendation results.

Description

Crowd dynamic diagram dynamic adjustment method based on deep learning

Technical Field

The application relates to the technical field of information, in particular to a crowd dynamic diagram dynamic adjustment method based on deep learning.

Background

With the continuous population migration and the acceleration of the urbanization process, population migration between different regions and different populations inevitably has a profound effect on business district competition trends. Conventional business turn competition analysis is typically based only on conventional market research, collecting extensive business data, analyzing market research and other business factors to explore business turn competition trends, and making predictions. However, these methods ignore population movements and also allow other merchants to learn about the opportunities and to have competing effects, failing to fully explain the nature of the business turn competition trends. Therefore, how to deeply analyze the influence of population flow on business district competition based on spatial information and population data has become one of the popular research directions in the commercial field.

Disclosure of Invention

The invention provides a crowd dynamic diagram dynamic adjustment method based on deep learning, which mainly comprises the following steps:

crowd data are acquired, and crowd line drawing is achieved; identifying sudden events and transient rules of crowd flow through real-time crowd flow trend prediction, wherein the sudden events and transient rules of crowd flow are identified through real-time crowd flow trend prediction, and the method specifically comprises the following steps: the ARIMA prediction algorithm predicts the influence trend of the emergency on the crowd flow, and predicts and recommends the crowd transient flow; judging whether the crowd movement brings business movement at the same time according to the moving trend of the business crowd moving line graph, and judging whether the movement is transient competition or long-term competition; the effective model transfer is carried out in different areas and among different groups, so that the general applicability of the crowd line graph trend model is ensured; analyzing consumption behaviors, frequency loyalty and life cycle values of the regional group in which the crowd is located after moving, and judging the advertising value of the crowd; combining merchant movement and crowd value, predicting the competition trend of the transferred merchant circles and the change of the advertisement input-output ratio; and dynamically adjusting the advertisement strategy according to the real-time change of the crowd dynamic diagram, and outputting advertisement recommendation results.

Further optionally, the obtaining crowd data, the implementing crowd line graph drawing includes:

acquiring the crowd number in a specific area by adopting an image recognition technology according to street and satellite shooting data; the moving direction and speed of the crowd are obtained by monitoring and tracking the images of the people; tracking and calculating the moving condition of the crowd to obtain the number of the moving crowd; the distribution situation of the crowd among different areas is obtained by analyzing the moving track of the crowd, so that the distribution percentage is determined; according to the data of the crowd quantity, the moving direction and the moving crowd quantity, the diversion percentage is combined, and the crowd moving situation is displayed in a small granularity of 450m x 450 m.

Further optionally, the identifying the emergency and transient rule of the crowd flow through real-time crowd flow trend prediction includes:

according to the crowd quantity, time, places, crowd attributes, vehicles, weather and events as influencing factors, a real-time data monitoring and predicting algorithm is adopted to predict crowd flow trend; firstly, determining a predicted range and a predicted time period by acquiring crowd quantity, time and place information; secondly, determining the flowing characteristics of the crowd according to the crowd attribute, the traffic means and the weather factors; then, the influence of the emergency on the crowd flow is identified through analysis and judgment of the event, and real-time monitoring and prediction are carried out; an ARIMA algorithm is adopted as a prediction algorithm, and the ARIMA algorithm carries out data preprocessing, feature extraction, model construction, model evaluation and real-time prediction aiming at influence factors, historical data and seasonal features, so as to judge whether the crowd flow is a conventional situation or a transient special situation; finally, determining the trend and rule of crowd flow by outputting a prediction result; comprising the following steps: the ARIMA prediction algorithm predicts the influence trend of the emergency on the crowd flow; predicting and recommending the transient flow of the crowd;

The ARIMA prediction algorithm predicts the influence trend of the emergency on the crowd flow, and specifically comprises the following steps:

by comparing and judging the emergency information with the crowd flow information, the influence trend of the emergency on the crowd flow is predicted by using statistics and a machine learning algorithm, and the method specifically comprises the following steps: an ARIMA prediction algorithm is adopted to predict the trend and the scale of the crowd flow: event type, event occurrence location, event size, and event duration data are obtained from an event monitoring system. The event data is checked, and if the data is not stable, differential processing is required to make the data stable. Parameters of the ARIMA model, including the order p of the AR model, the number of differences d, and the order q of the MA model, are determined using ACF and PACF. And fitting the ARIMA model by using the determined parameters to obtain a fitting result of the event data. And obtaining crowd characteristics, crowd quantity and crowd flow trend data from the crowd flow monitoring system. And comparing and judging the event data and the crowd flow data, and analyzing the influence of the event on the crowd flow. And predicting the influence trend and scale of the event on the crowd flow according to the fitted ARIMA model and crowd flow data. And checking the fitted model to ensure that the fitted model accords with the assumption of stationarity and white noise so as to ensure the accuracy of a prediction result. And obtaining a prediction result including a prediction value of the influence trend and the scale of the event on the crowd flow by an ARIMA prediction algorithm.

The method for predicting the transient flow of the crowd and recommending the business opportunity specifically comprises the following steps:

the crowd transient flow is predicted, business opportunity recommended data are preprocessed, data are obtained and cleaned according to crowd characteristics, geographic position information and behavior track data, different data sources are fused, characteristics are extracted, and preprocessed data are obtained. And selecting a proper prediction algorithm according to the preprocessed data by the model selection, and adopting an algorithm based on deep learning. The model training adopts a neural network method, the selected algorithm is trained, and parameter adjustment and model optimization are carried out according to the training result, so that an optimized model is obtained. The prediction and recommendation are carried out on new data by adopting a trained deep learning model, transient flow information of the crowd is obtained, and corresponding business opportunities are recommended.

Further optionally, the determining whether the crowd movement brings the merchant movement according to the moving trend of the merchant crowd moving line graph, and determining whether the movement is transient competition or permanent competition includes:

firstly, acquiring moving trend data of merchants and crowds, acquiring moving paths of the merchants and the crowds, and performing comparison analysis; determining whether the crowd movement brings the movement of the merchant at the same time by comparing the movement trend graphs of the merchant and the crowd; then, determining the moving purposes of the merchants and the crowd according to the moving trend of the merchants and the crowd; and judging whether the moving trend of the merchant is long-lasting competition or not by analyzing the moving aggregation period, the group type concentration degree and the product differentiation of the merchant.

Further optionally, the effective model transfer is performed between different areas and different groups, so that the general applicability of the crowd line graph trend model is ensured. Comprising the following steps:

determining the population activity ranges and travel modes of different areas and travel and activity links among the areas by acquiring population data, city planning data and traffic data of the areas; analyzing characteristics of different groups, including age, gender, income and occupation factors; determining the activity ranges and travel modes of different groups, and traveling and activity connection among the groups; judging the difference between different groups to ensure the accuracy of the model; judging cultural and habit differences between different areas and different groups, wherein the cultural and habit differences comprise customs, holiday culture and consumption habit factors, the economic development level, consumption capacity and price level of different areas, taking the factors as model special values, and judging the influence of the factors on crowd line; training a Bayesian classification model by taking the judging conditions as input features of the model, wherein the label value is general and special classification; after the data are marked, a model universal applicability model can be trained to judge the applicability of the original model.

Further optionally, the analysis of the consumption behavior, the loyalty and the life cycle value of the crowd in the area after the crowd moves, and the judgment of the advertisement value of the crowd. Comprising the following steps:

acquiring purchasing power, consumption habit and consumption preference information of new moving people, and attitude and response data of different products or services; according to the consumption track of the crowd, obtaining consumption frequency and loyalty data of the crowd; judging the trust degree of a crowd on a certain brand or product and the opportunity of repeated purchase; the method is used for evaluating the commercial value of the crowd and providing a data basis for advertising of brands or products; carrying out relation analysis among crowd consumption behaviors, loyalty and life cycle values by adopting an association rule algorithm, and obtaining crowd advertising values; firstly, data preprocessing, including cleaning, de-duplication and discretization of data; secondly, searching for a combination of which the occurrence frequency of the item sets exceeds a preset threshold by adopting a method for mining frequent item sets, and generating an association rule; obtaining association information among crowd consumption behaviors, loyalty and life cycle values according to an association rule algorithm; judging which commodities are purchased together frequently and which commodities have an effect of improving loyalty and life cycle value by analyzing frequent item sets and association rules; meanwhile, the correlation rule algorithm is used for obtaining the relevant attributes contained in the advertisement value of the crowd, including the hobbies, life style and region of the crowd; according to the analysis result, determining the relevant attribute of the crowd advertisement value, and formulating an accurate advertisement marketing strategy; different advertisement marketing schemes are designed for different consumer groups according to consumption behaviors and loyalty, promotion is conducted according to different attributes, and a targeted and accurate advertisement putting strategy is provided for advertisers.

Further optionally, combining the merchant movement and the crowd value, predicting the change of the competition trend of the merchant circles and the advertisement input-output ratio after the transfer. Comprising the following steps:

according to the classification of the moving destination and the crowd value of the merchant, historical data are obtained and time sequence analysis is carried out to predict the change of the business circle competition trend and the advertisement input-output ratio; firstly, collecting moving destination and crowd value classification information of merchants, and dividing a business circle into different areas based on the moving destination and crowd value classification information; then, collecting historical data, including sales data of merchants in the business circles, advertisement input data and number information of competitors in each time period; taking the data characteristics with the time sequences as input, and performing time sequence analysis, including analysis of stability test, white noise test, autocorrelation function and partial autocorrelation function of the time sequences; selecting a model to predict according to a time sequence analysis result, wherein the model comprises an ARIMA model and an exponential smoothing model; predicting the change conditions of business district competition trend and advertisement input-output ratio in a future period of time; judging the change condition of the input-output ratio according to the prediction result; if the competition intensity of the new business district is reduced after the business is moved, the advertisement input-output ratio is expected to be improved; in this case, a strategy of increasing the advertisement putting range and form is adopted to improve the input-output ratio; if the value of the destination crowd is low after the merchant moves, the advertisement input-output ratio is expected to be reduced; in this case, the strategy and the range of advertisement delivery are adjusted to reduce unnecessary investment, thereby improving the input-output ratio; if the advertisement input is increased after the merchant moves, but the input-output ratio is not obviously improved due to the strong competition, the input-output ratio is improved by reducing the advertisement input range, adjusting the input time period and other strategies.

Further optionally, the dynamically adjusting the advertisement strategy according to the real-time change of the crowd dynamic diagram, and outputting the advertisement recommendation result under the line includes:

recommending advertisements by adopting an algorithm based on geographic positions, crowd attributes, behavior habits and preference characteristics according to the real-time change of crowd dynamic patterns so as to dynamically adjust advertisement strategies; firstly, judging the characteristics and the trend of the area where the crowd is located and the flow and the distribution of the crowd by acquiring the geographical position information of the crowd; acquiring the current position of the crowd; determining the current position of the crowd by adopting a method based on the LBS geographic position, obtaining longitude and latitude information of the crowd, converting the longitude and latitude information into specific address information through a map API, determining the current position of the crowd, and predicting the subsequent position along with a time sequence; judging places where people go frequently and the time of stay at when and where by analyzing the behavior track; corresponding advertisements are recommended for the characteristics and trends of different areas to attract the attention of people.

The technical scheme provided by the embodiment of the invention can have the following beneficial effects:

the method analyzes and predicts the crowd moving picture, digs the transient flow of the crowd, avoids the misprediction of the crowd flow and provides help for the rapid response of the transient flow. And meanwhile, after capturing crowd moving images, judging the dynamic state of surrounding business district merchants, and judging the future competition trend so as to ensure the accurate prediction of business district analysis.

Drawings

Fig. 1 is a flow chart of a crowd dynamic diagram adjustment method based on deep learning.

Fig. 2 is a schematic diagram of a crowd moving picture dynamic adjustment method based on deep learning.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.

The crowd moving image dynamic adjustment method based on deep learning in the embodiment specifically comprises the following steps:

step 101, crowd data is obtained, and crowd line drawing is achieved.

And acquiring the crowd number in the specific area by adopting an image recognition technology according to the street and satellite camera data. By monitoring and tracking the images of the people, the moving direction and speed of the people are obtained. Tracking and calculating the moving condition of the crowd, and obtaining the moving crowd quantity. And analyzing the moving track of the crowd to obtain the diversion situation of the crowd among different areas, so as to determine the diversion percentage. According to the data of the crowd quantity, the moving direction and the moving crowd quantity, the diversion percentage is combined, and the crowd moving situation is displayed in a small granularity of 450m x 450 m. For example, based on street and satellite camera data, image recognition techniques are used to obtain the number of people in a particular area, for example 3000 people in an hour. By monitoring and tracking the images of the person, the direction and speed of movement of the crowd, for example, moving the crowd eastward during 9 to 10 am, is obtained, with an average speed of 10 meters per minute. The movement of the crowd is tracked and calculated to obtain the number of people moving, such as 1200 people leaving the area during the period of 4 to 5 pm. The diversion situation of the crowd between different areas is obtained by analyzing the moving track of the crowd, so that the diversion percentage is determined, for example, 60% of the crowd flows to a business area, 30% of the crowd flows to a residential area and 10% of the crowd flows to a public area in the period from 8 to 9 in the morning. According to the data of the crowd quantity, the moving direction and the moving crowd quantity, the diversion percentage is combined, the small granularity display crowd moving condition of 450m x 450m is realized, for example, the crowd quantity moving to the east is 800 people in a commercial area in the period of 2 to 3 pm, the crowd quantity moving to the west is 600 people, and the crowd quantity accounting for 40 percent and 30 percent of the total crowd quantity is marked by different colors on a map respectively.

Step 102, identifying sudden events and transient rules of crowd flow through real-time crowd flow trend prediction.

And predicting the crowd flow trend by adopting a real-time data monitoring and predicting algorithm according to the crowd quantity, time, places, crowd attributes, vehicles, weather and events as influencing factors. First, by acquiring crowd quantity, time and place information, a predicted range and time period are determined. Secondly, determining the flowing characteristics of the crowd according to the crowd attributes, the vehicles and the weather factors. Then, through analysis and judgment of the events, the influence of the emergency on the crowd flow is identified, and real-time monitoring and prediction are performed. An ARIMA algorithm is adopted as a prediction algorithm, and the ARIMA algorithm carries out data preprocessing, feature extraction, model construction, model evaluation and real-time prediction aiming at influence factors, historical data and seasonal features, so as to judge whether the crowd flow is a conventional situation or a transient special situation; and finally, determining the trend and rule of the crowd flow by outputting the prediction result. For example, in a business district in a city, about 5000 customers go to the business district for shopping every day based on data analysis of the past week. According to weather forecast, a strong rainfall weather appears at the end of the week, and a certain influence on the flow of people is expected. In combination with the characteristics of the business district and traffic conditions, it is expected that the number of people in the business district will drop to around 3000 people in this weather. Therefore, the weather condition and the crowd quantity can be monitored in real time, the customer flow of the business district on the weekend can be predicted to be reduced by about 40%, and the merchant can make corresponding countermeasures according to the prediction result, such as reducing the inventory, adjusting the business hours and the like, so as to adapt to the change of the customer flow. Meanwhile, by analyzing and evaluating the crowd flowing trend regularly, merchants can be helped to formulate more scientific and reasonable operation strategies, and production efficiency and economic benefits are improved.

The ARIMA prediction algorithm predicts the trend of the impact of an incident on crowd flow.

By comparing and judging the emergency information with the crowd flow information, the influence trend of the emergency on the crowd flow is predicted by using statistics and a machine learning algorithm, and the method specifically comprises the following steps: an ARIMA prediction algorithm is adopted to predict the trend and the scale of the crowd flow: event type, event occurrence location, event size, and event duration data are obtained from an event monitoring system. The event data is checked, and if the data is not stable, differential processing is required to make the data stable. Parameters of the ARIMA model, including the order p of the AR model, the number of differences d, and the order q of the MA model, are determined using ACF and PACF. And fitting the ARIMA model by using the determined parameters to obtain a fitting result of the event data. And obtaining crowd characteristics, crowd quantity and crowd flow trend data from the crowd flow monitoring system. And comparing and judging the event data and the crowd flow data, and analyzing the influence of the event on the crowd flow. And predicting the influence trend and scale of the event on the crowd flow according to the fitted ARIMA model and crowd flow data. And checking the fitted model to ensure that the fitted model accords with the assumption of stationarity and white noise so as to ensure the accuracy of a prediction result. And obtaining a prediction result including a prediction value of the influence trend and the scale of the event on the crowd flow by an ARIMA prediction algorithm. In another embodiment, data such as event type, event occurrence location, event size, and event duration of a business turn fire is obtained from an event monitoring system. Checking the stability of the data, checking the event scale data of the sudden fire, and if the data is not stable, carrying out one-time differential processing to ensure that the data is stable. Model parameters are determined using ACF and PACF to determine parameters of the ARIMA model, including the order p=2 of the AR model, the number of differences d=1, and the order q=1 of the MA model. And fitting the model, namely fitting the ARIMA model by using the determined parameters to obtain a fitting result of the emergency scale data. And acquiring crowd flow data, and acquiring data such as the number of people around the fire scene, the crowd characteristics, the crowd flow trend and the like from a crowd flow monitoring system. Comparing and judging, namely comparing and judging the event scale data of the sudden fire with the surrounding crowd quantity data, and analyzing the influence of the sudden fire on crowd flow. Predicting, according to the fitted ARIMA model and crowd flow data, predicting the influence trend and scale of the sudden fire on crowd flow. And (3) model inspection, namely inspecting the fitted model to ensure that the fitted model accords with the assumption of stationarity and white noise so as to ensure the accuracy of a prediction result. Outputting a prediction result, and obtaining the predicted value of the influence trend and scale of the sudden fire on the crowd flow by an ARIMA prediction algorithm, wherein the predicted value is as follows: the population flow will increase by 50 in the next 2 hours.

And predicting and recommending the transient flow of the crowd.

The crowd transient flow is predicted, business opportunity recommended data are preprocessed, data are obtained and cleaned according to crowd characteristics, geographic position information and behavior track data, different data sources are fused, characteristics are extracted, and preprocessed data are obtained. And selecting a proper prediction algorithm according to the preprocessed data by the model selection, and adopting an algorithm based on deep learning. The model training adopts a neural network method, the selected algorithm is trained, and parameter adjustment and model optimization are carried out according to the training result, so that an optimized model is obtained. The prediction and recommendation are carried out on new data by adopting a trained deep learning model, transient flow information of the crowd is obtained, and corresponding business opportunities are recommended. For example, based on crowd characteristics, geographical location information, and behavior trace data, after acquiring and cleansing the data, we obtained a set of data containing 100,000 samples. Wherein the characteristics include age, gender, occupation, income, marital status and the like, the geographic location information includes cities, counties, streets and the like, and the behavior trace data includes daily activities, consumption preferences, shopping habits and the like. In the data preprocessing process, a feature engineering technology is used for fusing and extracting features from different data sources, so that a simplified data set is obtained. In the model selection stage, a Convolutional Neural Network (CNN) based on deep learning is adopted for transient flow prediction and business recommendation. The CNN algorithm can mine potential rules and modes from a large amount of crowd characteristics and behavior data, and has high accuracy and stability. In the model training stage, a neural network method is used for training a CNN algorithm, and parameter adjustment and model optimization are carried out according to a training result, so that an optimized model is obtained. For example, during training, the data set is divided into training and testing sets using cross-validation techniques to avoid over-fitting and under-fitting problems. And finally, in the prediction and recommendation stage, predicting and recommending new data by using the trained deep learning model, obtaining transient flow information of the crowd, and recommending a corresponding business opportunity. For example, it may be predicted which consumption activities or shopping activities, such as purchasing umbrellas, are favored by the population in a certain area during a certain period of rain, and recommend related products or services to the merchant. Therefore, more accurate business opportunity matching can be realized, and the business operation effect is improved.

And step 103, judging whether the crowd movement brings the movement of the merchant at the same time according to the moving trend of the group dynamic diagram of the merchant, and judging whether the movement is transient competition or permanent competition.

Firstly, acquiring moving trend data of merchants and crowds, acquiring moving paths of the merchants and the crowds, and performing comparison analysis. And determining whether the crowd movement brings the movement of the merchant at the same time by comparing the movement trend graphs of the merchant and the crowd. And then, determining the moving purposes of the merchant and the crowd according to the moving trend of the merchant and the crowd. And judging whether the moving trend of the merchant is long-lasting competition or not by analyzing the moving aggregation period, the group type concentration degree and the product differentiation of the merchant. For example, based on the trend of the line of action diagram of the merchant population, we found that on the weekend, the trend of the population movement increased significantly, while the merchant also had a corresponding movement. By analysing the movement paths of the merchants and the people, we find that the movement paths of the people are mainly concentrated in the shopping malls and the dining streets of the businessman, whereas the movement paths of the merchants are mainly concentrated in the shopping malls and the shopping streets around the dining streets. It can be determined that the movement of the crowd does bring about the movement of the merchant. Assuming that we analyze the restaurant-merchant in a certain area, by counting the mobile aggregation period, it is found that the restaurant-merchant in the area does not only business on weekends, but also walks for one month, and the probability of long-term competition for migration is high. For example, if a restaurant in a home provides a unique spicy soup, has a unique taste, and no other businesses in the area provide similar products, the restaurant may attract more consumers and receive higher profits, and thus a high probability of long-lasting competition. This migration is not an exacerbation of transient contention, but rather is long lasting.

Step 104, performing effective model transfer between different areas and different groups, and ensuring the universal applicability of the crowd line graph trend model.

And determining the population activity ranges and travel modes of different areas and travel and activity links among the areas by acquiring population data, city planning data and traffic data of the areas. The characteristics of different groups are analyzed, including age, gender, income and occupational factors. Determining the activity ranges and travel modes of different groups, and traveling and activity connection among the groups; and judging the difference between different groups to ensure the accuracy of the model. Judging cultural and habit differences between different areas and different groups, wherein the cultural and habit differences comprise customs, holiday culture and consumption habit factors, the economic development level, consumption capacity and price level of different areas, taking the factors as model special values, and judging the influence of the factors on the crowd line. And training a Bayesian classification model by taking the judgment conditions as input features of the model, wherein the label value is of general classification and special classification. After the data are marked, a model universal applicability model can be trained to judge the applicability of the original model. For example, the population activity ranges and travel modes of different areas are determined by analyzing population data, city planning data and traffic data of Beijing city. And determining the activity ranges and the travel modes of different groups and the travel and activity relation among the groups. For example, by analyzing cultural and habitual attributes of beijing, it is known that beijing residents can go home and home to visit during spring festival, which can have a certain influence on crowd moving lines. In addition, economic attributes of Beijing are also important factors affecting crowd movement. For example, beijing has a high consumption capacity and a relatively high price level, which affects people's travel and activity. Through the analysis, the travel and activity modes of Beijing city residents can be predicted, so that a crowd line graph trend model with universal applicability is generated.

Step 105, analyzing the consumption behavior, the loyalty and the life cycle value of the regional group where the crowd is moving, and judging the advertisement value of the crowd.

Purchasing power, consumption habit and consumption preference information of new moving people are obtained, and attitude and response data of different products or services are obtained. And acquiring consumption frequency and loyalty data of the crowd according to the consumption track of the crowd. The trust degree of a crowd on a certain brand or product and the opportunity of repeated purchase are judged. For assessing the commercial value of a population and providing a data base for advertising of brands or products. And carrying out relation analysis among the crowd consumption behaviors, loyalty and life cycle values by adopting a correlation rule algorithm, and obtaining crowd advertising values. Firstly, data preprocessing, including cleaning, de-duplication and discretization of data; secondly, searching for a combination of which the occurrence frequency of the item sets exceeds a preset threshold by adopting a method for mining frequent item sets, and generating an association rule; and obtaining the association information among the crowd consumption behaviors, the loyalty and the life cycle value according to an association rule algorithm. By analyzing the frequent item sets and association rules, it is determined which commodities are purchased together and which have an effect on improving loyalty and life cycle value. Meanwhile, the correlation rule algorithm is used for obtaining the relevant attributes contained in the advertisement value of the crowd, including the hobbies, the life style and the region of the crowd. And determining the relevant attribute of the crowd advertising value according to the analysis result, and formulating an accurate advertising marketing strategy. Different advertisement marketing schemes are designed for different consumer groups according to consumption behaviors and loyalty, promotion is conducted according to different attributes, and a targeted and accurate advertisement putting strategy is provided for advertisers. For example, by statistical analysis, we obtain an average consumption of 500 yuan for a population in a certain area in one month. Further analysis found that these groups prefer to purchase household items and cosmetics, while the purchase will be relatively low for digital products. Therefore, people can obtain that the consumption preference of household articles and cosmetics is high, and guidance can be provided for advertisement delivery in the industries. For example, by analyzing the consumption trace data of a population, we find that a brand has 50% loyalty, i.e., half of the population will purchase the brand again. Further analysis found that these higher loyalty populations were mainly under 30 years of age and less than 5 ten thousand yuan in income. Therefore, people can obtain that the trust degree of the people on the brand is high, and guidance can be provided for advertisement delivery of the brand. For example, according to data mining techniques, association rule algorithms are employed to analyze relationships between crowd consumption behavior, loyalty, and life cycle value, and to obtain crowd advertising value. First, data preprocessing is performed, for example, 1000 purchase records are cleaned, 800 records remain after duplicate removal, and discretization processing is performed on the price. Secondly, a method of mining frequent item sets is adopted, combinations with the occurrence frequency of the item sets exceeding 5% are found, and association rules are generated. For example, it was found that apples and oranges are often purchased together, resulting in the rule { apple } - > { orange }, with 10% support, 80% confidence, and 2 improvement. And finally, evaluating and explaining the analysis result. According to the rule { apple } - > { orange }, it can be inferred that apples and oranges are mutually replaced commodities or mutually complementary commodities, and corresponding marketing strategies can be adopted to improve sales and loyalty. According to the association rule algorithm, people buying roast ducks and red wine are also found to come from Beijing areas frequently, which can help enterprises locate relevant attributes of the advertising value of the people and formulate targeted advertising marketing strategies.

And 106, predicting the competition trend of the business district and the change of the advertisement input-output ratio after the transfer by combining the movement of the business and the crowd value.

According to the classification of the moving destination and the crowd value of the merchant, historical data are obtained and time series analysis is carried out to predict the change of the business district competition trend and the advertisement input-output ratio. First, moving destination and crowd value classification information of merchants are collected, and on the basis of the moving destination and crowd value classification information, a business circle is divided into different areas. Next, historical data is collected, including sales data, advertising placement data, and competitor's quantity information for the merchants within the business circle over each time period. Taking the data characteristics with the time sequences as input, performing time sequence analysis, including analysis of stability test, white noise test, autocorrelation function and partial autocorrelation function of the time sequences. And selecting a model to predict according to the time sequence analysis result, wherein the model comprises an ARIMA model and an exponential smoothing model. And predicting the change conditions of business district competition trend and advertisement input-output ratio in a future period. And judging the change condition of the input-output ratio according to the prediction result. If the new business district competition intensity is reduced after the business is moved, the advertisement input-output ratio is expected to be improved. In this case, strategies are adopted to increase the advertising scope and form so as to improve the input-output ratio. If the value of the destination population is low after the merchant moves, the advertisement input-output ratio is expected to be reduced. In this case, the strategy and scope of advertisement delivery are adjusted to reduce unnecessary investment, thereby improving the input-output ratio. If the advertisement input is increased after the merchant moves, but the input-output ratio is not obviously improved due to the strong competition, the input-output ratio is improved by reducing the advertisement input range, adjusting the input time period and other strategies. And taking all time series data characteristics as input, and performing time series analysis. And judging whether the sequence is stable or not according to the stability test and the white noise test result of the time sequence. If not, it is converted into a plateau sequence by differential methods, otherwise this step is skipped. Next, an autocorrelation function and partial autocorrelation function analysis is performed to determine an appropriate model. The features of the time series are described by the order of the ARIMA model, including AR, MA and differential order, resulting in a suitable ARIMA model. And then, selecting an exponential smoothing model for prediction according to the characteristics of the historical data. Parameters of the smoothing coefficients and initialization values are determined and predicted as inputs. After the prediction result is obtained, the prediction result is compared with the actual result to judge the prediction capability of the model. And finally, carrying out model evaluation according to the prediction result. And judging the accuracy and the reliability of the model by comparing the error between the predicted result and the actual result. If the model needs to be improved, parameter adjustment is carried out according to the evaluation result, and prediction and evaluation are carried out again until a satisfactory prediction result is obtained. Through the steps, time series analysis and prediction can be carried out on sales data, advertisement investment data and competitor quantity information of merchants in the business district. According to the prediction result, merchants can adopt corresponding strategies to improve sales performance, strengthen competition between advertising investment and competitors, and improve the development level of business circles. If sales data of a restaurant in a business district in the past three months is [10000,12000,13000] yuan, advertisement input data is [2000,2500,3000] yuan, and the number of competitors is [3,4,5 ]. Through time sequence analysis, a suitable ARIMA model is ARIMA (1, 1), and an exponential smoothing model is Holt-windows seasonal model. And predicting according to the historical data and the model parameters to obtain a predicted sales value of 13759 yuan, a predicted advertisement investment value of 3150 yuan and a predicted number of competitors of 6. And (3) comparing the errors of the actual value and the predicted value, evaluating the accuracy and the reliability of the model, and if the errors are large, performing parameter adjustment and re-predicting and evaluating until a satisfactory predicted result is obtained. Merchants can formulate corresponding marketing strategies according to the prediction results, and sales performance and competitiveness are improved. For example, by increasing advertising investment and improving restaurant services, more customers are attracted to consume the product before, and meanwhile, competition with competitors is enhanced, so that the overall development level of the business is improved. Assuming that before a merchant moves, a certain advertisement is put into 1 ten thousand yuan, the sales amount is 10 ten thousand yuan, and the input-output ratio is 10; the degree of business district competition is medium. After the merchant moves, the destination population value increases, but the degree of competition increases. Through time series analysis, it is predicted that the advertisement input-output ratio will drop to 8. Thus, merchants may consider taking strategies to reduce the scope of advertising, only in the vicinity of the destination, while increasing the creative of the ad format and content to increase the input-to-output ratio. If the advertisement input is reduced to 8 ten thousand yuan after implementation and adjustment, the sales amount is 4 ten thousand yuan, the input-output ratio is 8, and the expected effect is met.

And 107, dynamically adjusting the advertisement strategy according to the real-time change of the crowd dynamic diagram, and outputting the advertisement recommendation result.

And recommending the advertisements by adopting an algorithm based on geographic positions, crowd attributes, behavior habits and preference characteristics according to the real-time change of the crowd dynamic diagram so as to dynamically adjust advertisement strategies. Firstly, by acquiring geographic position information of the crowd, the characteristics and the trend of the area where the crowd is located and the flow and the distribution of the crowd are judged. Acquiring the current position of the crowd; determining the current position of the crowd by adopting a method based on the LBS geographic position, obtaining longitude and latitude information of the crowd, converting the longitude and latitude information into specific address information through a map API, determining the current position of the crowd, and predicting the subsequent position along with a time sequence. By analyzing the behavior track, the places where people go frequently and the time when and where people stay are judged. Corresponding advertisements are recommended for the characteristics and trends of different areas to attract the attention of people. If population age distribution shows an aging trend in a certain area, people can recommend advertisements of medical health class so as to meet the health requirement of the aged; for example, by analyzing crowd attributes, a user is female, the age is between 20 and 30 years, the user likes sports and body building, and the user can recommend advertisements of sports and body building types to meet sports requirements; longitude and latitude information of a user on weekends is converted into specific address information through a map API. For example, from 2 pm to 4 pm on Saturday, 100 users stay in a cafe in a downtown area for at least 1 hour. By analyzing the behavior trace of the batch of users, we can find that the cafe is one of the places they often go to. For this case, we can recommend relevant advertisements, such as coupons or campaign information, to the group of users to attract their attention.

The preferred embodiments of the invention disclosed above are intended only to assist in the explanation of the invention. The preferred embodiments are not intended to be exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best understand and utilize the invention. The invention is limited only by the claims and the full scope and equivalents thereof.

Claims

1. The crowd dynamic diagram dynamic adjustment method based on deep learning is characterized by comprising the following steps:

2. The method of claim 1, wherein the obtaining crowd data to enable crowd line mapping comprises:

3. The method of claim 1, wherein the identifying sudden events and transient laws of crowd flow through real-time crowd flow trend prediction comprises:

according to the crowd quantity, time, places, crowd attributes, vehicles, weather and events as influencing factors, a real-time data monitoring and predicting algorithm is adopted to predict crowd flow trend; firstly, determining a predicted range and a predicted time period by acquiring crowd quantity, time and place information; secondly, determining the flowing characteristics of the crowd according to the crowd attribute, the traffic means and the weather factors; then, the influence of the emergency on the crowd flow is identified through analysis and judgment of the event, and real-time monitoring and prediction are carried out; an ARIMA algorithm is adopted as a prediction algorithm, and the ARIMA algorithm carries out data preprocessing, feature extraction, model construction, model evaluation and real-time prediction aiming at influence factors, historical data and seasonal features, so as to judge whether the crowd flow is a conventional situation or a transient special situation; finally, determining the trend and rule of crowd flow by outputting a prediction result; comprising the following steps: the ARIMA prediction algorithm predicts the influence trend of the emergency on the crowd flow; and predicting and recommending the transient flow of the crowd.

4. The method of claim 1, wherein the determining whether the crowd movement simultaneously brings the merchant movement according to the moving trend of the merchant crowd moving, and determining whether the movement is transient or long-lasting competition comprises:

5. The method of claim 1, wherein said effecting model transfer between different regions, different populations, ensures universal applicability of crowd-line graph trend models, comprising:

6. The method of claim 1, wherein analyzing consumption behavior, loyalty and life cycle value of the crowd in the area after the crowd moves, determining the crowd advertising value comprises:

7. The method of claim 1, wherein predicting the post-transfer business turn competition trends and advertisement input-output ratio changes in combination with merchant movement and crowd value comprises:

8. The method of claim 1, wherein the dynamically adjusting the advertisement strategy according to the real-time change of the crowd dynamic diagram, outputting the advertisement recommendation result comprises: