Electronic cigarette use frequency prediction system based on data analysis

Abstract

The invention relates to the technical field of data analysis and prediction, in particular to an electronic cigarette use frequency prediction system based on data analysis, which comprises a time interval division module, a behavior dynamic analysis module, an interval trend prediction module and a global prediction optimization module. According to the method, through data extraction and division based on a time sequence, segmented dynamic analysis is carried out on key data such as use frequency, short-term fluctuation and single interval, data distribution boundary and continuity are optimized, a data change rule is accurately captured, interference of data deviation on modeling is reduced, and analysis of segmented behavior data is combined, so that modeling accuracy is improved. According to the method, dynamic behavior offset parameters are extracted, a multi-dimensional feature data set is constructed, prediction is closer to an actual use situation, and prediction accuracy and decision value are remarkably improved through trend parameterization processing of high-frequency fluctuation data, recursion of interval trend features, optimization of global fluctuation amplitude and adjustment of behavior changes and enhancement of global and local data linkage.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis and prediction, and particularly to an electronic cigarette usage times prediction system based on data analysis. Background Art

[0002] The technical field of data analysis and prediction includes using data analysis, mining, and modeling methods to process, summarize, and predict information in a data set to solve various practical problems. The core content of this technical field includes extracting features from multi-source data through methods such as statistics, machine learning, and deep learning, analyzing the change trends, correlation relationships, and potential laws of the data. Data analysis and prediction technology are widely applied in multiple fields such as business prediction, health monitoring, and behavior analysis, mainly relying on data acquisition, cleaning, feature extraction, and model construction to provide a basis for decision-making through quantitative analysis.

[0003] Among them, the electronic cigarette usage times prediction system refers to a technical solution that predicts the change in its usage times by analyzing electronic cigarette-related data. This patent theme targets the usage behavior of electronic cigarette users and involves collecting multi-dimensional data such as usage frequency, time interval, and smoking habits of users' devices, extracting features and classifying and summarizing the data, and combining prediction models such as regression analysis models and classification models to perform predictive analysis of future usage times. Basic data is obtained through a data collection device, and a mathematical model is used to model and optimize the multi-dimensional data to generate a prediction result, thereby completing the accurate prediction of the electronic cigarette usage times.

[0004] In the prediction of electronic cigarette usage behavior in the prior art, it mainly relies on simple classification and model application of basic data such as usage frequency and time interval, lacking fine-grained dynamic analysis for short-term fluctuations and behavior deviations, and being unable to effectively capture complex fluctuation characteristics in the time series, resulting in insufficient understanding and representation of the data change law, and being prone to prediction errors in high-frequency fluctuation or behavior anomaly scenarios. The prior art focuses on the accuracy of local prediction in model optimization, ignoring the dynamics of the global and local data association, and it is difficult to balance the prediction consistency of the global trend and local changes. The lack of this global adjustment ability makes it easy to amplify the prediction error due to deviation accumulation in the recursion between multi-interval data, weakening the adaptability of the prediction result to complex behavior scenarios and resulting in a decline in the reliability of the prediction result in complex scenarios, limiting the support for actual decision-making. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art and propose an electronic cigarette usage times prediction system based on data analysis.

[0006] To achieve the above purpose, the present invention adopts the following technical solution: The electronic cigarette usage times prediction system based on data analysis includes:

[0007] Based on the time series of e-cigarette usage data, the time interval division module extracts the e-cigarette usage frequency, short-term fluctuation interval, and single-use interval duration data, analyzes the fluctuation law and change range of the time period, compares the offset values of the interval change amplitude and continuity, adjusts the interval boundary and distribution, and obtains the segmented usage dynamic structure;

[0008] Based on the segmented usage dynamic structure, the behavior dynamic analysis module captures the single inhalation volume, usage frequency, and behavior offset characteristics within each interval, analyzes the differential distribution of the frequency and inhalation interval within each interval, obtains the dynamic behavior offset parameters, determines the interval correlation of the dynamic behavior offset parameters, and generates a dynamic behavior characteristic data set;

[0009] Based on the dynamic behavior characteristic data set, the interval trend prediction module analyzes the usage peak and high-frequency time periods in the interval, performs parametric processing through the trend relationship between the frequency change value and the behavior offset amount, obtains the interval prediction characteristics, recursively derives the interval trend characteristics, adjusts the behavior offset amount of the trend characteristics, and generates a multi-interval trend prediction result;

[0010] Based on the multi-interval trend prediction result, the global prediction optimization module analyzes the offset amplitude and behavior change range in the global prediction, adjusts the high-frequency deviation of the interval trend prediction, obtains the global trend dynamic value, combines the global trend dynamic value to correct the local prediction data, and generates the e-cigarette global usage times prediction result.

[0011] As a further solution of the present invention, the specific steps for obtaining the segmented usage dynamic structure are as follows:

[0012] Based on the time series of e-cigarette usage data, extract the usage frequency and interval duration of each time period, analyze the usage frequency change characteristics of each time period, analyze the fluctuation amplitude, screen and eliminate data, and obtain the interval division result that conforms to the fluctuation law;

[0013] According to the interval division result that conforms to the fluctuation law, calculate the extreme value of each interval, compare the boundary value of each interval with the continuity of the adjacent interval, adjust the dynamic structure of the interval, and generate the adjusted interval distribution result;

[0014] Based on the adjusted interval distribution result, perform trend analysis on short-term and long-term fluctuations, combine the fluctuation interval boundary, and use the formula:

[0015]

[0016] Obtain the segmented usage dynamic structure;

[0017] Among them, S represents the segmented usage dynamic structure score, n represents the number of time periods, w iThe weight coefficient representing each fluctuation range, T end and T start respectively represent the start time and end time of the interval, and ΔT i represents the change amplitude within the time period, and abs is the absolute value function.

[0018] As a further solution of the present invention, the steps for obtaining the dynamic behavior offset parameter are specifically as follows:

[0019] Based on the segmented use of the dynamic structure, extract the single inhalation amount and usage frequency within each interval, record the single inhalation data points within each time interval, statistically analyze the distribution of inhalation interval times within each interval, calculate the variance and mean of the inhalation interval distribution, mark the intervals where the variance exceeds the distribution range, and obtain the interval results of the single inhalation amount and frequency;

[0020] Based on the interval results of the single inhalation amount and frequency, compare the relationship between the inhalation frequency and inhalation interval within each interval, and use the formula:

[0021]

[0022] Calculate the dynamic behavior offset value, weight the inhalation interval and reflect the degree of behavior offset, and generate the interval offset value distribution result;

[0023] where P is the dynamic behavior offset value, A j is the inhalation amount at the j-th inhalation point, I j is the inhalation interval time at the j-th inhalation point, α is the inhalation time adjustment parameter, β is the inhalation interval weight factor, and m is the number of inhalation points within the interval;

[0024] According to the interval offset value distribution result, screen the intervals where the offset value deviates from the mean, adjust the inhalation amount and frequency distribution of the abnormal intervals, and integrate them using the dynamic behavior offset amount to obtain the dynamic behavior offset parameter.

[0025] As a further solution of the present invention, the steps for obtaining the dynamic behavior characteristic data set are specifically as follows:

[0026] Based on the dynamic behavior offset parameter, perform time series acquisition, parse each item of the acquired time series data one by one, map each offset parameter value according to the preset interval division rule, statistically analyze the frequency and cumulative quantity within the interval, and combine the distribution density and upper and lower limit values to obtain the dynamic behavior interval distribution data;

[0027] Based on the dynamic behavior interval distribution data, identify the distribution density and total frequency within each interval, mark the data intervals according to the density ratio and screening threshold, eliminate the low-density sparse distribution intervals, and classify them in combination with the fluctuation range and interval interaction characteristics to obtain the behavior trend section data;

[0028] Based on the data of the behavioral trend section, conduct a comparative analysis of the numerical distribution characteristics in the interval, measure the transition trend of the interval and the adjacent correlation of the distribution matrix, integrate each interval characteristic parameter item by item, and combine with the interval distribution trend to generate a dynamic behavior characteristic data set.

[0029] As a further solution of the present invention, the specific steps for obtaining the interval prediction characteristics are as follows:

[0030] Based on the dynamic behavior characteristic data set, extract the usage peak and high-frequency period of each interval, analyze the inhalation frequency and the distribution of inhalation interval time within each interval, record the extreme values of the inhalation volume and usage frequency in the high-frequency paragraph, and use them as the time markers of the peak period to obtain the usage peak and high-frequency period distribution characteristics of the interval;

[0031] According to the usage peak and high-frequency period distribution characteristics of the interval, analyze the trend relationship between the frequency change value and the behavior offset amount, and use the formula:

[0032]

[0033] Perform collaborative weighting on the change ranges of the frequency change value and the behavior offset amount to generate a trend parameter of the frequency change value and the behavior offset amount;

[0034] where T is the trend parameter of the frequency change value and the behavior offset amount, F k is the frequency change value of the k-th interval, μ F is the mean value of the frequency change value, B k is the behavior offset amount of the k-th interval, μ B is the mean value of the behavior offset amount, and p is the total number of intervals;

[0035] According to the trend parameter of the frequency change value and the behavior offset amount, map the trend relationship between the frequency change value and the behavior offset amount to the high-frequency period and peak distribution, fit the trend parameter of the high-frequency period and the change characteristics of the interval peak, and conduct a quantitative analysis of the high-frequency paragraph distribution in the future time period of the interval in combination with the numerical range of the prediction characteristic parameter to obtain the interval prediction characteristics.

[0036] As a further solution of the present invention, the specific steps for obtaining the multi-interval trend prediction result are as follows:

[0037] Based on the interval prediction characteristics, divide the input data into multiple sub-intervals, analyze the change amplitude and change rate between data points, sort out and extract the data trend direction and change range, and integrate the interval trend direction and change range to obtain an interval trend characteristic set;

[0038] Based on the set of interval trend characteristics, extract and analyze the trend directions of each adjacent interval, conduct cross-analysis on the direction change points, perform hierarchical comparison based on the boundary values of the change range, embed the adjustment factor into the interval relationship model and integrate it to generate a trend characteristic association model;

[0039] Based on the trend characteristic association model, gradually deduce the trend characteristics of each interval, analyze the data change range, integrate the trend offset values in the association model, optimize the trend prediction, and generate a multi-interval trend prediction result.

[0040] As a further solution of the present invention, the specific steps for obtaining the global trend dynamic value are as follows:

[0041] Extract the offset amplitude and the behavior change range from the multi-interval trend prediction result, calculate the difference between the maximum value and the minimum value according to the absolute difference of the full-interval data, and calculate the distribution absolute difference through the change gradient of the sub-intervals for the behavior change range to generate an offset amplitude range and a behavior change range difference;

[0042] Based on the offset amplitude range and the behavior change range difference, perform a non-linear combination operation using a weighted adjustment coefficient and a dynamic smoothing factor, through the formula:

[0043] P h =k a ·|D max -D min |+k b ·|B max -B min |;

[0044] Obtain the high-frequency deviation value;

[0045] wherein, P h represents the high-frequency deviation value, k a represents the weighted adjustment coefficient of the offset amplitude, k b represents the weighted adjustment coefficient of the behavior change range, D max represents the maximum value of the offset amplitude, D min represents the minimum value of the offset amplitude, B max represents the maximum value of the behavior change range, B min represents the minimum value of the behavior change range;

[0046] Dynamically adjust the high-frequency deviation value, use non-linear difference operation to calculate the dynamic offset coefficient of the global trend, and correct the multi-interval trend prediction value through the dynamic offset coefficient to generate the global trend dynamic value.

[0047] As a further solution of the present invention, the specific steps for obtaining the prediction result of the global usage times of the electronic cigarette are as follows:

[0048] Based on the global trend dynamic value, extract the increase and decrease amplitude and change direction between adjacent numerical values, analyze the distribution of peak and low value nodes in the time series, identify the fluctuation regions, summarize the numerical trend and distribution characteristics, and generate the global trend data interval characteristic value;

[0049] Call the global trend data interval characteristic value, compare the deviation of the numerical values in the local prediction data, analyze the deviation source and the distribution of abnormal points, identify the numerical correction factor within the deviation range, adjust the deviation value to the global interval trend range, optimize the overall distribution of the adjusted data, and obtain the corrected local prediction data;

[0050] Analyze the interval range distribution of the corrected local prediction data and the global trend dynamic value, perform weighted processing on the corrected data, identify the predicted cumulative value within the global range, superimpose the cumulative value of each interval and summarize it into the complete total numerical range, and generate the prediction result of the global usage times of the electronic cigarette.

[0051] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0052] In the present invention, through data extraction and division based on the time series, segmental dynamic analysis is carried out on key data such as usage frequency, short-term fluctuation, and single interval, optimizing the accuracy of data distribution boundary and continuity analysis, accurately capturing the core law of data change, further reducing the interference of data deviation on subsequent modeling, combining in-depth analysis of segmental behavior data, focusing on the change characteristics of the difference in inhalation volume and frequency within the interval, through the extraction and correlation analysis of dynamic behavior offset parameters, effectively constructing a multi-dimensional behavior feature data set, making the prediction of interval dynamic changes closer to the actual usage scenario. In trend prediction, through trend parameterization processing of high-frequency fluctuation data, interval trend recursion and parameter optimization are realized, strengthening the adaptability of the prediction model to behavior trends. Combining trend prediction data of multiple intervals, optimize the adjustment of the fluctuation amplitude and behavior changes within the global range. Through the dynamic linkage of global and local data, further improve the stability and accuracy of the overall prediction. This processing logic based on dynamic data capture, characteristic parameter optimization, and global trend correction realizes the comprehensive quantification of multi-interval data fluctuations and behavior changes, significantly improving the accuracy and decision-making reference value of the prediction of the usage times of electronic cigarettes. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is the system flow chart of the present invention;

[0054] Figure 2 is the flow chart of the segmental usage dynamic structure in the present invention;

[0055] Figure 3 is the flow chart of the dynamic behavior offset parameter in the present invention;

[0056] Figure 4 is the flow chart of the dynamic behavior characteristic data set in the present invention;

[0057] Figure 5 is the flow chart of the interval prediction characteristic in the present invention;

[0058] Figure 6 is the flow chart of the multi - interval trend prediction result in the present invention;

[0059] Figure 7 is the flow chart of the global trend dynamic value in the present invention;

[0060] Figure 8 is the flow chart of the prediction result of the global usage times of e - cigarettes in the present invention. Specific Embodiments

[0061] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0062] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more unless otherwise specifically defined.

[0063] Please refer to Figure 1 , the e - cigarette usage times prediction system based on data analysis includes:

[0064] The time interval division module extracts the e - cigarette usage frequency, short - term fluctuation interval and single - use interval duration data based on the time series of e - cigarette usage data, analyzes the fluctuation law and change range of the time period, compares the offset values of the interval change amplitude and continuity, adjusts the interval boundary and distribution, and obtains the segmented usage dynamic structure;

[0065] The behavior dynamic analysis module captures the single - puff volume, usage frequency and behavior offset characteristics in each interval based on the segmented usage dynamic structure, analyzes the differential distribution of the frequency and inhalation interval in each interval, obtains the dynamic behavior offset parameters, judges the interval correlation of the dynamic behavior offset parameters, and generates the dynamic behavior characteristic data set;

[0066] Based on the dynamic behavior characteristic dataset, the interval trend prediction module analyzes the usage peak and high-frequency periods in the interval, performs parametric processing through the trend relationship between the frequency change value and the behavior offset, obtains the interval prediction characteristics, recursively derives the interval trend characteristics, adjusts the behavior offset of the trend characteristics, and generates the multi-interval trend prediction results;

[0067] Based on the multi-interval trend prediction results, the global prediction optimization module analyzes the offset amplitude and behavior change range in the global prediction, adjusts the high-frequency deviation of the interval trend prediction, obtains the global trend dynamic value, combines the global trend dynamic value to correct the local prediction data, and generates the prediction result of the global usage times of the electronic cigarette.

[0068] The segmented usage dynamic structure includes usage frequency distribution, short-term fluctuation interval range, single-use interval duration distribution, interval change amplitude, and interval continuity. The dynamic behavior offset parameters include single inhalation volume offset value, usage frequency difference value, inhalation interval difference value, and behavior offset characteristic value. The dynamic behavior characteristic dataset includes frequency distribution data, inhalation interval distribution data, dynamic behavior offset parameters, and interval correlation data. The interval prediction characteristics include usage peak characteristics, high-frequency period characteristics, frequency change trend parameters, and behavior offset trend parameters. The multi-interval trend prediction results include trend recursion characteristics, trend offset adjustment value, and interval correlation trend value. The global trend dynamic value includes global offset amplitude, global behavior change range, and global high-frequency deviation adjustment value. The prediction result of the global usage times of the electronic cigarette includes the corrected result of the global trend dynamic value, the corrected result of the local prediction data, and the predicted value of the global usage times.

[0069] Please refer to Figure 2 , and the specific steps for obtaining the segmented usage dynamic structure are as follows:

[0070] Based on the time series of electronic cigarette usage data, extract the usage frequency and interval duration of each time period, analyze the usage frequency change characteristics of each time period, analyze the fluctuation amplitude, screen and eliminate data, and obtain the interval division result that conforms to the fluctuation law;

[0071] First, obtain the record of each usage time in the time series data, calculate the interval duration of each usage according to the time difference, calculate the median and interquartile range of the interval duration distribution, judge whether there are outliers in the data, screen and mark the outliers through the points outside the interquartile range, eliminate the data according to the marked outliers, reconstruct the dataset, use the segmented analysis method to calculate the fluctuation amplitude of the remaining data, according to the statistical result of the fluctuation amplitude, screen the data segments with larger amplitudes and set their short-term fluctuation intervals, and classify and organize the usage frequencies through the time demarcation points of the set intervals to obtain the interval division result that conforms to the fluctuation law.

[0072] According to the interval division results that conform to the fluctuation law, calculate the extreme values of each interval, compare the boundary values of each interval with the continuity of adjacent intervals, adjust the dynamic structure of the intervals, and generate the adjusted interval distribution results;

[0073] First, obtain the maximum and minimum values of the usage frequencies based on the data points within each interval, record the results as the fluctuation boundary values of the intervals, further analyze the boundary value differences between each interval, calculate the mean and standard deviation of the boundary differences, judge the significance of the boundary differences according to the calculated difference normalization range, screen out the intervals with poor continuity by setting the threshold of the boundary value differences, for the screened intervals, use the sliding window method to recalculate their usage frequencies and fluctuation ranges, re-adjust the boundary values of their front and rear intervals by re-dividing and connecting them with adjacent intervals, and the updated boundary values will be used as the adjusted interval distribution results.

[0074] Based on the adjusted interval distribution results, conduct trend analysis on short-term and long-term fluctuations, and combine the fluctuation interval boundaries to use the formula:

[0075]

[0076] Obtain the segmented usage dynamic structure;

[0077] Among them, S represents the segmented usage dynamic structure score, n represents the number of time periods, w i represents the weight coefficient of each fluctuation interval, T end and T start respectively represent the start time and end time of the interval, ΔT i represents the change amplitude within the time period, and abs is the absolute value function;

[0078] The benefit of the formula is that by introducing the weight parameter w i adjusts the importance of each interval's fluctuation, and combines the interval duration and the absolute fluctuation amplitude to calculate the final dynamic structure score, improving the fineness of the segmented dynamic structure optimization;

[0079] In the formula, S represents the optimized usage dynamic structure score, n represents the number of time periods, calculated from the short-term fluctuation intervals divided in the time series data, w i represents the importance weight parameter of each fluctuation interval, set by analyzing the fluctuation range of the interval usage frequencies, calculated according to the statistical mean of the frequencies and fluctuation amplitudes of each interval, T end and T start respectively represent the start time and end time of the interval, obtained by recording the timestamps of each use in the time series data, ΔT i represents the change amplitude within the time period, calculated from the difference in the fluctuation frequencies of each interval;

[0080] Using time - series data, set n = 3, and the time boundaries of the intervals are T start,1 = 0, T end,1 = 10, T start,2 = 10, T end,2 = 20, T start,3 = 20, T end,3 = 30, and the weight parameter w of each interval 1 = 0.5, w 2 = 0.3, w 3 = 0.2, and the fluctuation amplitudes of each interval are ΔT 1 = 5, ΔT 2 = 8, ΔT 3 = 3;

[0081] Substitute into the formula to calculate S as follows:

[0082]

[0083] The result shows that the score of the dynamic segmentation structure is 18.33. The higher the score, the stronger the dynamics of the fluctuation interval, and it has a positive correlation with the optimization level of the segmentation result. According to this score, it can be used to further adjust the interval division or optimize the dynamic structure.

[0084] Please refer to Figure 3 , and the specific steps for obtaining the dynamic behavior offset parameter are as follows:

[0085] Based on the dynamic structure used for segmentation, extract the single - inhalation volume and usage frequency within each interval, record the single - inhalation data points within each time interval, statistically analyze the distribution of inhalation intervals within each interval, calculate the variance and mean of the inhalation - interval distribution, mark the intervals where the variance exceeds the distribution range, and obtain the interval results of the single - inhalation volume and frequency;

[0086] First, extract the boundary points of each time interval in the dynamic structure used for segmentation, record the timestamps of each inhalation behavior in the time series data, and compare them with the time ranges of each interval to divide the inhalation behavior data within each interval. For the data in each interval, count the number of inhalation points as the preliminary result of the usage frequency, and combine the recorded inhalation volume sensor data to record the flow rate value for each inhalation, obtaining the preliminary distribution of the inhalation volume within each interval. Calculate the time difference between every two inhalation points to obtain the inhalation interval time series, and perform statistics on this series to calculate the mean, variance, and interquartile range of the inhalation interval respectively, which are used to evaluate the stability of the inhalation behavior in each interval. By comparing the calculated variance value with the global variance of the inhalation behavior data in all intervals, eliminate the intervals that are significantly higher than the overall variance range. For the eliminated intervals, their inhalation volume and inhalation interval data will be marked as abnormal. Subsequently, through data reconstruction, recalculate the mean and variance of the inhalation volume and usage frequency, and update them to the valid interval data to complete the final statistics of the inhalation volume and frequency, and obtain the interval results of the single inhalation volume and frequency.

[0087] Based on the interval results of the single inhalation volume and frequency, compare the relationship between the inhalation frequency and inhalation interval within each interval, and use the formula:

[0088]

[0089] Calculate the dynamic behavior offset value, weight the inhalation interval, and reflect the degree of behavior offset to generate the distribution result of the interval offset value;

[0090] where P is the dynamic behavior offset value, A j is the inhalation volume at the j-th inhalation point, I j is the inhalation interval time at the j-th inhalation point, α is the inhalation time adjustment parameter, β is the inhalation interval weight factor, and m is the number of inhalation points within the interval;

[0091] The benefit of the formula is that by introducing the adjustment parameter α and the weight factor β, dynamically adjust the contribution of the inhalation interval to the offset value, and perform weighted calculation in combination with the relationship between the inhalation volume and the interval, so as to more accurately quantify the differential distribution of the inhalation behavior and avoid the deviation caused by too short or too long intervals to the results;

[0092] In the formula, P is the dynamic behavior offset value, which is calculated by weighting the relationship between the inhalation volume and the interval at each inhalation point. A j is the inhalation volume at the j-th inhalation point, and this data is recorded by the built-in sensor of the inhaler, representing the total airflow volume for each inhalation. I j$I_j$ is the inhalation interval time for the $j$-th inhalation point, calculated from the timestamps of each inhalation behavior; the adjustment parameter $\alpha$ is used to smooth the inhalation interval time to avoid unreasonable fluctuations in the results due to too short an interval time. Its value is dynamically set by the median and interquartile range of the inhalation interval distribution. The weight factor $\beta$ is used to correct the influence of the interval time on the offset value, calculated from the interval statistical results of historical behavior data. $m$ is the number of inhalation points within the interval, obtained by counting the inhalation behavior records within the time interval;

[0093] Substitute the monitoring data for actual calculation: Assume that within a certain interval, $m = 4$, the inhalation amounts $A$ 1 $= 3.2$, $A$ 2

[0094] $= 4.1$, $A$ 3 $= 2.8$, $A$ 4 $= 3.5$, the inhalation interval times $I$ 1 $= 2.1$, $I$ 2 $= 3.4$, $I$ 3 $= 1.8$, $I$ 4 $= 2.5$, the adjustment parameter $\alpha = 0.5$, the interval weight factor $\beta = 0.3$;

[0095] Substitute into the formula as follows:

[0096]

[0097] The result shows that the dynamic behavior offset value $P = 0.5745$ is used to represent the behavior offset degree of this interval. Through subsequent analysis, it can be compared with the offset values of other intervals to evaluate the inhalation behavior change rules in different time periods, and at the same time, the behavior data distribution can be further optimized.

[0098] According to the distribution results of the interval offset values, screen the intervals where the offset values deviate from the mean, adjust the inhalation amount and frequency distribution of the abnormal intervals, and integrate them using the dynamic behavior offset amount to obtain the dynamic behavior offset parameter;

[0099] First, perform statistics on the offset value list, analyze the overall situation of the distribution by calculating its mean and standard deviation, standardize the offset value of each interval, compare its standardized result with the set significant offset threshold, screen the intervals where the offset value is significantly higher or lower than the distribution mean range. For the screened abnormal intervals, record their inhalation behavior data, and readjust their inhalation amount and inhalation frequency distribution in a segmented reorganization manner. By recalculating the ratio of their inhalation interval time to inhalation frequency, optimize the behavior data of the abnormal intervals into a distribution state closer to the interval. The adjusted interval data is re-integrated into the dynamic behavior offset parameter for the final behavior dynamic analysis and data prediction analysis.

[0100] Please refer to Figure 4, the steps for obtaining the dynamic behavior characteristic dataset are specifically as follows:

[0101] Based on the dynamic behavior offset parameters, perform time series collection, parse the collected time series data item by item, map each offset parameter value according to the preset interval division rule, count the frequency and cumulative quantity within the interval, and combine the distribution density and upper and lower limit values to obtain the dynamic behavior interval distribution data;

[0102] First, record the time points of the number of times of e-cigarette use by setting the collection frequency and data interface, use the use frequency at each time point as the dynamic behavior offset parameter, parse the collected time series data item by item, determine the time stamp of each data point and the corresponding behavior offset value, and map each offset parameter value into the corresponding time interval according to the preset time interval (for example, each day is divided into four time periods). Use the counting function to count the use frequency within each time period, and at the same time accumulate the total number of times of e-cigarette use. Apply statistical analysis methods to the data, such as frequency distribution diagrams and cumulative density functions, calculate the use density within each time period, and combine the upper and lower use limit values of each time period to evaluate and determine the dynamic behavior interval distribution data of e-cigarette use, and finally output this dataset to reflect the user's usage behavior pattern.

[0103] Based on the dynamic behavior interval distribution data, identify the distribution density and total frequency within each interval, mark the data intervals according to the density ratio and the screening threshold, eliminate the low-density sparse distribution intervals, and classify them in combination with the fluctuation range and interval interaction characteristics to obtain the behavior trend section data;

[0104] First, comprehensively evaluate the use frequency and frequency within each time period, identify the use density and total use frequency within each time period, mark and eliminate the sparse distribution intervals that are lower than this threshold by setting a specific density ratio threshold (for example, each time period should contain at least 5% of the total number of uses). Considering the fluctuation range of e-cigarette use, that is, the change range of user use in different time periods, and the interaction characteristics between intervals, such as the comparison of the use amount in the morning and at night, classify and mark the significant use intervals by analyzing these data characteristics. This process involves performing cluster analysis on the data to identify time periods with similar usage patterns, and finally obtaining the behavior trend section data that can reflect the user's main activity pattern.

[0105] Based on the behavior trend section data, conduct a comparative analysis of the interval numerical distribution characteristics, measure the interval transition trend and the adjacent correlation of the distribution matrix, integrate each interval characteristic parameter item by item, and combine the interval distribution trend to generate the dynamic behavior characteristic dataset;

[0106] Use statistical methods such as box plots and violin plots to conduct comparative analysis on usage data in different time periods, thereby measuring the transition trends in each interval. Analyze the correlation between adjacent time periods using a distribution matrix, and examine how an increase in usage in one time period affects the usage frequency in the next time period. Conduct integrated analysis on the characteristic parameters within each time period. Considering the periodicity and randomness of e-cigarette usage, combine statistical models such as autoregressive moving average models to generate a dataset of dynamic behavior characteristics. This dataset will comprehensively consider the stability and volatility of user behavior within the time period, providing a scientific basis for further prediction of usage times.

[0107] Please refer to Figure 5 , and the steps for obtaining the interval prediction characteristics are specifically as follows:

[0108] Based on the dataset of dynamic behavior characteristics, extract the usage peak and high-frequency time periods for each interval, analyze the distribution of inhalation frequencies and inhalation intervals within each interval, record the extreme values of inhalation amounts and usage frequencies in the high-frequency segments, and use them as time markers for the peak time periods to obtain the distribution characteristics of usage peaks and high-frequency time periods for the intervals.

[0109] First, analyze the characteristics of dynamic behavior data within each time interval, extract the timestamps and inhalation amount data for each inhalation behavior, calculate the distribution of inhalation amounts and usage frequencies within the interval, obtain the maximum inhalation amount and maximum frequency, and use the time point corresponding to the maximum value as the high-frequency time period within the interval. Then, conduct statistical analysis on the dynamic data distribution of the interval, calculate the median and interquartile range of the inhalation amount and frequency distributions within each interval, identify abnormal data points within the distribution range, remove the abnormal points and recalculate the statistical distribution within the interval. Based on the cleaned data distribution, determine the dynamic characteristics of the high-frequency time period and generate the peak data for each interval. By jointly statistically analyzing the dynamic data of the high-frequency time period and the cleaned peak data, finally obtain the distribution characteristics of usage peaks and high-frequency time periods for the intervals.

[0110] According to the distribution characteristics of usage peaks and high-frequency time periods for the intervals, analyze the trend relationship between the frequency change value and the behavior offset amount, and use the formula:

[0111]

[0112] Perform collaborative weighting on the change ranges of the frequency change value and the behavior offset amount to generate a trend parameter for the frequency change value and the behavior offset amount;

[0113] where T is the trend parameter for the frequency change value and the behavior offset amount, F k is the frequency change value for the k-th interval, μ F is the mean value of the frequency change values, B k is the behavior offset amount for the k-th interval, μ Bis the mean of the behavior offset, and p is the total number of intervals;

[0114] The advantage of the formula is that by introducing the mean of the frequency change value and the behavior offset and the collaborative weighting processing method, the formula can more accurately reflect the dynamic relationship between the interval frequency change value and the behavior offset, avoiding the incompleteness of the description of the change correlation between the two by simple weighting. At the same time, the square root is used to dynamically correct the final trend parameter, making the calculation result smoother;

[0115] In the formula, T is the trend parameter of the frequency change value and the behavior offset, used to reflect the correlation between the two changes, and F k is the frequency change value of the k-th interval, obtained from the mean and range of the time difference of the inhalation frequency within the calculation interval, and μ F is the mean of the frequency change values, obtained by calculating the average level of the frequency change values of all intervals, and B k is the behavior offset of the k-th interval, obtained by calculating the dynamic behavior offset parameter, and μ B is the mean of the behavior offsets, obtained by calculating the mean of the behavior offsets of all intervals, and p is the total number of intervals, obtained by counting the number of all intervals;

[0116] Actual operation example: Assume that the total number of intervals p = 3, the frequency change values are F 1 = 1.2, F 2 = 1.8, F 3 = 2.4, the mean of the frequency change μ F = 1.8, the behavior offset is B 1 = 0.6, B 2 = 1.0, B 3 = 1.4, the mean of the behavior offset μ B = 1.0;

[0117] Substitute into the formula:

[0118] Calculate the numerator part:

[0119] (F 1 - μ F )·(B 1 - μ B ) = (1.2 - 1.8)·(0.6 - 1.0) = (-0.6)·(-0.4) = 0.24;

[0120] (F 2 - μ F )·(B 2 - μ B ) = (1.8 - 1.8)·(1.0 - 1.0) = 0·0 = 0;

[0121] (F3 -μ F )·(B 3 -μ B )=(2.4 - 1.8)·(1.4 - 1.0)=0.6·0.4=0.24;

[0122] Sum of the numerators:

[0123] Calculate the denominator part:

[0124] (F 1 -μ F ) 2 =(1.2 - 1.8) 2 =(-0.6) 2 =0.36;

[0125] (F 2 -μ F ) 2 =(1.8 - 1.8) 2 =0 2 =0;

[0126] (F 3 -μ F ) 2 =(2.4 - 1.8) 2 =(0.6) 2 =0.36;

[0127] Sum of the denominators:

[0128] Final calculation:

[0129] The result shows that the trend parameter of the frequency change value and the behavior offset is 0.816. This result can be used for subsequent fitting and optimization in combination with the high-frequency period distribution to further quantify the interval prediction characteristics.

[0130] According to the trend parameter of the frequency change value and the behavior offset, map the trend relationship between the frequency change value and the behavior offset to the high-frequency period and the peak distribution, fit the trend parameter of the high-frequency period and the change characteristics of the interval peak, and combine the numerical range of the prediction characteristic parameters to quantitatively analyze the high-frequency paragraph distribution in the future time period of the interval to obtain the interval prediction characteristics;

[0131] First, perform trend fitting on the high-frequency time period and peak distribution data, calculate the frequency change trend of each high-frequency time period and its distribution change within the interval. Then, quantify the peak dynamic characteristics of each interval, extract the change ratio between the frequency change and the peak characteristics, and combine the change ratio with the trend parameters to form a trend model for fitting the high-frequency distribution of the interval. By calculating the fitting error of the trend model, the high-frequency change characteristics of each interval, and finally the result optimized by the combined trend relationship, the interval prediction characteristics are obtained, which are used as the core characteristic parameters for subsequent distribution prediction.

[0132] Please refer to Figure 6 , and the steps for obtaining the multi-interval trend prediction result are specifically as follows:

[0133] Based on the interval prediction characteristics, divide the input data into multiple sub-intervals, analyze the change amplitude and change rate between data points, sort out and extract the trend direction and change range of the data, and integrate the interval trend direction and change range to obtain the interval trend characteristic set;

[0134] Divide the input e-cigarette usage data into multiple sub-intervals according to the preset time range. The usage frequency data within each sub-interval is regarded as an independent analysis unit. Respectively extract the change amplitude of the data points within the interval and the change rate of adjacent data points. Through statistical analysis of the fluctuation values of the data within each sub-interval, sort out and extract the trend direction of each interval, such as growth, decline or stable state. Compare and analyze the maximum and minimum values within the interval to calculate the data change range. Combine the relationship between the change range and the change rate, and further summarize the usage pattern characteristics of each interval. Integrate the trend direction and change range of each interval, and output an interval trend characteristic set with a clear trend characteristic identifier. This set can reflect the characteristics of e-cigarette usage behavior in different time intervals, such as the trend direction of the significant increase in the usage frequency of users in certain time periods and its fluctuation range, providing basic data support for subsequent prediction and modeling.

[0135] Based on the interval trend characteristic set, extract and analyze the trend direction of each adjacent interval, conduct cross-analysis on the direction change points, conduct hierarchical comparison according to the boundary values of the change range, adjust the factor to be embedded in the interval relationship model and integrate it to generate the trend characteristic association model;

[0136] By identifying the change points of the trend direction, such as the inflection point from growth to decline, combining with the change range in the trend characteristic set, accurately marking the change points, and based on the boundary values of the change range, stratifying the interval trend into high-frequency, medium-frequency, and low-frequency levels, and making a horizontal comparison of the trend changes at each level, analyzing the relationship between the trend levels. To enable the model to more accurately reflect the correlation between different intervals, an adjustment factor is embedded to assign dynamic weight values to the trends in different intervals. The setting of the adjustment factor is based on the significance of the trend within the interval and its influence on the overall trend. For example, a higher weight is assigned to the high-frequency level trend to reflect its dominant role in the overall model. Integrate the trend direction, change range, and adjustment factor after hierarchical comparison to generate a trend characteristic correlation model that can reflect the correlation between intervals, so as to depict the dynamic correlation characteristics of e-cigarette usage data among different time periods.

[0137] Based on the trend characteristic correlation model, gradually deduce the trend characteristics of each interval, analyze the data change range, integrate the trend offset values in the correlation model, optimize the trend prediction, and generate the multi-interval trend prediction results.

[0138] Perform step-by-step recursive processing on the trend characteristics of each interval. While analyzing the data change range of each interval, calculate the dynamic adjustment amplitude of each interval in combination with the trend offset value in the model. By integrating the trend offset values and dynamic adjustment amplitudes of each interval, gradually optimize the predicted value, so that the model can more accurately reflect the change characteristics of each time period. During the recursive process, the trend characteristic correlation model will layer by layer integrate the change data of each interval to obtain comprehensive prediction data covering the change laws of each interval, and finally generate multi-interval trend prediction results, clarifying the future change direction, amplitude, and total amount trend of the number of e-cigarette uses in each time interval, providing refined data support for the short-term and long-term prediction of e-cigarette usage behavior, and ensuring the accuracy and practicality of the prediction results.

[0139] Please refer to Figure 7 , and the specific steps for obtaining the global trend dynamic value are as follows:

[0140] Extract the offset amplitude and behavior change range from the multi-interval trend prediction results, calculate the difference between the maximum value and the minimum value according to the absolute difference of the full-interval data, and calculate the distribution absolute difference through the change gradient of each interval for the behavior change range to generate the offset amplitude range and the difference of the behavior change range.

[0141] First, for the calculation of the offset amplitude, using the distribution range of data points in each interval as the initial basis, by calculating the difference between the maximum value and the minimum value in each interval to obtain the offset amplitude difference of that interval, and then performing an absolute difference calculation on the offset amplitude differences of all intervals. The specific process includes re - sorting the full - interval data according to the distribution range, eliminating the intervals with significant distribution differences, and thus obtaining the overall absolute difference of the offset amplitude range in all intervals. At the same time, for the behavior change range, calculate its absolute difference according to the change gradient of each interval. First, calculate the difference between the behavior change values of two consecutive time points within each interval to obtain the initial gradient of each interval, and then obtain the distribution difference of the behavior change in that interval by accumulating the gradient values. Subsequently, perform a total difference operation on the distribution differences of the behavior change range across all intervals. Combining the offset amplitude and the absolute difference of behavior change in the multi - interval trend prediction, generate the offset amplitude range and the difference of the behavior change range respectively.

[0142] Based on the offset amplitude range and the difference of the behavior change range, perform a non - linear combination operation using a weighted adjustment coefficient and a dynamic smoothing factor. Through the formula:

[0143] P h =k a ·|D max -D min |+k b ·|B max -B min |;

[0144] Obtain the high - frequency deviation value;

[0145] Among them, P h represents the high - frequency deviation value, k a represents the weighted adjustment coefficient of the offset amplitude, k b represents the weighted adjustment coefficient of the behavior change range, D max represents the maximum value of the offset amplitude, D min represents the minimum value of the offset amplitude, B max represents the maximum value of the behavior change range, B min represents the minimum value of the behavior change range;

[0146] The advantage of the formula is that by assigning different weight parameters to the range difference of the offset amplitude and the difference of the behavior change range, it can more accurately reflect the high - frequency deviation in the multi - interval prediction result, thereby improving the dynamic capture ability of the impact of high - frequency fluctuations;

[0147] Suppose the calculation result of the offset amplitude range of a certain interval is D max =60 and D min =20, and the calculation result of the behavior change range is B max =50 and B min= 10, set the offset amplitude weight parameter k a = 0.6, the behavior change weight parameter k b = 0.4, then the calculation of the high-frequency deviation value is as follows:

[0148] The offset amplitude range difference is calculated as: |D max - D min | = |60 - 20| = 40;

[0149] The behavior change range difference is calculated as: |B max - B min | = |50 - 10| = 40;

[0150] The high-frequency deviation value is calculated as:

[0151] P h = 0.6·40 + 0.4·40;

[0152] P h = 24 + 16 = 40;

[0153] This result shows that the high-frequency deviation value is 40, indicating significant offset and change trends in the multi-interval prediction results, showing that the high-frequency deviation value is a comprehensive characterization of the dynamic characteristics of multi-interval fluctuations, and this value can be used to adjust the fluctuation impact in the global trend prediction in subsequent calculations.

[0154] Dynamically adjust the high-frequency deviation value, use non-linear difference operation, calculate the dynamic offset coefficient of the global trend, and correct the multi-interval trend prediction value through the dynamic offset coefficient to generate the global trend dynamic value;

[0155] First, perform a non-linear difference operation on the result of the high-frequency deviation value and the fluctuation impact coefficient in the multi-interval prediction, including calculating the change range of the fluctuation impact coefficient by analyzing the fluctuation changes of continuous data points in the full-interval prediction result, dynamically correcting the difference using non-linear smoothing processing, and then dynamically adjusting the offset coefficient of the global trend. The specific adjustment includes comparing the full-interval prediction data points with the overall trend, dynamically calculating the weight combination of the trend offset value and the high-frequency deviation value, and finally adjusting according to the comprehensive weight of the offset value and the high-frequency deviation to generate the global trend dynamic value.

[0156] Please refer to Figure 8 , the specific steps for obtaining the prediction result of the global usage times of e-cigarettes are as follows:

[0157] Based on the global trend dynamic value, extract the increase and decrease amplitude and change direction between adjacent values, analyze the distribution of peak and low-value nodes in the time series, identify the fluctuation regions, summarize the numerical trend and distribution characteristics, and generate the global trend data interval characteristic value;

[0158] Extract data point by point from time series data, calculate the increase or decrease amplitude and change direction between each adjacent data point, analyze the distribution characteristics of the increase or decrease amplitude item by item, refine the main direction of trend change, locate the peak and low value nodes in the time series one by one, identify the fluctuation area by setting a clear judgment condition, for example, mark the interval where the change amplitude of adjacent data points exceeds a certain threshold as the fluctuation area, and then subdivide the fluctuation range and interval of the entire time series. Combine the relative position relationship between the peak node and the low value node to summarize the change pattern of the numerical trend, such as discovering whether the trend shows periodic or progressive changes. Integrate the trend characteristics and the distribution of the fluctuation area to generate the global trend data interval characteristic value, mark the main characteristics within each time interval, and provide a complete dynamic trend reference basis for the global prediction of e-cigarette usage behavior.

[0159] Call the global trend data interval characteristic value, compare the deviation of the values in the local prediction data, analyze the source of the deviation and the distribution of abnormal points, identify the numerical correction factor within the deviation range, adjust the deviation value to the global interval trend range, optimize the overall distribution of the adjusted data, and obtain the corrected local prediction data;

[0160] Compare the deviation range between the local prediction data and the global trend. By calculating the difference value between the local data and the global trend, analyze the source of the deviation item by item, and judge whether it is caused by the fluctuation characteristics or the distribution of abnormal points. Precisely correct the data points within the deviation range. By calculating the correction factor, readjust the values in the local data that exceed the global interval to within the global interval range. The correction factor is calculated based on the proportional relationship between the dynamic value of the global trend and the local deviation. For example, allocate the adjustment amplitude proportionally for each deviation value to ensure that the adjustment of the local prediction data can match the global trend range. After completion of the correction, redistribute the optimized adjusted data, and verify the effect of the data adjustment by calculating characteristics such as the mean and variance of the distribution. Finally, obtain the corrected local prediction data to make it consistent and coordinated with the global trend data.

[0161] Analyze the interval range distribution of the corrected local prediction data and the global trend dynamic value, perform weighted processing on the corrected data, identify the predicted cumulative value within the global range, superimpose the cumulative value of each interval and summarize it into a complete total value range, and generate the global prediction result of the e-cigarette usage times.

[0162] The corrected data is weighted by partitioning, and weights are assigned to the correction values in each interval. The size of the weights is set according to the contribution ratio of the data in the interval to the global trend. For example, intervals with large fluctuations can be assigned lower weights, while intervals with stable growth or decline are assigned higher weights, so as to optimize the calculation process of the cumulative value. The weighted cumulative values of each interval are superimposed, and the complete total value range within the global scope is summarized and output. By synthesizing the weighted cumulative values of all intervals, the predicted result of the global usage times of the electronic cigarette is generated. This result can reflect the change in the usage frequency of users in different time intervals, and at the same time provide the global trend range of future usage times, providing data support and decision-making basis for further optimizing the prediction scheme.

[0163] The above is only the preferred embodiment of the present invention, and does not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention still belong to the protection scope of the technical solution of the present invention.

Claims

1. An electronic cigarette usage frequency prediction system based on data analysis, characterized in that: The system comprises: The time interval division module extracts the frequency of e-cigarette use, short-term fluctuation interval and single interval duration data based on the time series of e-cigarette use data, analyzes the fluctuation pattern and change range of the time period, compares the offset values ​​of the interval change amplitude and continuity, adjusts the interval boundary and distribution, and obtains the segmented use dynamic structure; The behavior dynamic analysis module captures the single inhalation volume, usage frequency and behavior deviation characteristics in each interval based on the segmented usage dynamic structure, analyzes the difference distribution of frequency and inhalation interval in each interval, obtains dynamic behavior deviation parameters, performs interval correlation judgment on the dynamic behavior deviation parameters, and generates a dynamic behavior characteristic data set; The interval trend prediction module analyzes the usage peak and high-frequency period in the interval based on the dynamic behavior characteristic data set, performs parameter processing through the trend relationship between the frequency change value and the behavior offset, obtains the interval prediction characteristics, recursively deduces the interval trend characteristics, adjusts the behavior offset of the trend characteristics, and generates multi-interval trend prediction results; The global prediction optimization module analyzes the offset amplitude and behavior change range in the global prediction based on the multi-interval trend prediction results, adjusts the high-frequency deviation of the interval trend prediction, obtains the global trend dynamic value, and corrects the local prediction data in combination with the global trend dynamic value to generate the global electronic cigarette usage frequency prediction result.

2. The electronic cigarette usage frequency prediction system based on data analysis according to claim 1, characterized in that: The steps for obtaining the segmented dynamic structure are as follows: Based on the time series of e-cigarette usage data, the usage frequency and interval duration of each time period are extracted, the usage frequency change characteristics of each time period are analyzed, the fluctuation amplitude is analyzed, and the data is screened and eliminated to obtain the interval division results that conform to the fluctuation law; According to the interval division result that conforms to the fluctuation law, the extreme value of each interval is calculated, the continuity of the boundary value of each interval and the adjacent interval is compared, the dynamic structure of the interval is adjusted, and the adjusted interval distribution result is generated; Based on the adjusted interval distribution results, trend analysis of short-term and long-term fluctuations is performed, combined with the fluctuation interval boundaries, using the formula: Get segmentation using dynamic structure; Among them, S represents the dynamic structure score used in the segment, n represents the number of time periods, and w i Represents the weight coefficient of each fluctuation range, T end and T start Respectively represent the start time and end time of the interval, ΔT i Represents the magnitude of change within a time period, and abs is the absolute value function.

3. The electronic cigarette usage frequency prediction system based on data analysis according to claim 2, characterized in that: The steps for obtaining the dynamic behavior offset parameter are specifically as follows: Based on the segmented dynamic structure, extract the single inhalation volume and usage frequency in each interval, record the single inhalation data points in each time interval, count the inhalation interval time distribution in each interval, calculate the variance and mean of the inhalation interval distribution, mark the interval where the variance exceeds the distribution range, and obtain the interval results of the single inhalation volume and frequency; Based on the interval results of the single inhalation volume and frequency, the relationship between the inhalation frequency and the inhalation interval in each interval is compared, and the formula is used: Calculate the dynamic behavior deviation value, weight the inhalation interval and reflect the degree of behavior deviation, and generate the interval deviation value distribution result; Among them, P is the dynamic behavior offset value, A j is the suction volume of the jth suction point, I j is the inhalation interval time of the jth inhalation point, α is the inhalation time adjustment parameter, β is the inhalation interval weight factor, and m is the number of inhalation points in the interval; According to the interval offset value distribution result, the interval where the offset value deviates from the mean is screened, the inhalation volume and frequency distribution of the abnormal interval are adjusted, and the dynamic behavior offset is used for integration to obtain the dynamic behavior offset parameter.

4. The electronic cigarette usage frequency prediction system based on data analysis according to claim 3, characterized in that: The steps for acquiring the dynamic behavior characteristic data set are specifically as follows: Based on the dynamic behavior offset parameter, time series collection is performed, the collected time series data is analyzed item by item, each offset parameter value is mapped according to the preset interval division rule, the frequency and cumulative quantity in the interval are counted, and the dynamic behavior interval distribution data is obtained by combining the distribution density and the upper and lower limits; Based on the dynamic behavior interval distribution data, identify the distribution density and total frequency in each interval, mark the data interval according to the density ratio and screening threshold, eliminate the low-density sparse distribution interval, combine the fluctuation range and interval interaction characteristics to classify, and obtain the behavior trend segment data; Based on the behavioral trend segment data, the interval numerical distribution characteristics are compared and analyzed, the interval transition trend and the adjacent correlation of the distribution matrix are calculated, each interval characteristic parameter is integrated item by item, and combined with the interval distribution trend to generate a dynamic behavioral characteristic data set.

5. The electronic cigarette usage frequency prediction system based on data analysis according to claim 4, characterized in that: The steps for obtaining the interval prediction characteristics are specifically as follows: Based on the dynamic behavior characteristic data set, the usage peak and high-frequency period of each interval are extracted, the inhalation frequency and inhalation interval time distribution in each interval are analyzed, the extreme values ​​of the inhalation volume and usage frequency of the high-frequency section are recorded and used as the time mark of the peak period, and the usage peak and high-frequency period distribution characteristics of the interval are obtained; According to the usage peak and high-frequency time distribution characteristics of the interval, the trend relationship between the frequency change value and the behavior offset is analyzed, and the formula is used: Co-weighting the change ranges of the frequency change value and the behavior offset to generate trend parameters of the frequency change value and the behavior offset; Among them, T is the trend parameter of frequency change value and behavior offset, F k is the frequency change value of the kth interval, μ F is the mean of the frequency change values, B k is the behavior offset of the kth interval, μ B is the mean of the behavior deviation, p is the total number of intervals; According to the trend parameters of the frequency change value and the behavior offset, the trend relationship of the frequency change value and the behavior offset is mapped to the high-frequency time period and the peak distribution, the trend parameters of the high-frequency time period and the change characteristics of the interval peak are fitted, and the high-frequency segment distribution in the future time period of the interval is quantitatively analyzed in combination with the numerical range of the prediction characteristic parameters to obtain the interval prediction characteristics.

6. The electronic cigarette usage frequency prediction system based on data analysis according to claim 5, characterized in that: The steps for obtaining the multi-interval trend prediction results are specifically as follows: Based on the interval prediction characteristics, the input data is divided into multiple sub-intervals, the change amplitude and change rate between data points are analyzed, the data trend direction and change range are sorted and extracted, and the interval trend direction and change range are integrated to obtain an interval trend characteristic set; Based on the interval trend characteristic set, the trend direction of each adjacent interval is extracted and analyzed, the direction change points are cross-analyzed, hierarchical comparison is performed according to the boundary values ​​of the change range, the adjustment factors are embedded in the interval relationship model and integrated to generate a trend characteristic association model; Based on the trend characteristic association model, the trend characteristics of each interval are gradually recursively deduced, the data variation range is analyzed, the trend offset value in the association model is integrated, the trend prediction is optimized, and a multi-interval trend prediction result is generated.

7. The electronic cigarette usage frequency prediction system based on data analysis according to claim 6, characterized in that: The steps for obtaining the global trend dynamic value are specifically as follows: Extract data on the offset amplitude and behavior change range in the multi-interval trend prediction results, calculate the difference between the maximum and minimum values ​​based on the absolute difference of the full interval data, calculate the distribution absolute difference by the change gradient between the sub-intervals for the behavior change range, and generate the offset amplitude range and the behavior change range difference; Based on the difference between the offset amplitude range and the behavior change range, a weighted adjustment coefficient and a dynamic smoothing factor are used to perform a nonlinear combination operation, and the formula is: P h =k a ·|D max -D min |+k b ·|B max -B min |; Get the high frequency deviation value; Among them, P h Represents the high frequency deviation value, k a Represents the weighted adjustment factor of the offset amplitude, k b The weighted adjustment factor representing the range of behavioral variation, D max Represents the maximum value of the offset amplitude, D min Represents the minimum value of the offset amplitude, B max represents the maximum value of the behavior change range, B min represents the minimum value of the range of behavioral changes; The high-frequency deviation value is dynamically adjusted, and a nonlinear difference operation is used to calculate the dynamic offset coefficient of the global trend. The multi-interval trend prediction value is corrected by the dynamic offset coefficient to generate a global trend dynamic value.

8. The electronic cigarette usage frequency prediction system based on data analysis according to claim 7, characterized in that: The steps for obtaining the prediction result of the global number of times the electronic cigarette is used are specifically as follows: Based on the global trend dynamic value, the increase and decrease amplitude and change direction between adjacent values ​​are extracted, the distribution of peak and low value nodes in the time series is analyzed, the fluctuation area is identified, the numerical trend and distribution characteristics are summarized, and the global trend data interval characteristic value is generated; Calling the global trend data interval characteristic value, comparing the deviation of the value in the local prediction data, analyzing the deviation source and the distribution of abnormal points, identifying the value correction factor within the deviation range, adjusting the deviation value to within the global interval trend range, optimizing the overall distribution of the adjusted data, and obtaining the corrected local prediction data; Analyze the interval range distribution of the revised local prediction data and the global trend dynamic value, perform weighted processing on the revised data, identify the predicted cumulative value in the global range, superimpose each interval cumulative value and summarize it into a complete total value range, and generate a global e-cigarette usage frequency prediction result.

Images