Intelligent cloud smoking cessation control system based on behavioral science model and method thereof

Abstract

The present application relates to the technical field of health management, in particular to an intelligent cloud smoking control system and method based on a behavior science model, which comprises a behavior data generation module, a coupling relationship extraction module, an abnormal action screening module, a relapse signal identification module and an intervention behavior generation module. In the present application, the behavior amplitude, continuous intensity and action convergence information in the whole process of the smoking action are collected to construct a sequence of action fragments with sequential characteristics, the coupling mode between high-frequency trigger scenes is established in combination with the label connection relationship in the daily behavior intensive period, the non-stable behavior trajectory containing the shaking amplitude and posture delay characteristics is identified on this basis, the repeated action sequence with abnormal holding time and smoking interval is further extracted, the behavior combination with relapse tendency is identified therefrom, and various forms of intervention means are dynamically matched according to the feature type and frequency of the repeated mode, so as to improve the pertinence, timeliness and behavior correction efficiency of the prompt content.

Description

Technical Field

[0001] This invention relates to the field of health management technology, and in particular to an intelligent cloud-based smoking cessation control system and method based on a behavioral science model. Background Technology

[0002] The field of health management technology mainly involves core aspects such as continuous monitoring of individual health status, behavior recording, risk identification, health pattern modeling, and health behavior intervention. Its technical system typically includes methods for collecting relevant physiological or behavioral data, methods for quantitatively representing health-related habits, techniques for identifying behavioral triggers, methods for analyzing trends in health behavior, and strategies for developing health guidance. This field emphasizes the systematic management of human health behavior using scientific models. Through correlation analysis of behavioral, environmental, and physiological factors, it provides users with actionable health guidance plans, encompassing fundamental research directions such as behavioral data monitoring technology, habit pattern extraction technology, mathematical modeling methods for health behavior, and methods for generating personalized intervention strategies based on user behavioral characteristics. Traditional intelligent cloud-based smoking cessation control systems refer to smoking cessation assistance systems developed for smoking behavior management. They typically use recording the time, quantity, and context of smoking as core data sources, controlling smoking cessation by setting fixed daily allowable smoking limits or reminding users to reduce smoking frequency according to a preset schedule. Traditionally, a built-in opening / closing detection structure in the cigarette pack is used to determine if a user has taken a cigarette. A counting structure records changes in the number of remaining cigarettes, and reminders are sent to the user at a preset schedule. Alternatively, users can manually record smoking times and emotions, and then infer smoking habits based on these records. Some traditional methods roughly calculate the average number of cigarettes smoked by counting daily smoking frequency and then provide a linearly decreasing smoking target based on a fixed ratio. Other methods use user-marked high-frequency smoking periods as trigger points, repeatedly pushing reminders during those periods as intervention. These traditional smart cloud-based smoking cessation control systems primarily rely on simple counting structures, time recording methods, and fixed-rule reminders to manage smoking cessation behavior.

[0003] Traditional intelligent cloud-based smoking cessation control systems rely on changes in the structure of cigarette boxes to judge behavior. Behavior recognition is limited to opening and closing records and counting, lacking structured analysis of the cigarette-taking process. The granularity of behavior recognition is coarse and cannot reveal the changing characteristics of the action itself. The system records behavior using static time settings and manual input, making it difficult to establish effective correlations between complex action sequences and sudden behaviors. It cannot accurately identify abnormal behaviors based on actual behavioral fluctuations. Intervention methods are based on pushing prompts at fixed frequencies, lacking the perception and differentiated feedback mechanism for repetitive behavior patterns. This can easily lead to excessive prompt interference or intervention failure, affecting the user's acceptance of system prompts and the sustainability of intervention results. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide an intelligent cloud-based smoking cessation control system and method based on a behavioral science model.

[0005] On the one hand, an intelligent cloud-based smoking cessation control system based on a behavioral science model is provided, which includes:

[0006] The behavior data generation module collects signal data from the accelerometer, infrared components and Hall device, divides the time axis segment from opening to closing the cover, extracts three behavior values: start amplitude, maintenance intensity and end convergence, and constructs a sequence of behavior segments of the smoke-taking action.

[0007] The coupling relationship extraction module extracts the action timestamps from the cigarette-taking action behavior segment sequence, locates the concentrated distribution points within a fixed time period each day, counts the number of connections between behavior scene tags within the time period, and constructs a trigger behavior coupling structure group based on tag combinations whose frequency exceeds a preset threshold.

[0008] The abnormal action filtering module extracts the jitter amplitude and attitude stop delay values ​​in the trigger behavior coupling structure group, filters point pairs with changing amplitude and close intervals, removes coordinate overlap points in the jitter path, and filters the set of unstable smoke-picking actions.

[0009] The relapse signal recognition module extracts the holding duration and cigarette-taking interval from the unstable cigarette-taking action set, and determines whether they are both below a preset stable threshold. If the condition is met, a high-susceptibility relapse behavior sequence is obtained.

[0010] The intervention behavior generation module extracts the initiation amplitude value and holding duration from the relapse high-susceptibility behavior sequence, counts the repetition frequency of the coupling mode, matches the mode type with a frequency higher than the threshold, specifies vibration reminder, sound blocking and image prompts, and generates a behavior reminder execution content group.

[0011] As a further aspect of the present invention, the cigarette-taking action behavior segment sequence includes initiation amplitude encoding, maintenance intensity encoding, and termination convergence encoding; the triggering behavior coupling structure group includes high-frequency tag combinations, number of connection relationships, and concentrated distribution time periods; the unstable cigarette-taking action set includes amplitude mutation point pairs, posture delay data, and deduplicated trajectory points; the relapse high-susceptibility behavior sequence includes low-value holding duration, short-interval cigarette-taking time, and repetitive behavior identifiers; and the behavior reminder execution content group includes vibration reminder signals, sound blocking commands, and image prompts.

[0012] As a further aspect of the present invention, the tag combination based on frequency exceeding a preset threshold refers to the combination of tag pairs whose number of behavioral tag connections exceeds a preset threshold within a specific time period, thereby constructing a behavioral coupling structure.

[0013] As a further aspect of the present invention, the determination of whether it is simultaneously below a preset stability threshold means that both the holding duration and the cigarette extraction interval are below the set stability threshold, thereby identifying high-susceptibility relapse behaviors.

[0014] As a further aspect of the present invention, the behavior data generation module includes:

[0015] The signal acquisition submodule acquires the signal data output by the accelerometer, infrared component and Hall device, synchronizes the acceleration time series, infrared trigger waveform and Hall magnetic field change signal to the same time axis, identifies the opening start point and closing end point according to the trigger level change value, extracts the signal segment corresponding to each cigarette retrieval behavior, and generates an action signal interval dataset.

[0016] The behavior parameter extraction submodule, based on each segment in the action signal interval dataset, calls the acceleration signal change rate, infrared reflection amplitude and Hall sensing intensity to divide each action interval into a start-up phase, a maintenance phase and an end phase, and extracts three types of parameters in each segment: peak amplitude, sustained mean and convergence rate, to generate a set of behavior phase parameters.

[0017] The action sequence encoding submodule constructs a three-dimensional vector representation for each action segment based on the three types of parameter combinations extracted from the action stage parameter set. It then constructs a behavior sequence by arranging the parameters in a fixed order and performs sliding window filtering and noise reduction on the repeated and abnormal vectors in the sequence to generate a sequence of cigarette-taking action segments.

[0018] As a further aspect of the present invention, the coupling relationship extraction module includes:

[0019] The timestamp positioning submodule obtains all timestamp information recorded in the sequence of cigarette-taking action segments, groups the timestamps by date, aligns the time values ​​in each group to the hour, divides each hour into independent segments according to the daily timeline, calculates the number of timestamps in each segment and determines whether it exceeds the preset action concentration threshold, and obtains the action concentration distribution segment index.

[0020] The tag connection statistics submodule, based on the segments in the action distribution segment index, calls the behavior scene tags corresponding to each segment in the original behavior segment record, constructs a connection relationship sequence for consecutive behavior segment tags within the same time segment, counts the number of occurrences of adjacent tag pairs in all segments, filters out tag combinations whose frequency does not reach the preset connection frequency threshold, and obtains a list of high-frequency tag connection pairs.

[0021] The coupling structure generation submodule constructs a connection graph structure composed of tag pairs based on the connection relationships combined in the high-frequency tag connection pair list. It uses sequential node arrangement and connection edge numbering to mark the directionality and frequency of occurrence of the connection edges and generate a coupling structure group that triggers behavior.

[0022] As a further aspect of the present invention, the abnormal action screening module includes:

[0023] The attitude index extraction submodule extracts the jitter amplitude change and attitude stop time delay value corresponding to the continuous behavior combination sequence recorded in the trigger behavior coupling structure group, constructs a two-dimensional index pair vector for each combination and binds the behavior path sequence identifier to generate an attitude stability index set.

[0024] The adjacent point pair screening submodule calls each set of index pair vectors in the attitude stability index set, compares the difference between the amplitude change value and the time delay value between consecutive point pairs, selects point pairs with amplitude difference not less than the set amplitude change threshold and time interval less than the set delay interval threshold, screens local action sequences with unstable characteristics, and obtains a set of unstable point pairs.

[0025] The path deduplication filtering submodule checks whether the coordinate points in the path appear repeatedly in time according to the original coordinate path corresponding to each pair of points in the unstable point pair set. It records the behavior combinations with duplicate coordinate points in the path as items to be eliminated, performs coordinate uniqueness filtering on all unstable action paths, and filters out the behavior set that meets the path continuity to obtain the unstable smoke retrieval action set.

[0026] As a further aspect of the present invention, the reabsorption signal identification module includes:

[0027] The time parameter extraction submodule extracts the corresponding holding duration and the smoking interval time between actions based on each group of smoking behavior records in the unstable smoking action set. It calculates the duration as the time difference between the holding start point and the end point, and calculates the interval time as the time difference between the end of the adjacent action and the start of the next action, generating a smoking time feature matrix.

[0028] The threshold condition determination submodule calls the holding duration and the cigarette retrieval interval in the cigarette retrieval time feature matrix. Based on the preset stable action threshold range, the holding time and the interval are compared with the holding stability threshold and the interval stability threshold, respectively, and the record combination that meets both low threshold conditions is filtered to obtain a list of suspected repeated behavior combinations.

[0029] The behavior sequence identification submodule retrieves behavior trajectory sequences from the original unstable behavior set based on the index position in the suspected repetitive behavior combination list, and marks them as suspicious repetitive action records to obtain high-suspicion relapse behavior sequences.

[0030] As a further aspect of the present invention, the intervention behavior generation module includes:

[0031] The joint pattern extraction submodule collects the corresponding initiation amplitude value and holding duration for each group of behavior records in the high-suspicion relapse behavior sequence, combines the two in chronological order to construct a two-dimensional joint data pair, records the position index of the joint data pair in the behavior sequence, constructs a joint action pattern sequence in the initiation and holding dimensions, and generates a set of joint behavior pattern sequences.

[0032] The high-frequency matching and recognition submodule calls all joint data pairs in the behavior joint pattern sequence set, accumulates the number of times each pattern appears in the sequence, filters joint patterns whose frequency exceeds a preset threshold, records the corresponding matching label and the frequency of occurrence, and obtains a list of high-frequency joint patterns.

[0033] The reminder content generation submodule, based on the pattern types already marked in the high-frequency joint pattern list, corresponds to three preset behavioral intervention content templates, binds the pattern tags and types to the three content channels of vibration reminder, sound blocking, and image prompt, combines the execution action instruction sets under the three channels and numbers them, establishes a mapping structure between behavior triggering and intervention actions, and generates a behavioral reminder execution content group.

[0034] On the other hand, a smart cloud-based smoking cessation control method based on a behavioral science model, which is executed based on the aforementioned smart cloud-based smoking cessation control system based on a behavioral science model, includes the following steps:

[0035] S1: Collect signal data from the accelerometer, infrared components and Hall device, divide the time axis segment from opening to closing the cover, extract the three behavioral values ​​of start-up amplitude, maintenance intensity and end convergence, and construct a sequence of cigarette-taking action behavior segments.

[0036] S2: Extract the action timestamps from the cigarette-taking action behavior segment sequence, locate the concentrated distribution points within a fixed time period each day, count the number of connections between behavior scene tags within the time period, and construct a trigger behavior coupling structure group based on tag combinations whose frequency exceeds a preset threshold.

[0037] S3: Extract the jitter amplitude and attitude stop delay values ​​in the trigger behavior coupling structure group, screen the point pairs with changing amplitude and close intervals, remove the coordinate coincidence points in the jitter path, and filter the set of unstable smoke-taking actions.

[0038] S4: Extract the holding duration and cigarette taking interval from the unstable cigarette taking action set, and determine whether they are both lower than the preset stable threshold. If the condition is met, obtain the high-susceptibility relapse behavior sequence.

[0039] S5: Extract the activation amplitude value and holding duration from the relapse high-susceptibility behavior sequence, count the repetition frequency of the coupling mode, match the mode type with a frequency higher than the threshold, specify vibration reminder, sound blocking and image prompt, and generate a behavior reminder execution content group.

[0040] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0041] By collecting information on the amplitude, duration, and convergence of actions throughout the entire process of cigarette retrieval, a sequence of action segments with sequential characteristics is constructed. Combined with the tag connections during peak daily activity periods, a behavioral coupling pattern between high-frequency trigger scenarios is established. Based on this, unstable behavioral trajectories containing characteristics of shaking amplitude and posture delay are identified. Furthermore, repetitive action sequences with abnormal grasping time and cigarette retrieval intervals are extracted to identify behavioral combinations with a tendency to relapse. Based on the characteristic type and frequency of the repetitive patterns, various forms of intervention are dynamically matched to improve the targeting, timeliness, and efficiency of the prompts and behavioral correction. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of the system of the present invention;

[0044] Figure 2 This is a flowchart of the behavioral data generation module in this invention;

[0045] Figure 3 This is a flowchart of the coupling relationship extraction module in this invention;

[0046] Figure 4 This is a flowchart of the abnormal action screening module in this invention;

[0047] Figure 5 This is a flowchart of the reabsorption signal recognition module in this invention;

[0048] Figure 6 This is a flowchart of the intervention behavior generation module in this invention;

[0049] Figure 7This is a structural diagram of the intelligent smoking cessation box system of the present invention. Detailed Implementation

[0050] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0051] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0052] This invention provides an intelligent cloud-based smoking cessation control system based on a behavioral science model, such as... Figure 1 The diagram shown illustrates an intelligent cloud-based smoking cessation control system based on a behavioral science model. The system includes:

[0053] The behavior data generation module collects signal data from the accelerometer, infrared components and Hall device, divides the time axis into segments according to the opening start point and closing end point, extracts three behavior values ​​in the action sequence: start amplitude, maintenance intensity and end convergence, and divides and encodes each continuous action segment to construct a sequence of cigarette taking action behavior segments.

[0054] The coupling relationship extraction module extracts the action timestamps from the continuous records based on the sequence of cigarette-picking action behavior segments, locates the concentrated distribution points of actions within a fixed time period each day, counts the number of connections between behavior scene tags in all marked time periods, and constructs a trigger behavior coupling structure group based on the tag combinations in the connection relationship that appear more frequently than a preset frequency threshold.

[0055] The abnormal action filtering module is based on the trigger behavior coupling structure group. It extracts the jitter amplitude and attitude stop delay values ​​included in the continuous behavior combination, screens point pairs with amplitude changes and close intervals, removes coordinate coincidence points in the jitter path, and filters the set of unstable smoke picking actions.

[0056] The relapse signal identification module is based on the set of unstable cigarette-taking actions. It extracts the holding duration and cigarette-taking interval included in each group of actions, and determines whether the two time values ​​are simultaneously below the preset stable threshold range. If both conditions are met, it is recorded as a suspected repetitive behavior combination, and the high-suspicion relapse behavior sequence is identified.

[0057] The intervention behavior generation module is based on the high-suspicion relapse behavior sequence. It extracts the joint pattern of all initiation amplitude values ​​and holding duration, counts the repetition frequency in the continuous behavior of the joint pattern, and specifies three preset action contents for the pattern matching type with a repetition frequency higher than the threshold: vibration reminder, sound blocking and image prompt, and generates a behavior reminder execution content group.

[0058] The sequence of cigarette-taking action segments includes initiation amplitude encoding, maintenance intensity encoding, and termination convergence encoding. The triggering behavior coupling structure group includes high-frequency label combinations, number of connection relationships, and concentrated distribution time periods. The set of unstable cigarette-taking actions includes amplitude mutation point pairs, posture delay data, and deduplicated trajectory points. The relapse high-susceptibility behavior sequence includes low-value holding duration, short-interval cigarette-taking time, and repetitive behavior identifiers. The behavior reminder execution content group includes vibration reminder signals, sound blocking commands, and image prompts.

[0059] Specifically, such as Figure 2 As shown, the behavior data generation module includes:

[0060] The signal acquisition submodule acquires the signal data output by the accelerometer, infrared component and Hall device, synchronizes the acceleration time series, infrared trigger waveform and Hall magnetic field change signal to the same time axis, identifies the opening start point and closing end point according to the trigger level change value, extracts the signal segment corresponding to each cigarette retrieval behavior, and generates an action signal interval dataset.

[0061] First, a physical connection channel with the hardware layer is established. The accelerometer, infrared component, and Hall effect sensor are initialized and configured. The accelerometer's sampling frequency is set to 50 Hz, with a range of ±2 gravitational acceleration. The infrared component's sampling period is 20 milliseconds, and the Hall effect sensor's magnetic field sensitivity is set to 1.5 mV / Gauss. This submodule establishes a unified system clock source as the reference for the main time axis, using a first-in-first-out (FIFO) queue mechanism to buffer the data streams from the three sensors. To address the data misalignment issue caused by inconsistent sampling rates of different sensors, the submodule executes a linear interpolation algorithm. Using the accelerometer's timestamp as a reference, it calculates the estimated values ​​of the infrared trigger waveform and the Hall effect magnetic field change signal at corresponding time points, forcibly mapping the three signals to the same nanosecond-level precision time axis. Subsequently, the submodule invokes preset level-triggered logic to monitor the absolute value change of the acceleration signal in real time. When the acceleration value exceeds the set threshold of 1.2 gravitational acceleration for three consecutive sampling points, and the infrared reflection voltage value jumps from a low level of 0.5 volts to a high level of 3.3 volts, it is determined as the opening start point and the current timestamp is recorded. When the Hall sensor output voltage recovers to the initial steady-state value of 2.5 volts and the rate of change of acceleration approaches 0, it is determined as the closing end point. Based on the identified start and end points, the submodule backtracks 100 sampling points as a pre-buffer and extends 100 sampling points forward as a post-buffer, completely capturing the three-dimensional signal data within this time period. During this process, the submodule also needs to perform noise reduction processing on the original signal, using a moving average filtering algorithm, setting the window size to 5, summing the 5 consecutive sample values ​​within the window and dividing by 5 to obtain the smoothed signal value. For example, if the continuously acquired acceleration values ​​are 1.1, 1.2, 1.3, 1.2, and 1.2 gravitational accelerations, the submodule adds them together to get 6.0, then divides by 5 to get 1.2 gravitational acceleration as the filtered value for that moment. Through the above processing, the submodule completes the conversion and alignment of the original physical signal into a structured digital signal, encapsulates the original waveform data of each smoke-collecting action into independent data units, and finally generates an action signal interval dataset.

[0062] The behavior parameter extraction submodule, based on each segment in the action signal interval dataset, calls the acceleration signal change rate, infrared reflection amplitude and Hall sensing intensity to divide each action interval into the initiation phase, maintenance phase and termination phase, and extracts three types of parameters in each segment: peak amplitude, sustained mean and convergence rate, to generate a set of behavior phase parameters.

[0063] The submodule calls upon each independent segment of the action signal interval dataset. First, it performs a first-order difference operation on the acceleration signal, calculating the difference between two adjacent sampling points and dividing it by the sampling time interval to obtain the acceleration signal rate of change curve. Simultaneously, it directly reads the voltage data of the infrared reflection amplitude and the magnetic flux data of the Hall effect sensing intensity. Based on the zero-crossing characteristics of the acceleration signal rate of change, the submodule divides the entire action interval into three logical stages: the interval where the rate of change increases positively from 0 until it reaches its first peak and then falls back to 0 is defined as the start-up stage; the interval where the rate of change fluctuates slightly around 0 for more than 0.5 seconds is defined as the maintenance stage; and the interval where the rate of change increases negatively and eventually converges to 0 is defined as the end stage. For the startup phase, the submodule iterates through all acceleration signal sampling points within this phase, compares the numerical values ​​to identify the maximum value, and extracts it as the peak amplitude. For the maintenance phase, the submodule accumulates all infrared reflection amplitudes within this phase and divides them by the total number of sampling points in this phase to calculate the continuous average value. For the termination phase, the submodule calculates the slope of the Hall sensing intensity decaying from the maximum value to the reference value, that is, dividing the difference between the maximum value and the reference value by the time consumed by the decay process to obtain the convergence rate. To verify the accuracy of parameter extraction, the submodule sets a set of benchmark data: Assuming the acceleration sampling point set for the startup phase is 1.0, 1.5, 2.0, 1.5, and 1.0 gravitational accelerations, the submodule extracts the peak amplitude as 2.0 gravitational acceleration using a comparison algorithm; assuming the infrared reflection amplitude for the maintenance phase is 3.0 volts, 3.2 volts, and 3.1 volts, with 3 sampling points, the submodule adds the three values ​​to obtain 9.3 volts, then divides by 3 to calculate the sustained average value of 3.1 volts; assuming the Hall signal decays from 5.0 millivolts to 1.0 millivolts in 2 seconds during the termination phase, the submodule subtracts 1.0 from 5.0 to obtain 4.0, then divides by 2 to calculate the convergence rate as 2.0 millivolts per second. By executing the above calculation process for each segment, the submodule transforms complex waveform data into specific physical characteristic parameters, ultimately generating a set of behavioral phase parameters.

[0064] The action sequence encoding submodule constructs a three-dimensional vector representation for each action segment based on the three types of parameter combinations extracted from the action stage parameter set. It constructs the action sequence by arranging them in a fixed order and performs sliding window filtering and noise reduction on the repeated and abnormal vectors in the sequence to generate a sequence of cigarette-taking action segments.

[0065] First, the submodule accesses the parameter set for the behavior phase and reads three key parameters: peak amplitude, sustained mean, and convergence rate. These three scalar values ​​are arranged in a fixed physical dimension order to construct a three-dimensional feature vector in the form of (peak amplitude, sustained mean, convergence rate). The submodule then concatenates all the constructed three-dimensional vectors sequentially according to the absolute chronological order of the actions to form the original behavior sequence. To eliminate abnormal data caused by sensor noise or accidental touches, the submodule introduces a sliding window filtering mechanism, setting the window length to 5 vector units and the step size to 1 vector unit. Within each window, the submodule calculates the average Euclidean distance between the window's center vector and its surrounding neighboring vectors. If this average distance exceeds a preset outlier threshold, the center vector is determined to be an abnormal noise point and is removed. For example, setting the outlier threshold to 1.5, with the center vector coordinates within the window at (2, 2, 2) and a neighboring vector at (1, 1, 1), the submodule first calculates the differences between the two vectors in the three dimensions as 1, 1, and 1 respectively. The sum of the squares of these differences yields 3, and the square root of 3 gives the Euclidean distance, approximately 1.732. Since 1.732 is greater than the set threshold of 1.5, and if the average distance between this vector and its neighboring vectors is still greater than 1.5, the submodule will perform a removal operation. Simultaneously, the submodule checks if there are consecutive identical vectors in the sequence. If so, it retains the first vector and removes subsequent duplicates to achieve data deduplication. After the sliding window cleaning and deduplication process, the remaining vector sequence accurately represents the user's actual cigarette-picking behavior trajectory. The submodule then repackages these filtered vectors according to their temporal index, ultimately generating a sequence of cigarette-picking action behavior segments.

[0066] Specifically, such as Figure 3 As shown, the coupling relationship extraction module includes:

[0067] The timestamp positioning submodule obtains all timestamp information recorded in the sequence of cigarette-picking action segments, groups the timestamps by date, aligns the time values ​​in each group to the hour, divides each hour into independent segments according to the daily timeline, calculates the number of timestamps in each segment and determines whether it exceeds the preset action set threshold, and obtains the action set distribution segment index.

[0068] First, the module iterates through the sequence of cigarette-retrieving action segments, parsing the 64-bit UNIX timestamp information in the header of each record and converting it into a standard date and time format (year-month-day hour:minute:second). The submodule categorizes all timestamps into corresponding date groups based on the "year-month-day" field. Then, within each date group, it performs secondary division by hour, cutting a 24-hour day into 24 independent logical segments. For each hourly segment, the submodule configures a counter variable. It iterates through all timestamps within that segment, incrementing the counter by 1 for each record belonging to that time period, thus calculating the total number of actions within that segment. The submodule pre-defines the calculation logic for the action set threshold: it selects the average number of actions in all hourly segments over the past 7 days, adding 1.5 times the standard deviation as a dynamic threshold. For example, if the historical average is 10 times per hour and the standard deviation is 2, the threshold is set to 10 plus 1.5 times 2, which is 13 times. The submodule compares the count value of the current hourly segment with the threshold. If the count value is greater than 13, the hourly segment is marked as a "high-frequency active segment" and its index position on the timeline of the whole day is recorded to obtain the index of the action concentration distribution segment.

[0069] The tag connection statistics submodule is based on the segments in the action distribution segment index. It calls the corresponding behavior scene tags for each segment in the original behavior segment record, constructs a connection relationship sequence for consecutive behavior segment tags within the same time segment, counts the number of occurrences of adjacent tag pairs in all segments, filters out tag combinations whose frequency does not reach the preset connection frequency threshold, and obtains a list of high-frequency tag connection pairs.

[0070] Based on the index of action concentration distribution segments, the submodule locates the data corresponding to the time period in the original behavior segment records and reads the pre-labeled behavior scene tags for each data segment, such as "office," "walking," and "driving." Within the same time segment, the submodule combines adjacent tags into a directed pair according to the order in which the behavior occurs. For example, if "office" appears before "smoking," a connection relationship of "office-smoking" is constructed. The submodule establishes a hash map table, using the name string of the tag pair as the key and its occurrence count as the value. It iterates through all adjacent tag pairs within the time segment, incrementing the value of the corresponding key by 1 each time the same combination appears. After the statistics are completed, the submodule sets a preset connection frequency threshold of 5 times, iterates through all records in the hash table, and directly removes tag pairs with a count less than 5, retaining only combinations with a count greater than or equal to 5. For example, if the "office-smoking" combination appears 8 times, while the "sleep-smoking" combination appears only 2 times, the submodule will keep the former and discard the latter, thereby filtering out statistically significant behavior patterns and obtaining a list of high-frequency tag connection pairs.

[0071] The coupling structure generation submodule constructs a connection graph structure composed of tag pairs based on the connection relationships combined in the high-frequency tag connection pair list. It uses sequential node arrangement and connection edge numbering to mark the directionality and frequency of occurrence of the connection edges and generate a coupling structure group that triggers behavior.

[0072] The submodule reads each combination from the list of high-frequency label connection pairs, defining the label as a node in the graph and the connection relationship as a directed edge between the nodes. For each label pair (node ​​A, node B), the submodule constructs an arrow pointing from node A to node B in memory and assigns the statistically obtained frequency value as the weight attribute of the edge. The submodule uses a depth-first search algorithm to traverse all constructed nodes and edges, identify whether there are closed loops or complex network structures, and sorts the connection edges according to the magnitude of their frequency weights, with edges with larger weights being numbered earlier. For example, if the weight of "office-smoking" is 20 and the weight of "driving-smoking" is 15, then the former is marked as connection number 1 and the latter as connection number 2. Through this structured mapping, the submodule transforms discrete label pairs into a network graph with topological features, intuitively displaying the association strength and flow direction of smoking behavior triggered by different scenarios, and finally generating a group of coupled structures for triggering behaviors.

[0073] Specifically, such as Figure 4 As shown, the abnormal action filtering module includes:

[0074] The attitude index extraction submodule extracts the jitter amplitude change and attitude stop time delay value corresponding to the continuous behavior combination sequence recorded in the trigger behavior coupling structure group. It constructs a two-dimensional index pair vector for each combination and binds the behavior path sequence identifier to generate a set of attitude stability indices.

[0075] For each continuous sequence of actions within the triggered behavior coupling structure group, the submodule deeply analyzes the corresponding raw sensor data waveforms. The submodule calculates the sum of the variances of the acceleration signal along the three axes to quantify the change in jitter amplitude during the action; simultaneously, it calculates the time difference between the end of one action and the start of the next, as the attitude stopping time delay value. The submodule encapsulates these two calculated values ​​into a two-dimensional vector (jitter amplitude, stopping delay) and binds it to the corresponding behavior path ID. For example, if the system collects X, Y, and Z axis acceleration variances of 0.01, 0.02, and 0.01 for a certain behavior segment, the submodule sums them to obtain a jitter amplitude change of 0.04 timestamp of gravitational acceleration; if the end timestamp of this behavior is 1000 seconds and the start timestamp of the next behavior is 1002 seconds, the calculated attitude stopping time delay value is 2 seconds. The submodule associates and stores this vector (0.04, 2) with the unique identifier of the behavior sequence. In this way, the abstract behavior sequence is transformed into a quantifiable stability evaluation index, generating a set of attitude stability indices.

[0076] The adjacent point pair screening submodule calls each set of index pair vectors in the attitude stability index set, compares the difference between the amplitude change value and the time delay value between consecutive point pairs, selects point pairs with amplitude difference not less than the set amplitude change threshold and time interval less than the set delay interval threshold, screens local action sequences with unstable characteristics, and obtains a set of unstable point pairs.

[0077] The module iterates through the attitude stability index set, setting the amplitude change threshold to 0.1 times the square of gravitational acceleration and the delay interval threshold to 3 seconds. The submodule reads the jitter amplitude and time delay value from each index pair vector, executing a dual-condition judgment logic: first, it checks if the jitter amplitude is greater than or equal to 0.1; second, it checks if the time delay value is less than 3. Only when both conditions are met is the submodule considered an unstable point. For example, for the vector (0.04, 2), since 0.04 is less than 0.1, the point is considered stable; while for the vector (0.15, 1.5), since 0.15 is greater than 0.1 and 1.5 is less than 3, the submodule considers it an unstable point. This logic aims to filter out rapid behaviors with drastic movements and almost no pauses or buffers, which usually reflect abnormal psychological or physiological states through physical characteristics. The submodule extracts the indices of all point pairs that meet the conditions and aggregates them into a set of unstable point pairs.

[0078] The path deduplication filtering submodule checks whether the coordinate points in the path are repeated in time based on the original coordinate path corresponding to each pair of points in the unstable point pair set. It records the behavior combinations with duplicate coordinate points in the path as items to be eliminated, performs coordinate uniqueness filtering on all unstable action paths, and filters out the behavior sets that meet the path continuity to obtain the unstable smoke retrieval action set.

[0079] Based on the indexes in the set of unstable points, the original path trajectories in the spatial coordinate system are retrieved in reverse order to obtain a path sequence composed of a series of three-dimensional coordinate points. The submodule establishes a coordinate point hash set, traverses the coordinate points in each path, and checks whether the current coordinate point already exists in the hash set. If a duplicate coordinate point is found, it indicates that the action has a wandering or repetitive trajectory in space, and the submodule marks the entire combination of behaviors containing the duplicate point as "to be removed". For example, if path A contains coordinate point (10, 20, 30), and this coordinate point has already appeared in the previous path B with a very short time interval, the submodule considers path A to be a redundant record. After completing the traversal and check, the submodule removes all records marked as "to be removed" from the set, retaining only those behavioral data that are unique and continuous in the spatial path, ensuring that the finally filtered abnormal behaviors are not misjudgments caused by sensor stationary drift or duplicate records, thus obtaining the set of unstable smoke-collecting actions.

[0080] Specifically, such as Figure 5 As shown, the re-inhalation signal recognition module includes:

[0081] The time parameter extraction submodule extracts the corresponding holding duration and the smoking interval between actions based on each group of smoking behavior records in the unstable smoking action set. The duration is calculated as the time difference between the holding start point and the end point, and the interval is calculated as the time difference between the end of the adjacent action and the start of the next action, generating a smoking time feature matrix.

[0082] For each record in the unstable smoke-retrieving action set, the start and end frames of the grasping action, as well as the end frame of the previous action and the start frame of the current action, are precisely located. The submodule performs subtraction operations: subtracting the start time from the end time of the current action to obtain the grasping duration; and subtracting the end time of the previous action from the start time of the current action to obtain the smoke-retrieval interval. To ensure the intuitiveness and usability of the data, the submodule organizes and stores the calculation results, as shown in Table 1 below:

[0083] Table 1 Calculation Table of Time Parameters for Cigarette Retrieval Behavior

[0084] Behavior Number Capture the starting time point (seconds) Capture termination time (seconds) End point of the previous action (seconds) Hold duration (seconds) Cigarette removal interval (seconds) 001 100 105 80 5 20 002 150 153 145 3 5 003 300 306 290 6 10

[0085] As shown in Table 1, the submodule precisely quantifies the temporal dimension features of each action through simple arithmetic difference operations. For example, for action number 001, the submodule subtracts 100 from 105 to obtain a holding duration of 5 seconds, and subtracts 80 from 100 to obtain an interval time of 20 seconds. These data constitute the basis matrix for subsequent discrimination, generating the cigarette retrieval time feature matrix.

[0086] The threshold condition determination submodule calls the holding duration and the cigarette retrieval interval in the cigarette retrieval time feature matrix. Based on the preset stable action threshold range, it performs interval judgment on the holding time and interval time with the holding stability threshold and the interval stability threshold, respectively, and filters the record combinations that meet both low threshold conditions at the same time to obtain a list of suspected repeated behavior combinations.

[0087] The system reads data from the cigarette-taking time feature matrix and loads preset stable action threshold parameters. The system sets the stable holding threshold range to 4 to 8 seconds and the stable interval threshold to 15 seconds. The submodule performs interval comparison logic for each record: first, it determines whether the holding duration falls within the closed interval of 4 to 8 seconds; then, it determines whether the cigarette-taking interval is greater than 15 seconds. If a record's holding time is within this interval, but the interval is less than 15 seconds, or both indicators show a short duration, the system will pay close attention. Specifically, for the "short interval, high frequency" characteristic of relapse behavior, the submodule's filtering rule is set as follows: if the holding duration is less than 4 seconds and the cigarette-taking interval is less than 15 seconds, it is judged as a suspected relapse. Substituting the data from Table 1 for calculation: For behavior 002, its holding time is 3 seconds less than 4 seconds, and its interval time is 5 seconds less than 15 seconds, satisfying both low threshold conditions. The submodule marks it as a hit. For behavior 001, although the holding time of 5 seconds is within the normal range, the interval time of 20 seconds is also greater than 15 seconds, so it is judged as normal. The submodule extracts the indexes of all records that meet the "double low" conditions to obtain a list of suspected duplicate behavior combinations.

[0088] The behavior sequence identification submodule retrieves behavior trajectory sequences from the original unstable behavior set based on the index position in the suspected repetitive behavior combination list, and marks them as suspicious repetitive action records to obtain high-suspicion relapse behavior sequences;

[0089] Based on the index number provided by the list of suspected recurring behavior combinations, the submodule traces back to the original set of unstable behaviors, retrieving the complete acceleration, infrared, and Hall signal trajectory sequences corresponding to that index. The submodule tags these trajectory sequences with "suspicious relapse" electronically and separates them from the original data stream, storing them separately in a high-priority memory area. This process not only identifies anomalies but also preserves the complete contextual data before and after the anomaly, allowing subsequent modules to perform deeper pattern matching. Through this index backtracking and tagging mechanism, the submodule achieves a leap from time feature screening to complete waveform identification, obtaining high-suspicion relapse behavior sequences.

[0090] Specifically, such as Figure 6 As shown, the intervention behavior generation module includes:

[0091] The joint pattern extraction submodule collects the corresponding initiation amplitude value and holding duration for each group of behavior records in the high-suspicion behavior sequence of relapse, combines the two in chronological order to construct a two-dimensional joint data pair, records the position index of the joint data pair in the behavior sequence, constructs a joint action pattern sequence in the initiation and holding dimensions, and generates a set of joint behavior pattern sequences.

[0092] First, the submodule accesses the high-suspicion relapse behavior sequence. For each behavior segment in the sequence, it extracts the peak acceleration value of its initiation phase as the initiation amplitude value, and the holding duration calculated by the preceding module. The submodule combines these two heterogeneous parameters into a two-dimensional joint data pair in the format of "(initiation amplitude value, holding duration)". Subsequently, the submodule traverses the entire sequence, recording the occurrence order index of each joint data pair on the time axis. For example, if the initiation amplitude of a certain behavior is extracted to be 2.5 gravitational acceleration and the holding duration is 3 seconds, then the data pair (2.5, 3) is constructed, and its index position is recorded as the 5th event. In this way, the submodule integrates single-dimensional parameters into a multi-dimensional behavior pattern descriptor, which can more comprehensively characterize the aggression and persistence of relapse behavior, generating a set of joint behavior pattern sequences.

[0093] The high-frequency matching and recognition submodule calls all joint data pairs in the behavior joint pattern sequence set, accumulates the number of times each pattern appears in the sequence, filters joint patterns whose frequency exceeds a preset threshold, records the corresponding matching label and frequency of occurrence, and obtains a list of high-frequency joint patterns.

[0094] A pattern counter is initialized, and all joint data pairs in the set of joint pattern sequences are traversed. The submodule uses an exact matching algorithm to group data pairs with the same initiation amplitude (with an error range of 0.1) and the same holding duration (with an error range of 0.5 seconds) into the same category and accumulate their occurrence counts. After the statistics are completed, the submodule calls a preset frequency threshold, for example, setting the threshold to 3 times. The submodule filters out joint patterns that have been counted more than 3 times, defines them as inertial re-inhalation patterns, and records their corresponding matching labels (such as "rapid short inhalation pattern") and specific occurrence frequencies. For example, if the pattern (2.5, 3) appears 5 times in the sequence, exceeding the threshold of 3, the submodule will output the pattern and its frequency of 5. This process aims to identify specific re-inhalation patterns that users unconsciously repeat, and to obtain a list of high-frequency joint patterns.

[0095] The reminder content generation submodule, based on the pattern types already marked in the high-frequency joint pattern list, corresponds to three preset behavioral intervention content templates, binds the pattern tags and types to the three content channels of vibration reminder, sound blocking and image prompt, combines the execution action instruction sets under the three channels and numbers them, establishes a mapping structure between behavior trigger and intervention action, and generates a behavioral reminder execution content group;

[0096] Based on the pattern type marked in the high-frequency joint pattern list, the pre-set intervention strategy library is invoked. The library contains vibration intensity levels, an audio file list, and image cues. The submodule establishes a mapping logic: if the pattern type is "rapid short inhalation," it binds the "high-intensity long vibration" instruction, the "sharp alarm sound" file, and the "red prohibition icon"; if the pattern type is "slow and frequent inhalation," it binds the "medium-intensity pulse vibration" instruction, the "voice persuasion" file, and the "yellow warning icon." The submodule packages the selected three-channel execution instructions and assigns a unique execution number. For example, for the aforementioned high-frequency pattern (2.5, 3), the system matches a first-level intervention strategy based on its high amplitude and short duration characteristics, generating an instruction package: start the motor to vibrate at 100% power for 2 seconds, play a 2000 Hz buzzer sound, and display a red icon with the number Img_Stop_01 on the screen. The submodule structurally encapsulates these instructions, ensuring that the execution end can parse and act without delay, generating a behavioral reminder execution content group.

[0097] The system collects users' smoking behavior data using a self-monitoring mechanism to establish personalized baseline states and leverages the awareness effect derived from the recorded behavior to achieve initial natural reduction. Subsequently, by analyzing habit loops, it identifies high-risk situations and psychological triggers that induce smoking, categorizing users into corresponding behavioral change stages. Building on this, the system abandons forced quitting and instead adopts a gradual behavior shaping strategy, setting smooth, phased reduction goals. When the urge to smoke occurs, it utilizes cognitive intervention and utility substitution mechanisms to push alternative behavioral suggestions with equivalent psychological satisfaction (such as deep breathing or drinking water) to break the original conditioned reflex chain. Finally, by dynamically assessing multidimensional factors such as stress and withdrawal reactions, the system predicts relapse risk in real time and adjusts the intervention intensity accordingly. Combined with supportive strategies and social support, it ultimately achieves scientific control from physiological detoxification to the elimination of psychological dependence.

[0098] The intelligent cloud-based smoking cessation control method based on behavioral science models is implemented based on the aforementioned intelligent cloud-based smoking cessation control system based on behavioral science models, and includes the following steps:

[0099] S1: Collect signal data from the accelerometer, infrared components and Hall device, divide the time axis segment from opening to closing the cover, extract the three behavioral values ​​of start-up amplitude, maintenance intensity and end convergence, and construct a sequence of cigarette-taking action behavior segments.

[0100] S2: Extract the action timestamps from the sequence of cigarette-picking action segments, locate the concentrated distribution points within a fixed time period each day, count the number of connections between action scene tags within the time period, and construct a trigger behavior coupling structure group based on tag combinations whose frequency exceeds a preset threshold.

[0101] S3: Extract the jitter amplitude and attitude stop delay values ​​in the trigger behavior coupling structure group, screen the point pairs with amplitude changes and close intervals, remove the coordinate coincidence points in the jitter path, and screen the set of unstable smoke picking actions.

[0102] S4: Extract the holding duration and cigarette-taking interval from the unstable cigarette-taking action set, and determine whether they are both below the preset stability threshold. If the condition is met, obtain the high-susceptibility relapse behavior sequence.

[0103] S5: Extract the initiation amplitude value and holding time from the high-susceptibility relapse behavior sequence, count the repetition frequency of the coupling pattern, match the pattern type with a frequency higher than the threshold, specify vibration reminder, sound blocking and image prompt, and generate a behavior reminder execution content group.

[0104] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the technical solution.

Claims

1. An intelligent cloud-based smoking cessation control system based on a behavioral science model, characterized in that, The system includes: The behavior data generation module collects signal data from the accelerometer, infrared components and Hall device, divides the time axis segment from opening to closing the cover, extracts three behavior values: start amplitude, maintenance intensity and end convergence, and constructs a sequence of behavior segments of the smoke-taking action. The coupling relationship extraction module extracts the action timestamps from the cigarette-taking action behavior segment sequence, locates the concentrated distribution points within a fixed time period each day, counts the number of connections between behavior scene tags within the time period, and constructs a trigger behavior coupling structure group based on tag combinations whose frequency exceeds a preset threshold. The abnormal action filtering module extracts the jitter amplitude and attitude stop delay values ​​in the trigger behavior coupling structure group, filters point pairs with changing amplitude and close intervals, removes coordinate coincidence points in the jitter path, and filters the set of unstable smoke-picking actions. The relapse signal recognition module extracts the holding duration and cigarette-taking interval from the unstable cigarette-taking action set, and determines whether they are both below a preset stable threshold. If the condition is met, a high-susceptibility relapse behavior sequence is obtained. The intervention behavior generation module extracts the initiation amplitude value and holding duration from the relapse high-susceptibility behavior sequence, counts the repetition frequency of the coupling mode, matches the mode type with a frequency higher than the threshold, specifies vibration reminder, sound blocking and image prompts, and generates a behavior reminder execution content group.

2. The intelligent cloud-based smoking cessation control system based on behavioral science models according to claim 1, characterized in that: The sequence of cigarette-taking action segments includes initiation amplitude encoding, maintenance intensity encoding, and termination convergence encoding. The triggering behavior coupling structure group includes high-frequency tag combinations, number of connection relationships, and concentrated distribution time periods. The set of unstable cigarette-taking actions includes amplitude mutation point pairs, posture delay data, and deduplicated trajectory points. The relapse high-susceptibility behavior sequence includes low-value holding duration, short-interval cigarette-taking time, and repetitive behavior identifiers. The behavior reminder execution content group includes vibration reminder signals, sound blocking commands, and image prompts.

3. The intelligent cloud-based smoking cessation control system based on behavioral science models according to claim 1, characterized in that: The tag combination based on frequency exceeding a preset threshold refers to the combination of tag pairs whose number of behavioral tag connections exceeds a preset threshold within a specific time period, thereby constructing a behavioral coupling structure.

4. The intelligent cloud-based smoking cessation control system based on behavioral science models according to claim 1, characterized in that: The determination of whether both the holding duration and the cigarette extraction interval are below the preset stability threshold means that both values ​​are below the set stability threshold, thus identifying high-susceptibility relapse behavior.

5. The intelligent cloud-based smoking cessation control system based on a behavioral science model according to claim 1, characterized in that, The behavior data generation module includes: The signal acquisition submodule acquires the signal data output by the accelerometer, infrared component and Hall device, synchronizes the acceleration time series, infrared trigger waveform and Hall magnetic field change signal to the same time axis, identifies the opening start point and closing end point according to the trigger level change value, extracts the signal segment corresponding to each cigarette retrieval behavior, and generates an action signal interval dataset. The behavior parameter extraction submodule, based on each segment in the action signal interval dataset, calls the acceleration signal change rate, infrared reflection amplitude and Hall sensing intensity to divide each action interval into a start-up phase, a maintenance phase and an end phase, and extracts three types of parameters in each segment: peak amplitude, sustained mean and convergence rate, to generate a set of behavior phase parameters. The action sequence encoding submodule constructs a three-dimensional vector representation for each action segment based on the three types of parameter combinations extracted from the action stage parameter set. It then constructs a behavior sequence by arranging the parameters in a fixed order and performs sliding window filtering and noise reduction on the repeated and abnormal vectors in the sequence to generate a sequence of cigarette-taking action segments.

6. The intelligent cloud-based smoking cessation control system based on a behavioral science model according to claim 1, characterized in that, The coupling relationship extraction module includes: The timestamp positioning submodule obtains all timestamp information recorded in the sequence of cigarette-taking action segments, groups the timestamps by date, aligns the time values ​​in each group to the hour, divides each hour into independent segments according to the daily timeline, calculates the number of timestamps in each segment and determines whether it exceeds the preset action concentration threshold, and obtains the action concentration distribution segment index. The tag connection statistics submodule, based on the segments in the action distribution segment index, calls the behavior scene tags corresponding to each segment in the original behavior segment record, constructs a connection relationship sequence for consecutive behavior segment tags within the same time segment, counts the number of occurrences of adjacent tag pairs in all segments, filters out tag combinations whose frequency does not reach the preset connection frequency threshold, and obtains a list of high-frequency tag connection pairs. The coupling structure generation submodule constructs a connection graph structure composed of tag pairs based on the connection relationships combined in the high-frequency tag connection pair list. It uses sequential node arrangement and connection edge numbering to mark the directionality and frequency of occurrence of the connection edges and generate a coupling structure group that triggers behavior.

7. The intelligent cloud-based smoking cessation control system based on a behavioral science model according to claim 1, characterized in that, The abnormal action filtering module includes: The attitude index extraction submodule extracts the jitter amplitude change and attitude stop time delay value corresponding to the continuous behavior combination sequence recorded in the trigger behavior coupling structure group, constructs a two-dimensional index pair vector for each combination and binds the behavior path sequence identifier to generate an attitude stability index set. The adjacent point pair screening submodule calls each set of index pair vectors in the attitude stability index set, compares the difference between the amplitude change value and the time delay value between consecutive point pairs, selects point pairs with amplitude difference not less than the set amplitude change threshold and time interval less than the set delay interval threshold, screens local action sequences with unstable characteristics, and obtains a set of unstable point pairs. The path deduplication filtering submodule checks whether the coordinate points in the path appear repeatedly in time according to the original coordinate path corresponding to each pair of points in the unstable point pair set. It records the behavior combinations with duplicate coordinate points in the path as items to be eliminated, performs coordinate uniqueness filtering on all unstable action paths, and filters out the behavior set that meets the path continuity to obtain the unstable smoke retrieval action set.

8. The intelligent cloud-based smoking cessation control system based on a behavioral science model according to claim 1, characterized in that, The reabsorption signal recognition module includes: The time parameter extraction submodule extracts the corresponding holding duration and the smoking interval time between actions based on each group of smoking behavior records in the unstable smoking action set. It calculates the duration as the time difference between the holding start point and the end point, and calculates the interval time as the time difference between the end of the adjacent action and the start of the next action, generating a smoking time feature matrix. The threshold condition determination submodule calls the holding duration and the cigarette retrieval interval in the cigarette retrieval time feature matrix. Based on the preset stable action threshold range, the holding time and the interval are compared with the holding stability threshold and the interval stability threshold, respectively, and the record combination that meets both low threshold conditions is filtered to obtain a list of suspected repeated behavior combinations. The behavior sequence identification submodule retrieves behavior trajectory sequences from the original unstable behavior set based on the index position in the suspected repetitive behavior combination list, and marks them as suspicious repetitive action records to obtain high-suspicion relapse behavior sequences.

9. The intelligent cloud-based smoking cessation control system based on a behavioral science model according to claim 1, characterized in that, The intervention behavior generation module includes: The joint pattern extraction submodule collects the corresponding initiation amplitude value and holding duration for each group of behavior records in the high-suspicion relapse behavior sequence, combines the two in chronological order to construct a two-dimensional joint data pair, records the position index of the joint data pair in the behavior sequence, constructs a joint action pattern sequence in the initiation and holding dimensions, and generates a set of joint behavior pattern sequences. The high-frequency matching and recognition submodule calls all joint data pairs in the behavior joint pattern sequence set, accumulates the number of times each pattern appears in the sequence, filters joint patterns whose frequency exceeds a preset threshold, records the corresponding matching label and the frequency of occurrence, and obtains a list of high-frequency joint patterns. The reminder content generation submodule, based on the pattern types already marked in the high-frequency joint pattern list, corresponds to three preset behavioral intervention content templates, binds the pattern tags and types to the three content channels of vibration reminder, sound blocking, and image prompt, combines the execution action instruction sets under the three channels and numbers them, establishes a mapping structure between behavior triggering and intervention actions, and generates a behavioral reminder execution content group.

10. A smart cloud-based smoking cessation control method based on a behavioral science model, characterized in that, The execution of the intelligent cloud-based smoking cessation control system based on behavioral science models according to any one of claims 1-9 includes the following steps: S1: Collect signal data from the accelerometer, infrared components and Hall device, divide the time axis segment from opening to closing the cover, extract the three behavioral values ​​of start-up amplitude, maintenance intensity and end convergence, and construct a sequence of cigarette-taking action behavior segments. S2: Extract the action timestamps from the cigarette-taking action behavior segment sequence, locate the concentrated distribution points within a fixed time period each day, count the number of connections between behavior scene tags within the time period, and construct a trigger behavior coupling structure group based on tag combinations whose frequency exceeds a preset threshold. S3: Extract the jitter amplitude and attitude stop delay values ​​in the trigger behavior coupling structure group, screen the point pairs with changing amplitude and close intervals, remove the coordinate coincidence points in the jitter path, and filter the set of unstable smoke-taking actions. S4: Extract the holding duration and cigarette taking interval from the unstable cigarette taking action set, and determine whether they are both lower than the preset stable threshold. If the condition is met, obtain the high-susceptibility relapse behavior sequence. S5: Extract the activation amplitude value and holding duration from the relapse high-susceptibility behavior sequence, count the repetition frequency of the coupling mode, match the mode type with a frequency higher than the threshold, specify vibration reminder, sound blocking and image prompt, and generate a behavior reminder execution content group.

Images