Behavioral path sequence mining application platform based on dynamic intervention timing selection

Abstract

This invention discloses a behavioral path sequence mining application platform based on dynamic intervention timing selection, belonging to the field of big data mining technology. It includes the following steps: a trend feature calculation module, a path propagation construction module, a path sequence correction module, an intervention timing selection module, and an intervention effectiveness determination module. The trend feature calculation module collects multi-source data from patrol terminals and extracts behavioral trend features within an adaptive sliding window; the path propagation construction module generates a dynamic boundary band and statistically analyzes the intervention propagation attenuation chain with a temporal forgetting factor; the path sequence correction module dynamically adjusts the boundary band in real time based on the cognitive load modulation coefficient and accelerates the effectiveness of the forgetting factor; the intervention timing selection module determines whether the trend feature continuously crosses the upper boundary to determine the intervention timing; and the intervention effectiveness determination module corrects the attenuation chain loss coefficient after intervention. This scheme can suppress false triggering caused by instantaneous noise, improving decision-making accuracy and resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of big data mining technology, and in particular to a behavioral path sequence mining application platform based on dynamic intervention timing selection. Background Technology

[0002] Currently, in key aspects such as path planning and dynamic intervention applied in forest patrol scenarios, existing technologies typically employ a two-stage processing approach combining offline modeling and online monitoring: by integrating multi-source data such as historical patrol logs, weather forecasts, and geographic information systems, models such as random forests are used to predict regional risk levels, and algorithms such as ant colony optimization or mixed-integer linear programming are used to generate initial optimal patrol paths and task recommendations; simultaneously, the positions of patrol personnel, vehicles, or drones are tracked in real time using GPS or BeiDou devices, and when abnormal events such as prolonged stays, serious deviations from preset paths, or sudden fires are detected, preset early warning schemes are triggered and the command center is notified to conduct manual intervention or implement dynamic replanning.

[0003] Based on the above technical solutions, it was found that existing behavioral path sequence mining technologies identify interveneable nodes and trigger intervention actions by independently calculating the contribution of each node based on historical behavioral data and setting static intervention thresholds. However, during actual patrol missions, since the contribution of each node is usually derived from historical statistics of other nodes that have not been intervened, it is difficult to effectively predict the effect of early intervention on the user's subsequent behavioral path. This leads to deviations in the original contribution assessment of later nodes in multi-node continuous intervention scenarios, and the system still triggers intervention instructions that may no longer be applicable as originally planned. At the same time, intervention decisions usually rely only on explicit behavioral characteristics such as location and dwell time, making it difficult to simultaneously assess whether the user is currently in a state of high cognitive load. This makes it easy for pushed intervention messages to be ignored or interrupted when the user's attention is highly focused, resulting in a waste of system resources and a gradual decrease in user trust in the intervention mechanism. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a behavioral path sequence mining application platform and system based on dynamic intervention timing selection. The technical solution is as follows: A behavioral path sequence mining application platform based on dynamic intervention timing selection is provided. This platform includes: a trend feature calculation module, used to collect multi-source data from patrol terminals during patrol mission execution, and calculate the behavioral trend features of patrol personnel within an adaptive sliding window based on the multi-source data; a path propagation construction module, used by patrol terminals to generate dynamic boundary bands based on behavioral trend features, and statistically analyze the intervention propagation attenuation chain, introducing a temporal forgetting factor to correct the intervention propagation attenuation chain; a path sequence correction module, used to calculate the cognitive load modulation coefficient of patrol personnel based on multi-source data, and dynamically adjust the dynamic boundary band and temporal forgetting factor in real time using the cognitive load modulation coefficient; an intervention timing selection module, used to determine whether the behavioral trend features cross the upper boundary of the dynamic boundary band, and select the intervention timing based on the crossing status; and an intervention effectiveness determination module, used to determine the effectiveness of each formal intervention, and correct the intervention propagation attenuation chain based on the effectiveness determination result, thus completing the behavioral path sequence mining based on dynamic intervention timing selection.

[0005] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. The behavioral path sequence mining application platform based on dynamic intervention timing selection provided by this invention collects multi-source data from patrol terminals through a trend feature calculation module and extracts behavioral trend features within an adaptive sliding window. A path propagation construction module generates a dynamic boundary band and statistically analyzes an intervention propagation attenuation chain with a temporal forgetting factor. A path sequence correction module then uses the cognitive load modulation coefficient to dynamically adjust the boundary band in real time and accelerate the effectiveness of the forgetting factor. An intervention timing selection module determines whether the trend feature continuously crosses the upper boundary to establish the intervention timing. Finally, an intervention effectiveness determination module corrects the loss coefficient in the attenuation chain after intervention, thus forming a complete dynamic intervention closed loop. This scheme can effectively suppress false triggering caused by instantaneous noise and continuously optimize the intervention priority between nodes through online feedback, improving the accuracy of intervention decisions and the efficiency of resource utilization.

[0006] 2. This invention integrates multi-source information such as voice activity detection, touchscreen interaction frequency, and terminal thermal status to generate a continuously variable cognitive load modulation coefficient. This allows the tightness of the intervention trigger boundary to match the patrol personnel's current attention span and workload level in real time. Under high load, the boundary is automatically widened to avoid sending unnecessary alerts when users are using walkie-talkies, climbing steep slopes, or when the terminal is overheating, reducing ineffective interruptions and system interference. Under low load, the boundary is automatically tightened to improve the response sensitivity to slight deviations or abnormal stops, guiding behavior back in a timely manner. This mechanism achieves adaptive coordination between intervention timing and the user's acceptable window without increasing the burden on additional sensors, significantly improving the user acceptance and actual guidance effect of the intervention system.

[0007] 3. Compared to existing behavioral intervention methods based on fixed thresholds or instantaneous feature comparisons, this technical solution replaces instantaneous limit-crossing detection with trend-crossing judgment, fundamentally suppressing false triggers and oscillation alarms caused by factors such as GPS drift and brief hesitation. It introduces an intervention propagation attenuation chain and a product-based loss accumulation mechanism, solving the problem of static node contribution becoming invalid after early intervention changes subsequent behavioral paths, making the decision-making logic in multi-node continuous intervention scenarios closer to the actual propagation effect. By integrating the continuous modulation coefficient of cognitive load and boundary dynamic drift, user state perception is upgraded from simple threshold masking to continuously variable control input. Simultaneously, through post-intervention behavior verification and online correction of the attenuation chain, a rapid closed-loop adaptive capability is formed without retraining the model. These innovations collectively improve the accuracy of intervention timing selection, the return on system resources, and the adaptive robustness of long-term deployment in dynamic and complex environments such as forest patrols. Attached Figure Description

[0008] 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.

[0009] Figure 1 This is a schematic diagram of the structure of a behavior path sequence mining application platform based on dynamic intervention timing selection provided in an embodiment of this application.

[0010] Figure 2 This is a flowchart of the intervention decision execution process for a hybrid architecture.

[0011] Figure 3 This is a schematic diagram of the hardware structure of the patrol terminal.

[0012] Figure 4 This is a schematic diagram of the interface of the behavior trajectory analysis module. Detailed Implementation

[0013] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0014] 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.

[0015] Example 1: This technical solution is applied to a forest patrol scenario: Patrol personnel carry intelligent patrol terminals to perform daily patrol tasks along a preset route. During the process, they may deviate from the route or stop abnormally due to complex terrain, distraction, or emergency. The intervention system needs to combine multi-dimensional information such as real-time location, movement status, voice communication, device interaction, and ambient temperature to determine whether it is appropriate to issue an intervention reminder. At the appropriate time, it should guide personnel back to the route or deal with risks in an appropriate manner, while avoiding frequent and ineffective disturbances, thereby improving patrol safety and task execution efficiency.

[0016] like Figure 1 The diagram shown is a structural schematic of a behavioral path sequence mining application platform based on dynamic intervention timing selection provided in this embodiment of the invention. The platform includes: a trend feature calculation module, a path propagation construction module, a path sequence correction module, an intervention timing selection module, an intervention effectiveness determination module, and a database. The database stores preset values ​​for various parameters in this embodiment of the invention.

[0017] The trend feature calculation module is used to collect multi-source data from patrol terminals during patrol mission execution, and calculate the behavioral trend features of patrol personnel within an adaptive sliding window based on the multi-source data.

[0018] The aforementioned patrol terminal refers to a smart handheld device carried by patrol personnel, integrating hardware such as a main processor, positioning subsystem, accelerometer, microphone, and temperature sensor. It is used to collect multi-source data in real time during patrols, including location trajectory, movement status, voice activity, interactive operations, and device thermal status, and to perform local computing tasks such as behavioral trend analysis, cognitive load assessment, and dynamic intervention decision-making. Figure 3 As shown, Figure 3 This is a schematic diagram of the hardware structure of the patrol terminal. The processor is used to perform calculation tasks such as trend feature calculation, dynamic boundary band generation, cognitive load modulation coefficient fusion, intervention timing selection, and attenuation chain correction. The memory is used to store multi-source data, historical statistical values ​​of behavioral trend features, attenuation coefficients of intervention propagation attenuation chains, and various preset threshold parameters. The communication interface is used to interact with the command center or receive monitoring information.

[0019] In this embodiment, the behavioral trend characteristics of patrol personnel are calculated, and the specific calculation process is as follows: Multi-source data, including GPS positioning coordinates of patrol personnel at each sampling time within an adaptive sliding window, instantaneous movement speed at each sampling time, and step frequency sequence of the patrol terminal's pedometer.

[0020] A sliding window is a technique used in continuous time series data processing where a fixed-length time interval or range of sampling points is defined. This interval moves forward over time, and after each movement, the window contains the latest data while discarding the oldest data, ensuring that analysis is always based on data from the most recent period. In this scheme, the length of the sliding window is adaptively adjusted according to the movement speed of the patrol personnel.

[0021] The adaptive sliding window is specifically limited as follows: The current movement speed of the patrol personnel is obtained, specifically monitored by the speed sensor of the patrol terminal. The current movement speed of the patrol personnel is compared with predefined speed threshold intervals, where each speed threshold interval corresponds to a sliding window length, to determine the speed threshold interval to which the current movement speed belongs, and thus obtain the corresponding sliding window length.

[0022] The GPS positioning coordinates and instantaneous movement speed are obtained through the positioning subsystem and speed sensor integrated into the patrol terminal, respectively. The step frequency sequence refers to a series of step frequency values ​​arranged in chronological order, output by the accelerometer pedometer hardware unit built into the terminal. Step frequency refers to the number of steps taken per unit of time, usually measured in steps per minute. The pedometer hardware calculates the current step frequency in real time by detecting the periodic acceleration waveform changes generated when a person walks, and records the step frequency value at each sampling moment or each step cycle in sequence to form a step frequency sequence that changes over time.

[0023] The deviation trend intensity and cumulative deviation area are obtained by correlating the GPS positioning coordinates at each sampling time; the dwell trend coefficient is obtained by correlating the instantaneous movement speed at each sampling time; and the rhythm abnormality index is obtained by correlating the pedometer step frequency sequence.

[0024] Specifically, the deviation trend strength is calculated by taking each sampling moment within the window and calculating the shortest vertical distance between the GPS positioning coordinates and the baseline patrol path coordinate sequence at that moment, thus obtaining the deviation distance sequence within the window. The least squares method is used to linearly fit the deviation distance sequence, with the timestamp as the independent variable and the deviation distance as the dependent variable. The slope of the regression line is then used as the deviation trend strength.

[0025] The cumulative deviation area is calculated by performing a trapezoidal integral of the deviation distance as a function of time at each sampling time within a sliding window, where time is the time interval between each pair of adjacent sampling times, and the integral result is used as the cumulative deviation area.

[0026] The dwell time trend coefficient is calculated by taking the ratio of the number of sampling moments in the sliding window where the instantaneous motion speed is lower than the predefined low speed threshold to the total number of sampling moments as the dwell time proportion, and by taking the ratio of the standard deviation to the mean of the instantaneous motion speed sequence in the sliding window as the speed variation coefficient. The dwell time proportion and the speed variation coefficient are used together as the dwell time trend coefficient.

[0027] The rhythm anomaly index is calculated by averaging the step frequency sequence within the sliding window to obtain the current average step frequency, and then calculating the percentage deviation of the current average step frequency relative to the step frequency baseline value. This percentage deviation is used as the rhythm anomaly index.

[0028] Behavioral trend characteristics include the intensity of deviation trend, cumulative deviation area, dwell trend coefficient, and rhythm abnormality index of patrol personnel within the adaptive sliding window.

[0029] The path propagation building module is used by patrol terminals to generate dynamic boundary bands based on behavioral trend characteristics, and to statistically analyze the intervention propagation attenuation chain. A temporal forgetting factor is introduced to correct the intervention propagation attenuation chain.

[0030] Furthermore, a dynamic boundary band is generated, which includes an upper boundary and a lower boundary. The initial values ​​for the generation of the upper and lower boundaries of the dynamic boundary band are set according to the preset quantiles of the corresponding behavioral trend features in the historical benchmark patrol data.

[0031] Specifically, the dynamic boundary band generation process includes two stages: offline initialization and online drifting. In the offline stage, the patrol terminal extracts normal patrol task samples without abnormal events or intervention records from historical baseline patrol data. For each behavioral trend feature, including cumulative deviation area, deviation trend intensity, dwell trend coefficient, and rhythm anomaly index, its numerical distribution in all normal samples is statistically analyzed. A preset high baseline quantile is taken as the initial value of the upper boundary of the feature, and a preset low baseline quantile is taken as the initial value of the lower boundary of the feature. These initial boundary values ​​are stored in the patrol terminal. In the online operation stage, the patrol terminal dynamically drifts the initial upper and lower boundaries based on the cognitive load modulation coefficient and thermal state factor calculated in real time.

[0032] Furthermore, the intervention propagation attenuation chain is used to record the probability of path change and contribution reduction coefficient between any two behavioral nodes.

[0033] The patrol terminal pre-defines the patrol area according to a set spatial granularity. For example, it divides the area into multiple action nodes at fixed intervals or based on landmarks, with each action node corresponding to a key location point on the patrol path. Each patrol task in the historical patrol log is represented as a sequence of action nodes arranged in chronological order.

[0034] The initial statistical process for determining the probability of path change is as follows: Baseline statistics under no-intervention conditions: Patrol mission records without any intervention are selected from historical patrol logs. For each pair of behavior nodes, preceding behavior node i and subsequent behavior node j, the number of times the patrol personnel actually passed through behavior node j after starting from behavior node i is counted in all no-intervention missions, and the ratio of this number to the total number of backward transfers of behavior node i is calculated to obtain the path traversal probability under no-intervention conditions.

[0035] Path change statistics under intervention conditions: Patrol mission records where intervention was performed at behavior node i were selected from historical patrol logs. For these missions, the number of times patrol personnel actually passed through subsequent behavior node j after intervention at behavior node i was counted and compared with the baseline number of passes without intervention: the ratio of the number of times behavior node j was passed after intervention at behavior node i to the number of times behavior node j was passed without intervention was calculated to obtain the path change probability.

[0036] The initial value statistical process for the contribution reduction coefficient is as follows: After intervention at behavior node i, the actual marginal contribution of behavior node j to the final success of the task is compared with the marginal contribution of behavior node j without intervention. The ratio deviation between the two is calculated to obtain the contribution reduction coefficient.

[0037] Specifically, the actual marginal contribution mentioned above involves the patrol terminal extracting data from completed patrol tasks from historical patrol logs. Each task record includes the sequence of behavioral nodes traversed during the task and the final success indicator, such as whether a fire hazard was discovered, whether the entire patrol mileage was completed, and whether an anomaly was reported within the specified time. For behavioral node j to be evaluated, all tasks containing behavioral node j are used as the first sample set, and all tasks not containing behavioral node j are used as the second sample set. The success rates of tasks in the first and second sample sets are calculated, and the difference between the two is taken as the baseline marginal contribution of behavioral node j under no-intervention conditions. Tasks that underwent intervention at behavioral node i and subsequently passed through behavioral node j are selected from the historical logs. The success rate of these tasks is calculated, and the baseline marginal contribution of behavioral node j under no-intervention conditions is subtracted from this post-intervention success rate. This difference between the success rate of tasks containing and not containing behavioral node j yields the incremental contribution brought about by the intervention. Finally, this incremental contribution is added to the baseline marginal contribution to obtain the actual marginal contribution of behavioral node j under intervention conditions.

[0038] When any action node performs an intervention, the reduction of the original contribution of the intervention node to the subsequent action node is accumulated in the form of a product to obtain the real-time effective contribution of the subsequent action node. The calculation method is to multiply the original static contribution by the unit value minus the reduction coefficients corresponding to all intervention nodes, and limit the attenuation upper limit threshold so that the final effective contribution is not lower than the original static contribution multiplied by the attenuation upper limit threshold.

[0039] It should be explained that the above calculation process is used to quantify the cumulative impact of multiple intervention events on the decision value of subsequent behavioral nodes: during the patrol mission, if intervention operations are performed successively at multiple preceding behavioral nodes, each intervention node will have a certain weakening effect on the original static contribution of subsequent behavioral nodes based on the attenuation coefficient obtained from offline statistics. At the same time, to prevent the effective contribution from approaching zero due to excessive multiplication, thus causing the intervention system to completely lose its intervention tendency towards that behavioral node, a decay upper limit threshold is set to ensure that the final effective contribution is not lower than the product of the original static contribution and this threshold.

[0040] In this embodiment, the calculation of real-time effective contribution can truly reflect the effect of early intervention on subsequent behavioral paths, dynamically adjust the intervention priority of subsequent behavioral nodes, and avoid the intervention system mechanically triggering intervention instructions that are no longer applicable according to the original static plan. At the same time, through the lower limit protection mechanism, it is ensured that even after multiple interventions, behavioral nodes that still have basic value for task success still retain the possibility of being intervened, thereby improving the adaptability and robustness of intervention decisions in complex dynamic environments.

[0041] In this embodiment, the intervention propagation attenuation chain is modified by introducing a temporal forgetting factor. The temporal forgetting factor works as follows: Before updating the real-time effective contribution of subsequent action nodes, the patrol terminal obtains the time difference between the current time and the intervention execution time of each already executed intervention node, and compares this time difference with a preset forgetting period: For intervention nodes whose time difference does not exceed the preset forgetting period, their temporal forgetting factor is set to one, indicating that the loss coefficient of the intervention node does not decay.

[0042] It's important to explain that the preset forgetting period is a time-length parameter set by the patrol terminal for each executed intervention node in the intervention propagation decay chain. It determines the starting point at which the diminishing impact of that intervention node on subsequent behavioral nodes begins to decay. Specifically, starting from the moment the intervention is executed, within this preset forgetting period, the diminishing coefficient of the intervention node remains intact, and its impact is fully preserved. Once the time difference between the current moment and the intervention execution moment exceeds this preset forgetting period, a temporal forgetting factor is triggered, causing the diminishing coefficient to decay over time. This parameter serves to balance the long-term impact of historical interventions with the natural recovery of user behavior, preventing excessively distant intervention events from having an unnecessary dominant influence on current decisions, thus ensuring that the calculation of real-time effective contribution more closely reflects the actual state of recent task execution.

[0043] For intervention nodes whose time difference exceeds a preset forgetting period, the patrol terminal calculates the ratio of the time difference exceeding the preset forgetting period to a preset decay time span to obtain a temporal forgetting factor. The original loss coefficient of the intervention node is multiplied by the temporal forgetting factor to obtain a decayed loss coefficient, which is then used in the calculation of a product. This ensures that the earlier the intervention node is, the smaller the impact on the contribution of subsequent behavioral nodes. When the time difference exceeds or reaches the decay time span, the temporal forgetting factor becomes zero, and the loss effect of the early intervention node is completely eliminated.

[0044] The path sequence correction module is used to calculate the cognitive load modulation coefficient of patrol personnel based on multi-source data, and to use the real-time drift boundary band and temporal forgetting factor of the cognitive load modulation coefficient.

[0045] Furthermore, the cognitive load modulation coefficient of patrol personnel is calculated as follows: Multi-source data also includes the communication occupancy mapping value of patrol terminals, touch interaction frequency, and thermal state factor during the standard fusion period.

[0046] The standard fusion period refers to a short, independent time window specifically set by the patrol terminal for calculating the cognitive load modulation coefficient. This period is fixed and does not change with the patrol personnel's movement speed or GPS sampling rate. At the end of each fusion period, the patrol terminal collects data such as the communication occupancy status and the number of touch screen interaction events accumulated during that period, performs weighted fusion to obtain the cognitive load modulation coefficient at the current moment, and starts the statistics again in the next period.

[0047] The communication occupancy status indicates the number of consecutive time windows that the patrol terminal determines as voice activity. When the number of consecutive voice activity windows exceeds a preset frame threshold, the communication occupancy status is recorded as "communication occupancy". It should be noted that the above voice activity determination is based on the output of a 1 or 0 from the audio encoder integrated in the patrol terminal.

[0048] The patrol terminal monitors touchscreen press or release events, counts the number of touchscreen events occurring within a standard fusion period, and calculates the ratio of the number of events to the length of the standard fusion period to obtain the touch interaction frequency of the patrol terminal.

[0049] The patrol terminal reads the temperature output values ​​from the battery temperature sensor and the central processing unit (CPU) temperature sensor, respectively. It then compares the battery temperature value and the CPU temperature value with their corresponding battery temperature thresholds to obtain the battery temperature deviation and CPU temperature deviation, respectively. The maximum of these two deviations is taken as the overall temperature deviation and compared with its corresponding deviation threshold. When the overall temperature deviation equals the first deviation threshold, the thermal state factor takes the first quantized value; when the overall temperature deviation equals the second deviation threshold, the thermal state factor takes the second quantized value; wherein the second preset value is less than the first preset value; for any overall temperature deviation between the first and second deviation thresholds, its thermal state factor is obtained by substituting into the expression: Thermal state factor = first preset value - ×(First preset value - Second preset value). When the overall temperature deviation equals the first preset deviation threshold, the numerator is 0, and the thermal state factor equals the first preset value; when the overall temperature deviation equals the second preset deviation threshold, the ratio equals 1, and the thermal state factor equals the first preset value minus the difference between the first and second preset values, i.e., the second preset value. For temperature deviations between these two values, the thermal state factor decreases linearly with temperature.

[0050] The communication occupancy mapping value, touch interaction frequency, and thermal state factor of the patrol terminal are normalized for their respective contributions. The normalized contribution results are then multiplied by their corresponding predefined weights and the product results are accumulated to obtain the preliminary modulation coefficients.

[0051] The cognitive load modulation coefficient is calculated as follows: In this embodiment, the contribution normalization process is as follows: the communication occupancy status includes communication occupancy and communication non-occupancy. When the communication occupancy status is displayed as communication occupancy, it is mapped to the first mapping value; otherwise, it is mapped to the second mapping value. The ratio of the mapping value to the preset maximum mapping value is recorded as the communication occupancy contribution value.

[0052] Divide the touchscreen interaction frequency by the preset reference frequency upper limit, and truncate any values ​​exceeding one to one, to obtain the normalized interaction frequency contribution value.

[0053] The thermal state factor is normalized to a standard interval according to a preset linear mapping relationship to obtain the thermal state contribution value corresponding to the standard interval.

[0054] Specifically, the normalized contribution results are the communication occupancy contribution value C. com Interaction frequency contribution value C int Thermal state contribution value C the The corresponding predefined weights are w1, w2, and w3, respectively. Substituting these into the calculation formula for the cognitive load modulation coefficient: Among them, C raw denoted as the cognitive load modulation coefficient of patrol personnel, s represents the preset scaling factor, w1 represents the predefined weight corresponding to the communication occupancy contribution value, w2 represents the predefined weight corresponding to the interaction frequency contribution value, and w3 represents the predefined weight corresponding to the hot state contribution value.

[0055] The upper and lower boundaries of the real-time drift dynamic boundary band and the effective speed of the drift-sequence forgetting factor are determined by the cognitive load modulation coefficient.

[0056] In this embodiment, the upper and lower boundaries of the drifting dynamic boundary band are specifically determined by: the patrol terminal acquiring the current cognitive load modulation coefficient at each sampling time, multiplying the initial upper boundary of the dynamic boundary band by the cognitive load modulation coefficient to obtain the real-time upper boundary, and performing a ratio calculation on the initial lower boundary of the dynamic boundary band and the cognitive load modulation coefficient to obtain the real-time lower boundary.

[0057] Specifically, when the cognitive load modulation coefficient is greater than one, the real-time upper boundary expands relative to the initial upper boundary and the real-time lower boundary shrinks relative to the initial lower boundary, making the intervention triggering conditions more lenient; when the cognitive load modulation coefficient is less than one, the real-time upper boundary shrinks and the real-time lower boundary expands, making the intervention triggering conditions more stringent.

[0058] In this embodiment, the cognitive load modulation coefficient is used as a dynamic adjustment factor to match the intervention triggering condition with the user's current cognitive load state in real time. When the cognitive load modulation coefficient is greater than one, the upper boundary is expanded and the lower boundary is narrowed, which is equivalent to raising the threshold for intervention triggering. This ensures that the intervention system only intervenes when the deviation behavior is significant, thereby avoiding sending non-essential reminders that interfere with the user's critical operations when their attention is highly focused, reducing the rate of invalid interruptions and increasing the user's trust in the intervention system. When the cognitive load modulation coefficient is less than one, the upper boundary is narrowed and the lower boundary is expanded, which is equivalent to lowering the threshold for intervention triggering. This allows the intervention system to more sensitively detect slight deviations or abnormal pauses and provide timely guidance. This mechanism achieves adaptive coordination between intervention sensitivity and the user's acceptable intervention window. While ensuring the timeliness of safety warnings, it minimizes interference with the user's normal task execution, thereby improving the user acceptance and actual guidance effect of the intervention system.

[0059] The effective speed of the drift-series forgetting factor is specifically achieved by the patrol terminal comparing the current cognitive load modulation coefficient with the load threshold. When the cognitive load modulation coefficient continuously exceeds the load threshold for a preset duration, the preset forgetting period of the time-series forgetting factor is temporarily shortened to a defined short period. This advances the start time for the time-series forgetting factor to begin decaying. Therefore, under high cognitive load conditions, the diminishing impact of early intervention nodes on the contribution of subsequent behavioral nodes will enter the decay phase earlier. The ratio of the defined short period to the preset forgetting period is used as a scaling factor and multiplied by the decay time span to obtain the temporary decay time span. The speed at which the loss coefficient decays from full value to zero is accelerated. This allows the residual impact of early intervention nodes to be eliminated more quickly and completely. The intervention system's intervention decisions for subsequent behavioral nodes will be able to break free from the constraints of historical interventions more quickly, thus responding more promptly to the actual needs of the current patrol task. When the cognitive load modulation coefficient falls below the load threshold, the preset forgetting period and decay time span are restored to their original values.

[0060] The intervention timing selection module is used to determine whether the behavioral trend characteristics cross the upper boundary of the dynamic boundary zone, and select the intervention timing based on the crossing status.

[0061] Furthermore, to determine whether the behavioral trend characteristics cross the upper boundary of the dynamic boundary zone, the specific determination process is as follows: The patrol terminal defines a crossing duration counter for each behavioral trend feature, including the patrol personnel's deviation trend intensity, cumulative deviation area, dwell trend coefficient, and rhythm abnormality index. The crossing duration counter is used to count the crossing duration of the patrol personnel's behavioral trend features.

[0062] If the current behavior trend feature value is greater than the real-time upper boundary, it is determined that the patrol personnel are in a crossing state, and the crossing duration counter corresponding to the current behavior trend feature value is incremented by a sampling interval value; otherwise, it is determined that the patrol personnel are not in a crossing state, and the corresponding crossing duration counter is cleared to zero.

[0063] The patrol terminal compares the current value of the continuous patrol counter with a time threshold to determine when to intervene.

[0064] In the dynamic intervention decision-making process of this invention embodiment, the original data after sliding window updates and trend feature calculations can be backtracked and verified through a list of behavioral trajectories. For example... Figure 4 As shown, Figure 4 This is a schematic diagram of the behavior trajectory analysis module interface. The diagram includes a query condition setting area such as region, user, and frequency. The frequency represents the time resolution option of the adaptive sliding window for GPS sampling. The behavior trajectory list below displays the ID, region, participating users, sampling frequency, start time, and status information of each patrol task. The status bar indicating "in progress" or "created" reflects the real-time output results of the crossing status judgment module in this technical solution, i.e., whether the current patrol personnel are in a normal movement state within the dynamic boundary zone, thus providing a data basis for selecting the timing of subsequent intervention.

[0065] Furthermore, determining the timing of intervention involves: When the cumulative value of the continuous crossing counter reaches or exceeds a predefined first time threshold for the first time, the patrol terminal enters a pre-intervention state and starts a confirmation window timer to continue monitoring within the confirmation window. The length of the confirmation window is a preset confirmation duration.

[0066] During the confirmation window, the patrol terminal continues to perform a crossing judgment at each sampling moment: if the behavioral trend characteristic value falls below the real-time upper boundary at any sampling moment within the confirmation window, it immediately exits the pre-intervention state, the confirmation window timer is cleared, and no intervention is triggered.

[0067] If the confirmation window timer completes the entire confirmation period, and the behavioral trend characteristic value does not fall below the real-time upper boundary during this period, the confirmation window ends, and the formal intervention judgment process begins.

[0068] Formal intervention is triggered as follows: After the confirmation window ends, the patrol terminal continues to monitor the cumulative deviation area characteristic value of the patrol personnel. If the cumulative deviation area continues to exceed the real-time upper boundary, and the total duration accumulated since the first crossing, i.e., the cumulative value of the crossing duration counter, reaches a predefined second time threshold, then formal intervention is triggered. The second time threshold is greater than the first time threshold.

[0069] Once the formal intervention is triggered, the patrol terminal selects the intervention level and performs dynamic intervention based on the current cognitive load modulation coefficient and the real-time effective contribution.

[0070] Specifically, the real-time effective contribution of subsequent action nodes is compared with a preset low contribution threshold: if the real-time effective contribution is lower than the low contribution threshold, the intervention is abandoned and no intervention action is performed.

[0071] It's important to explain that when a subsequent action node, after experiencing cumulative attenuation from multiple preceding intervention nodes, has a real-time effective contribution below a preset low contribution threshold, it indicates that the actual contribution of that action node to the final success of the task is extremely limited. Even if intervention is performed at this point, it will not bring significant value improvement. In this situation, the system proactively abandons intervention to avoid consuming system resources such as communication bandwidth, terminal battery power, and command center attention due to low-value interventions, while also reducing unnecessary disturbance to patrol personnel. This mechanism ensures that resource investment in intervention execution is always concentrated on action nodes with clear positive value, improving the overall return on investment of the intervention strategy and preventing resource waste and a decline in user experience caused by the multiplicative effect of the decay chain leading to the system mechanically triggering interventions on low-contribution action nodes.

[0072] If the real-time effective contribution is not lower than the low contribution threshold, the cognitive load modulation coefficient is further determined: when the cognitive load modulation coefficient of the patrol personnel is greater than or equal to the load threshold, a level 1 intervention is executed. This level of intervention only displays a non-blocking prompt icon on the terminal interface and does not involve physical intervention such as sound or vibration.

[0073] When the cognitive load modulation coefficient is less than the load threshold but greater than the preset second load threshold, a secondary intervention is implemented, sending navigation suggestions or a brief voice reminder. The load threshold is greater than the preset second load threshold.

[0074] When the cognitive load modulation coefficient is less than or equal to the preset second load threshold, and the cumulative deviation area exceeds a predetermined multiple of the real-time upper boundary, a level 3 intervention is implemented, a strong reminder is sent and synchronized to the command center; if the cumulative deviation area does not exceed the preset multiple, a level 2 intervention is still implemented.

[0075] After the formal intervention is implemented, the patrol terminal will reset the crossover duration counter corresponding to the behavioral trend characteristics to zero, exit the pre-intervention state, and wait for the next sampling cycle to start monitoring again.

[0076] The intervention effectiveness determination module is used to determine the effectiveness of the intervention after each formal intervention, correct the intervention propagation attenuation chain based on the effectiveness determination results, and complete the mining of behavioral path sequences based on dynamic intervention timing selection.

[0077] Furthermore, the intervention transmission attenuation chain was revised based on the effectiveness assessment results. The specific process is as follows: After each formal intervention, the patrol terminal starts an observation window. The length of the observation window is timed in real time by the clock hardware. At the end of the window, the degree of improvement of the behavioral trend characteristics after the intervention is calculated. The degree of improvement includes the reduction of the deviation area and the decrease in the percentage of dwell time. If the reduction of the deviation area is greater than the area improvement threshold or the decrease in the percentage of dwell time is greater than the percentage decrease threshold, it is determined to be an effective intervention; otherwise, it is an ineffective intervention.

[0078] The reduction in deviation area is calculated as follows: At the moment of formal intervention, the patrol terminal records the cumulative deviation area at the last sampling time before the intervention as the pre-intervention baseline value. At the end of the observation window, the patrol terminal recalculates the cumulative deviation area at the current sampling time as the post-intervention measurement value. The reduction in cumulative deviation area is equal to the pre-intervention baseline value minus the post-intervention measurement value.

[0079] Decrease in dwell time percentage: At the moment of formal intervention, the terminal device records the dwell time percentage within the last sliding window before intervention as the pre-intervention baseline. At the end of the observation window, the terminal device recalculates the dwell time percentage within the current sliding window as the post-intervention measurement. The decrease in dwell time percentage is equal to the pre-intervention baseline minus the post-intervention measurement.

[0080] The intervention propagation attenuation chain is corrected by: if the intervention is deemed effective, the patrol terminal reduces the contribution reduction coefficient of the behavioral nodes related to the intervention node in the intervention propagation attenuation chain by a preset first correction step; if the intervention is deemed ineffective, a preset second correction step is added; wherein the first correction step and the second correction step are both preset positive values, and the first correction step is less than or equal to the second correction step.

[0081] In this embodiment, the contribution reduction coefficient between node pairs is updated incrementally by using the actual observed improvement in behavior after each intervention as a feedback signal. This allows the decay chain to continuously learn and self-optimize based on real intervention results. After effective intervention, the corresponding reduction coefficient decreases appropriately, meaning that the impact of the reduction between the behavior node pairs is reduced in subsequent similar scenarios. The system retains the intervention tendency for subsequent nodes, avoiding missing valuable intervention opportunities due to the conservative tendency of historical statistical data. After ineffective intervention, the reduction coefficient increases appropriately, automatically suppressing the intervention priority of subsequent behavior nodes after intervention at the preceding behavior node in subsequent tasks, reducing resource consumption caused by repeated ineffective interventions. This correction mechanism makes the intervention propagation decay chain no longer limited by the one-time characteristics of offline statistics, and can adapt to new patrol areas, new behavior patterns, and individual differences. As the number of patrol tasks accumulates, the reduction coefficient gradually converges to a value closer to the actual scenario, and the accuracy of intervention timing selection and resource utilization efficiency continue to improve.

[0082] Example 2: Under the condition that other conditions remain unchanged in Example 1, if the area involved in the current patrol task is a new patrol area, the contribution reduction coefficient between any two behavioral nodes in the intervention propagation attenuation chain cannot be obtained through offline statistics because the area lacks historical patrol logs.

[0083] Therefore, the patrol terminal uses the group average default value as the contribution reduction coefficient between each node pair: based on the same type of terrain in the new patrol area, such as mountains, plains, dense forest paths, etc., a reference area with existing historical statistical data is matched, and the arithmetic mean of the contribution reduction coefficients of all node pairs in the reference area is used as the group average default value for the same type of terrain; this default value is assigned to the corresponding node pair in the new patrol area and stored in the intervention propagation attenuation chain.

[0084] During subsequent patrol missions, whenever an intervention is performed in the new patrol area and an effectiveness determination is completed, the patrol terminal updates the loss coefficient of the corresponding node pair according to the incremental correction method in Example 1, so that it gradually converges from the group average default value to the actual statistical value.

[0085] Example 3: Under the condition that other conditions remain unchanged in Example 1 or Example 2, when the patrol terminal detects that it is currently in an external power supply state through the system power management interface, such as being connected to a charger or power bank, the patrol terminal determines that the patrol terminal is in a charging state.

[0086] Therefore, the patrol terminal multiplies the initial thermal state factor obtained according to Embodiment 1 or Embodiment 2 by a preset charging gain coefficient to obtain the final thermal state factor. The charging gain coefficient is a preset positive number greater than 1. After multiplying the initial thermal state factor by this coefficient, the final thermal state factor value when in the charging state is higher than that when in the non-charging state under the same comprehensive temperature deviation conditions.

[0087] It should be explained that the intervention system provided by this technical solution adopts a concurrent architecture that combines event-driven and timed polling. For example... Figure 2 As shown, Figure 2This is a flowchart illustrating the intervention decision execution process of a hybrid architecture. When the positioning subsystem outputs GPS sampling, the main processor triggers a sliding window update, thereby completing trend feature calculation and crossing status judgment. Simultaneously, the fusion timer independently triggers the main processor's cognitive load modulation coefficient calculation and uses this to drift the dynamic boundary band in real time, feeding back the real-time boundary value to the crossing status judgment. The intervention system's background timer advances the forgetting cycle and decay time span timing according to actual time. When the cognitive load modulation coefficient continuously exceeds the high load threshold and the duration exceeds the preset duration, the system temporarily shortens the forgetting cycle and reduces the decay time span. After an intervention event occurs, the system starts a confirmation window. If crossing still occurs by the end of the window, the intervention action is executed; if the features decline within the window, the intervention is canceled. After the intervention action is executed, an observation window is started. At the end of the window, the degree of behavioral improvement is evaluated, and the decay chain loss coefficient is adjusted based on the effective or ineffective results. The confirmation window, observation window, and background timer each operate on an independent hardware clock source, running in parallel with the main loop. This ensures the decoupling and real-time performance of multi-timescale tasks, avoiding blocking or resource conflicts.

[0088] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0089] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0090] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A behavioral path sequence mining application platform based on dynamic intervention timing selection, characterized in that, Includes the following steps: The trend feature calculation module is used to collect multi-source data from patrol terminals during patrol mission execution, and calculate the behavioral trend features of patrol personnel within an adaptive sliding window based on the multi-source data. The path propagation construction module is used by patrol terminals to generate dynamic boundary bands based on behavioral trend characteristics, and to statistically analyze the intervention propagation attenuation chain. A temporal forgetting factor is introduced to correct the intervention propagation attenuation chain. The path sequence correction module is used to calculate the cognitive load modulation coefficient of patrol personnel based on multi-source data, and to use the real-time drift dynamic boundary band and temporal forgetting factor of the cognitive load modulation coefficient. The intervention timing selection module is used to determine whether the behavioral trend characteristics cross the upper boundary of the dynamic boundary zone, and select the intervention timing based on the crossing status. The intervention effectiveness determination module is used to determine the effectiveness of the intervention after each formal intervention, correct the intervention propagation attenuation chain based on the effectiveness determination results, and complete the mining of behavioral path sequences based on dynamic intervention timing selection.

2. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 1, characterized in that: The specific calculation process for the behavioral trend characteristics of the patrol personnel is as follows: The multi-source data includes the GPS positioning coordinates of the patrol personnel at each sampling time within the adaptive sliding window, the instantaneous movement speed at each sampling time, and the step frequency sequence of the patrol terminal's pedometer. The behavioral trend features include the deviation trend intensity, cumulative deviation area, dwell trend coefficient, and rhythm abnormality index of patrol personnel within the adaptive sliding window; The deviation trend intensity and cumulative deviation area are obtained by correlating the GPS positioning coordinates at each sampling time, the dwell trend coefficient is obtained by correlating the instantaneous movement speed at each sampling time, and the rhythm abnormality index is obtained by correlating the pedometer step frequency sequence.

3. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 1, characterized in that: The dynamic boundary band is generated, which includes an upper boundary and a lower boundary. The initial values ​​of the upper and lower boundaries of the dynamic boundary band are set according to the preset quantiles of the corresponding behavioral trend features in the historical benchmark patrol data.

4. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 1, characterized in that: The intervention propagation attenuation chain is used to record the probability of path change and contribution reduction coefficient between any two behavioral nodes. The initial value statistical process for the path change probability and contribution reduction coefficient is as follows: the path traversal probability of each pair of preceding and following action nodes under no intervention conditions and after intervention is performed at the preceding action node is calculated from the historical patrol log, and the path change probability and contribution reduction coefficient are calculated and stored in the intervention propagation attenuation chain. When any action node performs an intervention, the reduction of the original contribution of the intervention node to the subsequent action node is accumulated in the form of a product to obtain the real-time effective contribution of the subsequent action node. The calculation method is to multiply the original contribution by the unit value minus the reduction coefficients corresponding to all intervention nodes. A temporal forgetting factor is introduced to correct the intervention propagation decay chain.

5. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 4, characterized in that: The temporal forgetting factor operates as follows: Before updating the real-time effective contribution of subsequent action nodes, the patrol terminal obtains the time difference between the current time and the intervention execution time of each already executed intervention node, and compares this time difference with a preset forgetting period: For intervention nodes whose time difference does not exceed the preset forgetting period, their temporal forgetting factor is set to one, indicating that the loss coefficient of the intervention node does not decay; For intervention nodes whose time difference exceeds a preset forgetting period, the patrol terminal calculates the ratio of the time difference exceeding the preset forgetting period to the preset decay time span to obtain a temporal forgetting factor. The original loss coefficient of the intervention node is multiplied by the temporal forgetting factor to obtain the decayed loss coefficient. The decayed loss coefficient is then used in the calculation of the product.

6. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 1, characterized in that: The cognitive load modulation coefficient of the patrol personnel is calculated as follows: The multi-source data also includes the communication occupancy mapping value, touch interaction frequency, and thermal state factor of the patrol terminal during the standard fusion period. The communication occupancy mapping value, touch interaction frequency, and thermal state factor of the patrol terminal are normalized for their respective contributions. The normalized contribution results are multiplied by the corresponding predefined weights and the product results are accumulated to obtain the preliminary modulation coefficients. The initial modulation coefficient is multiplied by the preset scaling factor to obtain the cognitive load modulation coefficient of the patrol personnel. The upper and lower boundaries of the drift dynamic boundary band and the effective speed of the drift time forgetting factor are determined by the cognitive load modulation coefficient.

7. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 6, characterized in that, Includes the following steps: The upper and lower boundaries of the drifting dynamic boundary band are specifically determined by: the patrol terminal acquiring the current cognitive load modulation coefficient at each sampling time, multiplying the initial upper boundary of the dynamic boundary band by the cognitive load modulation coefficient to obtain the real-time upper boundary, and performing a ratio calculation on the initial lower boundary of the dynamic boundary band by the cognitive load modulation coefficient to obtain the real-time lower boundary. The effective speed of the drift-sequence forgetting factor is specifically determined by: the patrol terminal comparing the current cognitive load modulation coefficient with the load threshold; when the cognitive load modulation coefficient continuously exceeds the load threshold for a preset duration, the preset forgetting period of the time-sequence forgetting factor is temporarily shortened to a short-term threshold, and the ratio of the short-term threshold to the preset forgetting period is used as a scaling factor and multiplied by the decay time span to obtain the temporary decay time span.

8. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 1, characterized in that: The specific process for determining whether the behavioral trend feature crosses the upper boundary of the dynamic boundary zone is as follows: The patrol terminal defines a crossing duration counter for each behavioral trend feature, where the crossing duration counter is used to count the crossing duration of the behavioral trend feature; If the current behavior trend feature value is greater than the real-time upper boundary, it is determined to be in the crossing state, and the corresponding crossing duration counter is incremented by a sampling interval value; otherwise, it is determined not to be in the crossing state, and the corresponding crossing duration counter is cleared to zero. The patrol terminal compares the current value of the traversal duration counter with a time threshold to determine when to intervene.

9. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 8, characterized in that: The determination of the timing of intervention is specifically as follows: When the cumulative value of the continuous crossing counter reaches or exceeds the predefined first time threshold for the first time, the patrol terminal enters the pre-intervention state and starts the confirmation window timer to continue monitoring within the confirmation window; After the confirmation window ends, the patrol terminal continues to monitor the cumulative deviation area feature value: if the cumulative deviation area continues to exceed the real-time upper boundary, and the total duration accumulated from the first crossing reaches the predefined second time threshold, then formal intervention is triggered. Once the formal intervention is triggered, the patrol terminal selects the intervention level and performs dynamic intervention based on the current cognitive load modulation coefficient and the real-time effective contribution.

10. The behavioral path sequence mining application platform based on dynamic intervention timing selection as described in claim 1, characterized in that: The specific process for correcting the intervention propagation attenuation chain based on the effectiveness determination result is as follows: After each formal intervention is executed, the patrol terminal starts an observation window and recalculates the degree of improvement of the behavioral trend characteristics after the intervention at the end of the window. The degree of improvement includes the reduction of the deviation area and the decrease in the percentage of dwell time. If the reduction of the deviation area is greater than the preset improvement threshold or the decrease in the percentage of dwell time is greater than the preset decrease threshold, it is determined to be an effective intervention; otherwise, it is an ineffective intervention. The modified intervention propagation attenuation chain is specifically modified by: if it is determined to be an effective intervention, the patrol terminal reduces the contribution loss coefficient of the behavioral nodes related to the intervention node in the intervention propagation attenuation chain by a first correction step; otherwise, it increases the second correction step.

Images