River valley pollution arrival judgment method based on deep learning

Abstract

Aiming at the problems of transportation time sequence distortion of river valley atmospheric pollutants under the effects of valley wind and geometric bottlenecks and difficulty in judging the arrival time of a downstream urban area, the invention provides an arrival judgment method based on deep learning, a directed river valley topology and a bottleneck stage sequence are taken as a main chain, and an upstream rising edge, a peak front anchor point and a falling anchor point are utilized to judge the arrival time of the downstream urban area. The method comprises the following steps: constructing a time transport flow bridge network constrained by valley wind and bottleneck, realizing one-way mapping and explicit parameterization bottleneck broadening, outputting travel time distribution and a time distortion function of each section, generating a time map of the day and a first arrival candidate, further establishing a bottleneck phase chain clock, forming a stage arrival window capable of realizing online translation and broadening, and obtaining a first arrival window. The last half window of the last phase is extracted as an urban area door window, arrival hours are judged according to an accumulated probability threshold value, pre-deployment and control advance hours are obtained in combination with preparation duration backtracking, and judgment is formed. Physical constraints and causal monotony are considered, the online correction capability is achieved, and the timeliness and accuracy of river valley environment air quality forecasting and early warning can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of environmental monitoring and air pollution early warning, and particularly relates to a valley pollution arrival determination method based on deep learning. BACKGROUND

[0002] The application belongs to the technical field of environmental monitoring and air pollution early warning, and faces air pollution transmission and arrival determination in mountain valleys. Valley topography can guide near-surface wind to flow along the valley axis, and pollutants are transmitted along the path under the action of valley wind. When the valley channel is geometrically contracted to form a bottleneck, the rising flank, peak front and back of the pollution process are stretched or broadened in time, making the actual occurrence time of the downstream (such as the boundary monitoring point of the city) more uncertain. That is, it is difficult to determine the arrival of pollution, and it is impossible to confirm that a certain upstream pollution process reaches and enters the warning concern period at the boundary monitoring point of the city.

[0003] Although the along-river monitoring network can provide upstream concentration time series, under the constraints of valley wind direction and one-way path, how to stably and traceably convert the upstream time series pattern into "executable arrival hours" at the city boundary, and at the same time give the advance preparation time matching the early warning deployment, is the core technical problem of the current valley scene. The time broadening caused by the bottleneck can significantly distort the travel time of the segment and the whole section, increase the uncertainty of arrival determination, and make it difficult to be directly used for action alignment.

[0004] The existing technology mainly adopts the following ideas: numerical model based on hydrodynamic or advection-diffusion equation to calculate transmission time and concentration field; approximate model based on Gaussian plume and empirical wind field to estimate downstream arrival; using multi-station time series correlation / time lag analysis or dynamic time warping to align upstream and downstream signals; approximating travel time by fixed flow rate or segmented parameters and triggering warning by threshold; and statistical or machine learning model based on historical samples to predict downstream arrival time or arrival probability from upstream concentration curve. These methods can provide arrival time or probability reference, but there is still a unified gap in the key link in the valley scene: lack of a phased time determination framework for the along-valley one-way path, unable to explicitly depict the time broadening under the physical conditions of bottleneck constraint and wind direction consistency, and to map the stable upstream pattern to the arrival hours and corresponding advance preparation time at the city boundary.

[0005] Therefore, there is an urgent need for an arrival determination method that combines pattern information and physical constraints, which can not only depict the transmission time of the segment and the whole section, but also output executable results of arrival hours and advance preparation time without violating the causal monotony and segment boundary, providing a stable time reference for environmental air quality early warning and deployment of river valley urban agglomerations. SUMMARY

[0006] An object of the present application is to provide a deep learning-based valley pollution arrival determination method, aiming to convert the rising edge, pre-peak, and falling edge of the upstream monitoring sequence into the arrival hour at the city gate and the corresponding pre-deployment advance hour in a stable and traceable manner while following the one-way mapping and stage boundary constraints in the valley scene dominated by the valley wind and with geometric bottlenecks, and to trigger the deviation at the intermediate station online by shifting or widening the arrival window to maintain the sequence and causality without being destroyed, thereby providing a daily time reference for action alignment.

[0007] According to an embodiment of the present application, a deep learning-based valley pollution arrival determination method comprises:

[0008] S1, receiving a river monitoring sequence and a valley wind, extracting an upstream time profile, generating a directed valley topology and a bottleneck stage sequence;

[0009] S2, constructing a conditional time transport flow bridge network based on the directed valley topology, the bottleneck stage sequence, and the upstream time profile, and anchoring and aligning the rising edge, pre-peak, and falling edge shape point under the one-way mapping and stage expansion factor constraints, and outputting the whole travel time distribution, the segmented travel time distribution, and the time distortion function, and giving the candidate first arrival time, and forming a daily time map;

[0010] S3, establishing a bottleneck phase chain clock according to the segmented travel time distribution and the time distortion function: projecting the upstream rising edge trigger time to the center and width of the first phase determination window, recursively following the bottleneck stage sequence, calibrating the next phase window according to the median and dispersion of each stage distribution, binding the window parameters and the corresponding distribution, generating a stage time table, and solidifying the phase sequence for action alignment;

[0011] S4, detecting the trigger according to the stage time table, adjusting the window according to the segmented travel time distribution if it is not in the window, and generating a city gate window;

[0012] S5, determining the arrival hour and outputting the arrival ruling according to the city gate window and the time distortion function;

[0013] S6, according to the arrival ruling and the stage time table, combining the preparation time parameter, and backtracking to calculate the pre-deployment advance hour, and outputting the ruling pair.

[0014] Optionally, S1 is specifically:

[0015] Taking the river monitoring sequence, the valley wind, and the geometry as input, unifying the time reference and the space reference, selecting the monitoring stations located upstream, determining the upstream boundary and the dominant valley wind direction, and obtaining a data subset for the upstream time profile;

[0016] calibrating the data subset, extracting rising edge, pre-peak, and falling edge of the upstream time profile according to preset threshold, setting the three types of morphological moments as morphological anchor points of the upstream time profile, and forming the upstream time profile;

[0017] According to the valley wind and geometry, calibrate the valley node and link, give unidirectionality according to the dominant valley wind direction, set the link length and sequence relationship, and construct the directed valley topology.

[0018] Based on the directed valley topology and geometry, the geometric contraction index and the valley wind index on the link are jointly thresholded to identify the bottleneck position and limit the upstream link and downstream link corresponding to each bottleneck.

[0019] According to the upstream to downstream sequence of the directed valley topology, the morphological anchor points of the upstream time profile are divided into stages to determine the boundary time and duration of each stage, and the bottleneck stage sequence is generated.

[0020] Optionally, S2 is specifically:

[0021] The directed valley topology, bottleneck stage sequence and upstream time profile are taken as input to construct the mapping structure of the conditional time transport flow bridge network, establish the one-way mapping framework from the upstream time profile to the urban gate time axis, and mark the stage boundary corresponding to the bottleneck in the network;

[0022] Taking the valley wind and bottleneck as conditions, a stage expansion factor is configured for each bottleneck stage to limit the one-way mapping not to backtrack and not to cross the stage boundary, and an explicit parameterization item is established in the network for time expansion at the bottleneck;

[0023] According to the morphological points of rising edge, pre-peak and falling edge extracted from the upstream time profile, anchor alignment is adopted to map the three types of morphological points to the target time of the corresponding stage according to the order of the directed valley topology, and the target time is adjusted by the valley wind and bottleneck conditions;

[0024] A continuous and monotonic time warping function is established between adjacent anchor points, which follows the one-way mapping and stage boundary constraints in each stage, and the influence of time expansion at the bottleneck on the time warping function is weighted by the stage expansion factor;

[0025] Based on the time warping function and the stage expansion factor, the travel time of each stage is calculated to generate the segmented travel time distribution corresponding to each bottleneck stage, and the segmented travel time distribution is bound one by one with the bottleneck stage sequence;

[0026] The segmented travel time distribution is sequentially synthesized according to the directed valley topology to obtain the whole travel time distribution, and the first arrival time candidate is determined according to the preset quantile threshold and the time warping function.

[0027] The whole travel time distribution, the segmented travel time distribution, the time warping function and the candidate of the first arrival time are summarized as the time map of the day.

[0028] Optionally, S3 is specifically:

[0029] The segmented travel time distribution, the time warping function, the sequence of bottleneck phases, the directed valley topology and the upstream time profile are taken as inputs to initialize the phase set of the bottleneck phase chain clock, limit the unidirectional advancement of the phase along the directed valley topology, set the phase boundary correspondence relationship, and receive the candidate of the first arrival time in the time map of the day as a reference contact point;

[0030] The rising edge trigger time of the upstream time profile is projected to the first phase through the time warping function to obtain the center initial value of the first phase, the center initial value is corrected according to the median and dispersion of the segmented travel time distribution corresponding to the phase, and the window width is converted according to a preset proportion coefficient to form the phase arrival window of the first phase;

[0031] The phase arrival window of the first phase is obtained by recursively calculating the window center and window width of each subsequent phase according to the median and dispersion of the segmented travel time distribution corresponding to the current phase and the end time of the phase arrival window of the previous phase, and limiting the start and end times of the window to be not earlier than the end time of the previous phase;

[0032] The phase arrival window of each phase is bound with the corresponding segmented travel time distribution, and the window parameters are corresponded with the phase boundaries to prohibit cross-phase jumping and time backtracking;

[0033] The phase arrival window of each phase is bound with the corresponding segmented travel time distribution, and the window parameters are corresponded with the phase boundaries to prohibit cross-phase jumping and time backtracking;

[0034] The phase arrival window of each phase is bound with the corresponding segmented travel time distribution, and the window parameters are corresponded with the phase boundaries to prohibit cross-phase jumping and time backtracking;

[0035] The phase arrival window of each phase is bound with the corresponding segmented travel time distribution, and the window parameters are corresponded with the phase boundaries to prohibit cross-phase jumping and time backtracking;

[0036] Optionally, S4 is specifically:

[0037] The stage time table and the intermediate station trigger time are taken as inputs, the stage arrival window of the current phase is located, it is judged whether the trigger time is located in the window, the phase is promoted to the next phase when the trigger time is located in the window, and the window is adjusted when the trigger time is not located in the window;

[0038] When the trigger time is not located in the window, the stage arrival window is translated or widened according to the median and dispersion of the segmented travel time distribution corresponding to the current phase, the translation amount and the widening amount are calculated according to the deviation of the trigger time from the center of the window and the preset threshold, and the updated window is limited not to cross the stage boundary;

[0039] When the phase is promoted to the last phase of the bottleneck phase chain clock, the latter half window of the last phase stage arrival window is extracted as the urban gate window, and if the window is adjusted, the urban gate window is determined according to the adjusted window center and window width.

[0040] When the stage arrival window is widened or translated, the stage arrival windows of the subsequent phases along the bottleneck stage sequence are sequentially synchronized, the window center and window width are recalculated according to the end time of the previous phase and the median and dispersion of the segmented travel time distribution corresponding to the current phase, and the phase sequence and time monotonicity are ensured not to be destroyed.

[0041] Optionally, S5 specifically comprises:

[0042] The urban gate window and the time warping function are taken as inputs, the left boundary of the latter half window of the urban gate window is selected as the start point of the latter half window, a preset threshold is set as a judgment threshold according to the cumulative arrival probability of the whole section travel time distribution in the urban gate window, and the one-way mapping in the window is checked by the time warping function.

[0043] The earliest time that meets the condition that the cumulative arrival probability is greater than or equal to the judgment threshold is searched in the urban gate window, and the time is taken as the judgment time, and when no time that meets the condition is found, the start point of the latter half window is taken as the judgment time.

[0044] The judgment time is mapped into an arrival hour, the arrival hour is determined according to the natural hour in which the judgment time is located, the arrival hour is bound with the urban gate window, and the arrival ruling is output.

[0045] When the urban gate window is the adjusted urban gate window, the judgment is performed by using the adjusted window center and window width, and the arrival hour is limited to be not earlier than the natural hour in which the start point of the latter half window is located.

[0046] When multiple judgment time candidates are generated in the urban gate window by the time warping function and the whole section travel time distribution, the judgment time is selected according to the principle that the probability density of the whole section travel time distribution in the window is maximum, and the one-way causal order with the upstream time profile is checked by the time warping function, and when the checking is not satisfied, the judgment time is reselected according to the principle that the probability density is second maximum.

[0047] Optionally, S6 is specifically:

[0048] Taking the arrival decision phase schedule and the preparation duration parameter as inputs, locating the phase corresponding to the arrival decision and the phase arrival window, and extracting the arrival hour as the backtracking end point;

[0049] According to the preparation duration parameter, the pre-control advance hour is calculated from the arrival hour, when the backtracking crosses the phase boundary, the last phase end time of the phase schedule and the remaining duration of the preparation duration parameter are continued to backtrack, and the pre-control advance hour is limited to be not earlier than the first phase start hour of the phase schedule;

[0050] When the arrival decision is adjusted to cause the arrival hour to change, the backtracking calculation is repeated and the pre-control advance hour is updated to keep consistent with the phase schedule;

[0051] The pre-control advance hour and the arrival hour are combined into a decision pair, and the decision pair is bound with the phase schedule;

[0052] When the pre-control advance hour and the arrival hour are located in the same natural hour and are insufficient to complete the preparation duration parameter, the end of the last natural hour is postponed and the pre-control advance hour is recalculated until the pre-control advance hour can meet the preparation duration parameter.

[0053] The beneficial effects of the present application are:

[0054] The present application proposes an improved conditional time transport flow bridge network and time distortion function construction method, which combines the directed valley topology and the bottleneck phase sequence, and anchors and aligns the form anchor (rising edge, pre-peak, falling) under the constraints of one-way mapping and phase boundary, and parameterizes the time expansion at the bottleneck with the phase expansion factor. The improvement lies in controllable weighting of the time sequence stretching caused by the bottleneck, generating segmented and whole travel time distribution and time distortion function, and giving the first arrival time candidate to form the daily time map. While maintaining causal monotony and topological order, the representation consistency and traceability of time distortion under the action of valley geometry and wind field are improved, providing a stable time basis for subsequent determination and scheduling.

[0055] The present proposal proposes a brand-new bottleneck phase chain clock and online deviation adjustment technique, which takes the time-of-day map as input, converts the median and dispersion of the segmented travel time distribution into the center and width of the stage arrival window, recursively binds the stage boundary along the directed topology, and forms the urban gate window after half-window extraction at the last phase; when the intermediate station triggers deviation, only the window is translated or widened, and the subsequent window is synchronized, without rewriting the time warp function and distribution parameters. The method stably converts upstream morphological information into arrival hour judgment basis at the urban gate side, maintains the phase order and unidirectional causality, and has online adaptation capability to cope with uncertain disturbances in river valley transportation, so as to output executable time constraints for field application.

[0056] The present proposal proposes an integrated method of arrival judgment and pre-control backtracking, which determines the arrival time and maps it to the arrival hour according to the urban gate window and the cumulative probability threshold of the whole travel time distribution and checks the unidirectional mapping with the time warp function; then, the pre-control advance hour is backtracked and calculated in combination with the stage schedule and the preparation time length parameter, forming a judgment pair bound with the stage schedule. The whole chain is bound in the order of "time-of-day map-stage schedule-urban gate window-judgment pair", providing a unique time reference for action alignment. Compared with traditional algorithms that only output time or probability, it better meets the needs of river valley scenarios for arrival hour decision and pre-control, while ensuring that the stage boundary is not crossed, the time is monotonic, and the result is reviewable. BRIEF DESCRIPTION OF DRAWINGS

[0057] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, which together with the embodiments of the application, serve to explain the application, and do not constitute a limitation on the application. In the drawings:

[0058] Figure 1 A flowchart of a river valley pollution arrival judgment method based on deep learning proposed by the present application;

[0059] Figure 2 An upstream profile and directed topology construction flowchart of a river valley pollution arrival judgment method based on deep learning proposed by the present application;

[0060] Figure 3 A conditional time transport flow bridge network flowchart of a river valley pollution arrival judgment method based on deep learning proposed by the present application;

[0061] Figure 4 A bottleneck phase chain clock flowchart of a river valley pollution arrival judgment method based on deep learning proposed by the present application;

[0062] Figure 5 An urban gate window and arrival judgment flowchart of a river valley pollution arrival judgment method based on deep learning proposed by the present application;

[0063] Figure 6 A time map along a river valley pollution arrival determination method based on deep learning according to the present application is shown in the following schematic flow chart:

[0064] Figure 7 A bottleneck phase chain clock action alignment flow chart of a river valley pollution arrival determination method based on deep learning according to the present application is shown in the following schematic flow chart:

[0065] Figure 8 A bottleneck expansion before and after comparison schematic of a river valley pollution arrival determination method based on deep learning according to the present application is shown in the following schematic flow chart: DETAILED DESCRIPTION

[0066] The present application will now be further described in detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams and only show the basic structure of the present application in a schematic manner, and thus only show the components related to the present application.

[0067] REFERENCE Figures 1 to 8 A river valley pollution arrival determination method based on deep learning, comprising:

[0068] S1, receiving a river monitoring sequence and a valley wind, extracting an upstream time profile, generating a directed valley topology and a bottleneck stage sequence;

[0069] S2, constructing a conditional time transport flow bridge network based on the directed valley topology, the bottleneck stage sequence and the upstream time profile, under the constraints of one-way mapping and stage expansion factor, anchoring and aligning the rising edge, pre-peak and falling edge shape points with the valley wind and the bottleneck as conditions, explicitly parameterizing the time expansion at the bottleneck, outputting the whole section travel time distribution, the section travel time distribution and the time distortion function, and giving the candidate time of first arrival, forming the time map of the day;

[0070] S3, establishing a bottleneck phase chain clock according to the section travel time distribution and the time distortion function: projecting the upstream rising edge trigger time to the center and width of the first phase determination window, recursively along the bottleneck stage sequence, calibrating the next phase window according to the median and dispersion of each stage distribution, and binding the window parameters and the corresponding distribution, generating a stage time table, and solidifying the phase order for action alignment;

[0071] S4, detecting the trigger according to the stage time table, adjusting the window according to the section travel time distribution when it is not in the window, and generating the urban gate window;

[0072] S5, determining the arrival hour and outputting the arrival decision according to the urban gate window and the time distortion function when the threshold is reached or the second half of the window is reached;

[0073] S6, according to the arrival of the decision-making and stage schedule, combined with the preparation of the length parameter, backtracking calculation of pre-control hours, output decision-making.

[0074] In this embodiment, step S1 is specifically:

[0075] The along-river monitoring sequence is denoted as , where is the monitoring station index, and the along-valley wind is denoted as , the geometry is denoted as , including the center line, cross section and elevation, a unified time reference is established , a fixed step is set , a spatial reference is established , the center line of the geometry is taken as a reference to generate a one-dimensional arc length coordinate, which is used for linear ordering and distance calculation of the monitoring station position;

[0076] , , and are taken as inputs, and is aligned with and , and is projected into , and the dominant along-valley wind direction is calculated based on the cumulative amount of the along-valley component on ; ; ;

[0077] The minimum position of in the direction is found to determine the upstream boundary , the monitoring stations that meet and have complete signal coverage are screened to form a data subset for the upstream time profile , and and , , are taken as inputs for subsequent steps;

[0078] The baseline correction and time alignment processing are performed on corresponding to , the baseline is determined as the mean value of the stable interval before triggering, and the preset threshold , , and the first-order difference threshold are set, the rising edge triggering time is determined by using the sliding window and the first-order difference criterion, the condition is that the window mean value exceeds the baseline plus and the first-order difference is continuously positive within the window, the stable platform before the local maximum value is taken as the pre-peak platform, and the criterion is that the first-order difference absolute value is lower than​ And the amplitude stability meets To return the signal to the baseline Furthermore, the fact that the first-order difference remains negative indicates a pullback pattern. These three types of pattern moments are then combined into a set of pattern anchor points. ,Will Inside By weight Perform weighted aggregation, with weights based on... relatively The distance and monitoring stability were normalized to obtain the upstream time profile. and will Bound to ;

[0079] Based on valley winds and geometry, valley nodes and links are identified, in order to Centerline sampling key locations generate valley node set This includes the location of monitoring stations and locations of significant geometric changes, and constructs a set of links between adjacent nodes. And calculate the link length as the difference in arc length between adjacent nodes, based on for Assigning unidirectionality, link sequence according to Incremental determination, forming a directed valley topology. ,exist The system records the start and end nodes, length, and sequence index of each link to ensure subsequent stage division in the unidirectional upstream to downstream direction;

[0080] Based on the directed valley topology and geometry, the geometric contraction index and valley wind index of the link are calculated. Defined as the normalized contraction rate of the minimum cross-sectional width sequence within a link relative to the mean of the previous upstream link, along the valley wind index. Defined as exist The time-averaged projection intensity of the valley component on the link is used to set a joint threshold. and ,when and In real time, bottleneck locations are identified. For each identified bottleneck, the corresponding upstream link is defined as the link preceding the current link, and the corresponding downstream link is defined as the link following the current link. The index relationship between the bottleneck and adjacent links is recorded in the middle;

[0081] Press the bottleneck location The bottlenecks are arranged sequentially from upstream to downstream, generating an ordered set. Combined with the morphological anchor points of the upstream time profile, the spatial segments between the bottlenecks are... Corresponding to the time periods in the text, using In 、 、 As a reference of nominal boundary time, the link order of determines the boundary time for each bottleneck stage, and the difference between adjacent boundary times determines the duration time, the spatial range, the nominal boundary time and the duration time of each bottleneck stage form a record unit, which is sequentially concatenated into a bottleneck stage sequence and is indexed and bound with the directed valley topology, which is used for subsequent construction of conditional time transport flow bridge network and calculation of projection to phase window under the constraints of one-way mapping and stage boundary. The upstream time profile, the directed valley topology and the bottleneck stage sequence formed by the above steps are consistent in time and space reference, providing a unique input reference and stage order constraint for the subsequent steps.

[0082] In this embodiment, step S2 is specifically:

[0083] This step constructs a conditional time transport flow bridge network to generate a daily time map. The directed valley topology is denoted as , the bottleneck stage sequence is denoted as , the upstream time profile is denoted as , the set of morphological anchor points of the upstream time profile is denoted as , the city gate time axis is denoted as , the , and are taken as inputs to build the mapping structure of the conditional time transport flow bridge network, the input layer is set as the morphological anchor points and the stage conditions, the stage transformation layer is set as the bottleneck constraints and the stage expansion factor, the output alignment layer is set as the time warp function and the travel time distribution, and the mapping structure is sequentially labeled according to The stage boundaries on are sequentially labeled, the one-way mapping framework from to is established, and the index relationship between each stage and the corresponding bottleneck is recorded;

[0084] Taking the valley wind and the bottleneck as the conditional input, the geometric contraction index and the valley wind index on the stage are extracted, and the stage expansion factor is calculated. The stage expansion factor is used to control the time expansion at the bottleneck and the local slope of the time warp function within the stage. To meet the constraints of one-way mapping and not crossing the stage boundary, the lower limit and the upper limit of the stage expansion factor are set, and the stage index in is bound. The calculation of the stage expansion factor uses a single explicit expression, which ensures that the expression only depends on the directly available geometric contraction index and valley wind index, and can be uniformly applied to all stages.

[0085] According to the morphological anchor points of the upstream time profile, the anchor alignment is performed to project to The target time as the starting point of the time warp function And In Two target times are formed in the first stage, and the local warp interval is determined. For subsequent stages , the target end of the previous stage is taken as the starting point, and the three target times of the current stage are calculated in turn, and the target time is aligned to the inside of the stage boundary. The target time is prohibited to fall outside the boundary of the adjacent stage. When the time period of the valley wind appears in the direction inconsistent with , the advance of the target time is frozen in this time period, and the alignment is continued only after the valley wind resumes in the direction consistent with , ensuring unidirectionality;

[0086] After the anchor point alignment is completed, the time warp function is constructed, and the three target times in each stage are taken as the segmentation nodes. The continuous and monotonic time warp function segment is generated between adjacent target times using a segmented monotonic spline, and the first-order continuous splicing is performed at the boundary of each segment. The stage expansion factor is applied to the local slope and local time length of the corresponding segment, so that the time expansion at the bottleneck is explicitly written into the time warp function. If there is a risk of segment overlap or backtracking due to the stage expansion factor, the stage expansion factor is reduced in the pre-set order within the stage until the backtracking is removed and the adjacent segments are only tangent at the stage boundary without intersection.

[0087] The time warp function is used to generate a segmented travel time distribution. For each stage , the difference between the stage entry time and the exit time on the time warp function is taken as the center parameter of the travel time, and the stage expansion factor is used to amplify or converge the stage dispersion parameter to obtain the center and dispersion of the segmented travel time distribution. The segmented travel time distribution is bound with the stage index , and its relationship with the corresponding bottleneck is recorded. The whole segmented travel time distribution is composed of the stage order synthesis strategy. The center parameter is synthesized according to the addition rule, and the dispersion parameter is synthesized according to the square sum rule, so that the whole travel time distribution is consistent with the segmented travel time distribution in the parameter level, and the traceability is maintained.

[0088] When the segmented travel time distribution is updated in a certain stage due to observation deviation, only the center and dispersion are recalculated in the stage, and the entry time and exit time of the subsequent stage are sequentially updated, keeping the unidirectional mapping and stage boundary constraint unchanged, without rewriting the time warp function in the confirmed segment. If the update causes the boundary between stages to be close, the stage expansion factor is reduced in the pre-set order within the close stage to avoid cross-stage jumping.

[0089] Based on the overall travel time distribution and the time warp function, candidate first arrival times are calculated. Using a preset quantile threshold as the criterion, the earliest time satisfying the threshold is searched across the entire travel time distribution, and its accuracy is verified using the time warp function. If the unidirectional causal relationship fails the verification, the quantile threshold is sequentially increased until the verification is passed, or the threshold is rolled back to the minimum value of the stage expansion factor and the target time of the stage is recalculated to ensure that the candidate first arrival time is consistent with the unidirectional mapping.

[0090] The overall travel time distribution, segmented travel time distribution, time warp function, and candidate first arrival times are summarized into a daily time map. This daily time map is indexed by stage order and includes the central parameter, dispersion parameter, stage expansion factor, and anchor point target time for each stage, along with the city gate time axis. The continuous expression of the time warp function is given above. The time map of the day is used as the only time series input for the subsequent generation of the bottleneck phase chain clock and the arrival hour determination, ensuring that the time series reference is consistent between the upstream time profile, the directed valley topology and the bottleneck stage sequence.

[0091] To ensure that the stage expansion factor has clear computability and operability, the following unique formula is used to determine each stage. Stage expansion factor:

[0092] ;

[0093] in, For stage expansion factor, and These are the upper and lower bounds of the stage expansion factor. and These are the weighting coefficients. For the stage Geometric contraction index, For the stage The time average of the valley wind index, To reference the valley wind intensity, the local slope of the time warp function and the dispersion of the segmented travel time distribution within the stage are uniformly scaled using the stage expansion factor, so that the time broadening at the bottleneck is explicitly parameterized, and the one-way mapping and stage boundary constraints are not violated in any stage.

[0094] The above implementation ensures that the conditional time transport flow bridge network generates a time map of the day under the premise of consistent input, clear boundaries and single parameters, and provides an executable timing basis for subsequent steps.

[0095] In this embodiment, step S3 specifically includes:

[0096] establishing the bottleneck phase chain clock, and converting the time-of-day map into an executable phase schedule, denoted as , the bottleneck phase sequence is denoted as , the parameters of the piecewise travel time distribution over phase are denoted as median and dispersion , the time warping function is denoted as , the rising edge trigger time of the upstream time profile is denoted as , the first-arrival time candidate in the time-of-day map is denoted as , the phase set is defined as , each phase corresponds to a phase arrival window , the center of which is denoted as , and the width is denoted as , the time boundaries for each phase are set as and , and correspond to the spatial boundaries of the bottleneck phase sequence;

[0097] Based on the piecewise travel time distribution, the time warping function, the bottleneck phase sequence, the directed river valley topology, and the upstream time profile, the phase set is initialized, the phase index is established in the order of the bottleneck phase sequence on the directed river valley topology, the phases are unidirectionally advanced from upstream to downstream, and is taken as the reference touch point, which is only used to check that the center of the first phase is not earlier than the time-of-day earliest arrival candidate, to prevent early triggering under weak wind or observation jitter;

[0098] The phase arrival window of the first phase is calculated, and is projected to the urban gate time axis by the time warping function to obtain the initial value of the center of the first phase, and then the center and the width are converted according to the median and the dispersion of the piecewise travel time distribution of the first phase to obtain , when the initial value of the center is earlier than and the deviation exceeds the preset threshold, the initial value of the center is replaced by and the center and the width are reconverted , to ensure that the first phase is consistent with the time-of-day map Figure 1 ;

[0099] For subsequent phases, the phase arrival window is determined in a recursive manner, the end time of the previous phase is taken as the starting reference of the current phase, the median and the dispersion of the current phase are converted into the center and the width of the window, and is limited to not be earlier than and not to cross the phase boundary , to avoid window overlap, is limited to be not less than , and if necessary, the convergence is prioritized to the boundary allowed range, then sequentially translate , maintain time monotonicity;

[0100] For the window calculation of unifying the first phase and the recursive phase, the following unique formula is used to generate the full-link phase arrival window:

[0101] ;

[0102] wherein, is the window center of phase , is the window width, is the window start, and is the window end, and is the phase boundary of phase , is the median of the segment travel time distribution, is the dispersion of the segment travel time distribution, and is the preset proportion coefficient, is the time warping function, which is guaranteed by the above calculation and , , , so as to solidify the one-way advance and the phase boundary constraint,

[0103] Bind the phase arrival window of each phase with the corresponding segment travel time distribution, and correspond the window parameters with the phase boundaries. Record the phase index , window parameters , segment travel time distribution parameters and phase boundaries in the binding relationship, prohibit cross-phase jumping and time backtracking, generate the phase timetable according to the bottleneck phase sequence and the directed river valley topology solidification sequence of the phase arrival window of all phases , and take the phase timetable as the unique time reference of action alignment;

[0104] Configure online deviation adjustment rules for the phase timetable, record the intermediate station trigger time as , and record the deviation threshold as , wherein is the preset proportion coefficient, when , shift leftward as a whole to contain and satisfy , when , preferentially shift rightward to contain , if the trigger time touches after the shift, fix and minimum amplitude stretching by containing , the upper limit of the stretching does not exceed The above adjustment does not rewrite the parameters of the time warp function and the segmented travel time distribution, only updates on the execution side When is adjusted, the phase of is recalculated in sequence and , keep and ;

[0105] In each phase, the boundary of the second half window is calibrated, and the second half window is defined as , and it is bound with For the last phase , take as the pre-constraint interval of the urban gate window, which is used for window generation of S4 and arrival hour judgment of S5. Through the above process, the bottleneck phase chain clock completes the conversion from the time map of the day to the stage timetable, realizes the one-way time sequence control driven by the segmented travel time distribution, projected by the time warp function, and executed by the phase sequence, and ensures that the urban gate window has unique resolvability and executability in phase advancement;

[0106] In this embodiment, step S4 is specifically:

[0107] In the execution side of the bottleneck phase chain clock, trigger judgment, window adjustment and urban gate window generation are completed. The stage timetable is recorded as , where the center of the stage arrival window is , and the width is The parameters of the segmented travel time distribution on the stage are recorded as the median and the dispersion The stage boundary is recorded as and The intermediate station trigger time is recorded as Take the earliest phase that has not completed the advancement as the current phase, locate and determine whether is located in , if it is located in the window, advance the phase to , and take as the starting reference of the next phase, if it is not located in the window, enter the window adjustment process;

[0108] In the window adjustment, the deviation amount is calculated first, the deviation threshold is set as , and ​The translation and expansion of the window are given by the product of the window width and the preset proportion coefficient , the magnitude of translation and expansion is determined by the unified adjustment amount calculation formula, and the scale reference is provided by the median and dispersion of the segmented travel time distribution, and the adjustment is triggered only when the deviation exceeds the threshold value:

[0109] ;

[0110] ;

[0111] wherein, is the translation amount of the window center, is the expansion amount of the window width, is the deviation amount of the trigger time relative to the window center, is the deviation threshold, is the threshold proportion coefficient, and is the preset proportion coefficient of translation and expansion, is the sign function, is the non-negative truncation function, and the updated center and width are obtained according to and , and the constraint clipping is performed: the calculation and is limited , , when is reached, the is reduced preferentially, when or is reached, the is reduced preferentially, and if all the constraints cannot be met simultaneously, the order of translation first and then expansion is followed to reduce until all the constraints are met, if is less than the minimum allowed width converted from the dispersion , the is raised to the preset lower limit and the is kept not less than ;

[0112] When is located within and the deviation does not exceed , the window is not adjusted, and the phase is directly advanced, when appears multiple times and across , they are processed in time sequence one by one, ensuring that each time only the current phase is updated once, and backtracking across phases is prohibited, after completing the window update of the current phase, the is generated and replaces , and then the sequential synchronization is entered: for the phase of the phase, using the generation rule of the stage time table, taking the initial reference, combining the segmented travel time distribution of the phase , , , , and limiting , , when the boundary conflict occurs, converging first, and then sequentially translating , until the one-way advance and the stage boundary correspondence are met.

[0113] When the phase advances to the last phase of the bottleneck phase chain clock, the second half window is directly extracted from as the urban gate window, if the last phase window has been translated or widened on the execution side, then and are replaced by and , generating as the urban gate window, and binding it with the segmented travel time distribution parameters of the last phase, to avoid conflicts, if the start point of the urban gate window is later than is not enough to form the second half window, then converging to the minimum amplitude in the last phase first, and then moving to the right without crossing , to ensure the existence and uniqueness of the second half window.

[0114] In the loop execution, for each newly arrived intermediate station trigger time, repeat the positioning, judgment, adjustment and synchronization process described above, until the urban gate window is determined, and only the window parameters are updated on the execution side during the entire process, without rewriting the parameters of the time warping function and the segmented travel time distribution, to ensure that the phase order of the stage time table and the time monotonicity are not destroyed, and to provide a unique and stable time interval input for subsequent arrival hour judgment according to the urban gate window.

[0115] In this embodiment, step S5 is specifically:

[0116] Forming an arrival decision on the urban gate time axis, recording the urban gate window as , when the window is not adjusted , when the window is the adjusted urban gate window , recording the start point of the second half window as , whose value is the left boundary of , recording the entire travel time distribution as , which is defined on the urban gate time axis , recording the limited form of the cumulative arrival probability function within as Let the time warp function be denoted as The preset quantile threshold is denoted as The upstream time profile is denoted as Mapping natural hours as This is used to assign any city gate time to the corresponding natural hour interval;

[0117] based on and Establish candidate search intervals, As the starting point for the search, The right boundary is used as the search endpoint, forming the search range of the second half of the window. ,exist Internal By performing limited accumulation, we obtain This cumulative total only includes statistics. Internal probability quality, and in left end Zero-based benchmarking is performed using a fixed time step. Perform discrete traversal with a time step no larger than the time resolution of the stage timetable to ensure synchronization with the preceding steps.

[0118] in accordance with and The initial selection of the execution decision time, along The search from left to right satisfies Not less than The earliest time, let this time be recorded as ,when If there is no time in the interval that satisfies the condition, then Simply record as Called in each search iteration Verify the one-way mapping, pass The corresponding time point on the upstream timeline is compared with an established upstream propagation reference, which includes at least... If the rising edge trigger time and the upstream nominal boundary time that the stage timetable has advanced to, If the corresponding upstream time is earlier than this reference, then in The process proceeds sequentially to the next time point, repeating the accumulation and verification until the unidirectional property is satisfied. Or return to use ;

[0119] Will Mapped to arrival hours, for application ,get The natural hour in which it occurs is denoted as ,Will and Binding is performed to form the basis for reaching the judgment, when When adjusting the city's entrance and exit windows, the above applies. calculate and and will The restriction is no earlier than If the natural hour is If it is earlier than that natural hour, then The execution will be postponed to that calendar hour and re-executed within that calendar hour. Cumulative searches within;

[0120] when Internally and Together they generate multiple satisfactions Not less than When determining the candidate time, a secondary screening sequence is constructed, first in The position with the highest probability density within the time frame is selected as the primary candidate. If multiple candidates with the same probability density exist, the earlier one is selected as a temporary candidate. The temporary candidate is then processed... Projected onto the upstream timeline, and The morphological anchor sequence is checked for causal order, requiring that the projection time is no later than the upstream end reference of the executed phase and no earlier than the rising edge trigger reference of the upstream time profile. If the check fails, temporary candidates are replaced in descending order of probability density and the check is repeated until a causal order is satisfied. If all candidates fail, then the following will be adopted. As And based on this, determine ;

[0121] In the output phase, and The data is combined into an arrival decision, and internally recorded for generating the arrival decision. , and When subsequent execution is caused by a new trigger When adjusted again, repeat the above search, verification, and mapping process, using only the latest version. For now, I will not write back. and The parameters are kept consistent with the unidirectional advancement of the stage timetable. Through the above process, under the premise of constraining the second half window, cumulative arrival probability and unidirectional mapping, a unique arrival hour and a bound city gate window are generated, forming an arrival decision that can be executed directly.

[0122] In this embodiment, step S6 specifically includes:

[0123] Record the arrival of the ruling as ,in To reach the hour, For the city's gateway, the timetable for each phase is recorded as follows: ,in For the stage arrival window, the preparation time parameter is denoted as... The time granularity adopts Mapping natural hours as Its derived hour start and end functions and Define the starting hour of the first phase. In this implementation, the timetable for each stage and the parameters of the city gate windows are not rewritten; only the advance deployment hours are calculated on the execution side.

[0124] Based on the arrival decision, phase timetable, and preparation time parameters, locate the phase and phase arrival window corresponding to the arrival decision, and then... As the endpoint of backtracking, if lie in If internal, then let the current phase index be... ,when lie in and During the interval, the current phase index is called back to the position containing... The phase ensures that the backtracking starts from the actual landing point and extracts... As the endpoint hour of the retrospective, incorrect. Adjust to the right to avoid disrupting unidirectional propulsion;

[0125] Perform segmented backtracking on the stage timetable based on the preparation time parameter, assuming the remaining time... The initial value is and the cursor The initial value is set to In the current phase Internal computation can be backtracked for a period of time. and The difference between them, if the difference is not less than Then in phase The internal backtracking point is obtained And stop backtracking if the difference is less than Then deduct the difference. ,Will Set as The phase index decreases to Continuing to trace back, when Decrease to and Still insufficient to cover At that time, Pincer , and set to zero, and mapped to natural hours, resulting in pre-deployment advance hours , and limiting ;

[0126] When the arrival decision is adjusted on the execution side resulting in changes, repeat the above piecewise backtracking process, new replace old values, reposition the starting phase and calculate and , if the phase schedule has been updated in the order synchronization at S4 or , then keep consistent with the phase order using the updated as the only basis;

[0127] When and are in the same natural hour and the remaining duration from to in that hour is less than , then forward the hour boundary, move to the left to , use that time as the new backtracking end point, recalculate , if still not enough to cover , continue to forward to the end of an earlier natural hour and repeat the piecewise backtracking until the preparation duration parameter is met, the above forward always follows the limit of , not entering before the first phase starting hour;

[0128] After completing the backtracking, generate the decision pair, combine the pre-deployment advance hours with the arrival hours into a decision pair, and bind it with the phase index of the phase schedule, so that the execution side can complete the preparation before and action alignment within , if is the adjusted city gate window, then still use in the arrival decision as the backtracking end point, without making any modifications to the parameters of the time warp function and the piecewise travel time distribution;

[0129] Through the above process, the pre-deployment advance hours are uniquely determined within the boundaries of the phase schedule, and the one-way advancement and the constraint of not being earlier than the first phase starting hour are maintained in the cross-phase and cross-hour backtracking, thus forming a directly executable decision pair for deployment and response.

[0130] The above merely describes preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art, according to the technical solution and inventive concept of the present application, makes equivalent replacement or change within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for determining the arrival of pollution in a river valley based on deep learning, characterized by, The method comprises the following steps: S1, receiving the river monitoring sequence and the valley wind, geometry, extracting the upstream time profile, generating the directed river valley topology and the bottleneck stage sequence; S2, based on the directed river valley topology, the bottleneck stage sequence and the upstream time profile, constructing a conditional time transport flow bridge network, under the constraints of one-way mapping and stage expansion factor, taking the valley wind and the bottleneck as conditions, anchoring and aligning the rising edge, peak front and falling shape points, explicitly parameterizing the time expansion at the bottleneck, outputting the whole section travel time distribution, the section travel time distribution and the time distortion function, and giving the candidate first arrival time, forming the time map of the day; S3, according to the section travel time distribution and the time distortion function, establishing the bottleneck phase chain clock: projecting the upstream rising edge trigger time to the center and width of the first phase determination window, recursively along the bottleneck stage sequence, calibrating the next phase window according to the median and dispersion of each stage distribution, binding the window parameters and the corresponding distribution, generating the stage time table, and fixing the phase order for action alignment; S4, detecting the trigger according to the stage time table, adjusting the window according to the section travel time distribution if it is not in the window, and generating the urban gate window; S5, according to the urban gate window and the time distortion function, when reaching the threshold or the second half of the window, determining the arrival hour and outputting the arrival decision; S6, according to the arrival decision and the stage time table, combined with the preparation time parameter, backtracking to calculate the pre-control advance hour, and outputting the decision pair. 2.The valley pollution arrival determination method based on deep learning according to claim 1, wherein, S1 is specifically: Taking the river monitoring sequence and the valley wind, geometry as input, unifying the time reference and the space reference, selecting the monitoring stations located in the upstream, determining the upstream boundary and the dominant valley wind direction, and obtaining the data subset for the upstream time profile; Calibrating the data subset, extracting the rising edge, peak front and falling shape time of the upstream time profile according to the preset threshold, setting the three types of shape time as the shape anchor points of the upstream time profile, and forming the upstream time profile; According to the valley wind and the geometry, the river valley nodes and links are calibrated, the one-way direction is given according to the dominant valley wind direction, the link length and the sequence relationship are set, and the directed river valley topology is constructed; Based on the directed river valley topology and the geometry, the joint threshold judgment of the geometric contraction index and the valley wind index on the link is carried out, the bottleneck position is identified, and the upstream link and the downstream link corresponding to each bottleneck are limited; The bottleneck position is divided into stages according to the upstream to downstream order of the directed river valley topology combined with the shape anchor points of the upstream time profile, the boundary time and the duration of each stage are determined, and the bottleneck stage sequence is generated. 3.The valley pollution arrival determination method based on deep learning according to claim 1, wherein, S2 is specifically: Taking the directed river valley topology, the bottleneck stage sequence and the upstream time profile as input, constructing the mapping structure of the conditional time transport flow bridge network, establishing the one-way mapping framework from the upstream time profile to the urban gate time axis, and marking the stage boundary corresponding to the bottleneck in the network; Taking the valley wind and the bottleneck as conditions, configuring the stage expansion factor for each bottleneck stage, limiting the one-way mapping not to backtrack and not to cross the stage boundary, and establishing the explicit parameterization item for the time expansion at the bottleneck in the network; According to the rising edge, the front peak and the falling edge of the upstream time profile, the three types of morphological points are mapped to the target time of the corresponding stage in the order of the directed valley topology by using anchor alignment, and the target time is adjusted by the along-valley wind and the bottleneck condition; A continuous and monotonic time warp function is established between adjacent anchor points, the time warp function follows the one-way mapping and stage boundary constraints in each stage, and the influence of the time expansion at the bottleneck on the time warp function is weighted by the stage expansion factor; Based on the time warp function and the stage expansion factor, the travel time of each stage is calculated, the segmented travel time distribution corresponding to each bottleneck stage is generated, and the segmented travel time distribution and the bottleneck stage sequence are bound one by one; The segmented travel time distribution is sequentially synthesized according to the directed valley topology to obtain the whole travel time distribution, and the first arrival time candidate is determined according to the preset quantile threshold and the time warp function; The whole travel time distribution, the segmented travel time distribution, the time warp function and the first arrival time candidate are summarized as the daily time map. 4.The valley pollution arrival determination method based on deep learning according to claim 1, wherein, S3 specifically is: The segmented travel time distribution, the time warp function, the bottleneck stage sequence, the directed valley topology and the upstream time profile are taken as inputs to initialize the phase set of the bottleneck phase chain clock, limit the one-way advancement of the phase along the directed valley topology, set the stage boundary correspondence, and receive the first arrival time candidate in the daily time map as a reference touch; The rising edge trigger time of the upstream time profile is projected to the first phase through the time warp function to obtain the center initial value of the first phase, the center initial value is corrected by a preset proportion coefficient according to the median and dispersion of the segmented travel time distribution corresponding to the phase, and the window width is converted to form the stage arrival window of the first phase; Along the bottleneck stage sequence, for each subsequent phase, the window center and window width of the current phase are calculated according to the end time of the stage arrival window of the previous phase and the median and dispersion of the segmented travel time distribution of the current phase, and the window start and end are limited to be not earlier than the end time of the previous phase to obtain the stage arrival window of the current phase; The stage arrival windows of each phase are bound with the corresponding segmented travel time distribution, and the window parameters are bound with the stage boundaries to prohibit cross-stage jumping and time backtracking; The stage time table is generated by solidifying the phase order according to the bottleneck stage sequence and the directed valley topology, and the stage time table is taken as the only time reference for action alignment; The online deviation adjustment rule is configured for the stage time table, when the trigger time of the intermediate station located on the link of the directed valley topology deviates from the current stage arrival window by more than a preset threshold, the corresponding segmented travel time distribution is called to expand or translate the window, and the parameters of the time warp function and the segmented travel time distribution are not rewritten; In each phase, the second half window boundary is marked, and the parameters of the second half window and the stage arrival window are bound as the pre-clock constraint for the city gate window generation and arrival hour determination. 5.The valley pollution arrival determination method based on deep learning according to claim 1, wherein, S4 specifically is: The stage schedule and the intermediate station trigger time are taken as inputs to locate the stage arrival window of the current phase, to determine whether the trigger time is located in the window, to advance the phase to the next phase when the trigger time is located in the window, and to enter window adjustment when the trigger time is not located in the window; When the trigger time is not located in the window, the stage arrival window is shifted or widened according to the median and dispersion of the segmented travel time distribution corresponding to the current phase, the shift amount and the widening amount are calculated according to the deviation of the trigger time from the center of the window and a preset threshold, and the updated window is limited not to exceed the stage boundary; When the phase is advanced to the last phase of the bottleneck phase chain clock, the latter half window of the last phase stage arrival window is extracted as the urban gate window, and if the window is adjusted, the urban gate window is determined according to the adjusted window center and window width; When the stage arrival window is widened or shifted, the stage arrival windows of the subsequent phases along the bottleneck stage sequence are sequentially synchronized, the window center and window width are recalculated according to the end time of the previous phase and the median and dispersion of the segmented travel time distribution corresponding to the current phase, and the phase sequence and time monotonicity are ensured not to be destroyed. 6.The valley pollution arrival determination method based on deep learning according to claim 1, wherein, S5 specifically is: Taking the urban gate window and the time warping function as inputs, the left boundary of the latter half window of the urban gate window is selected as the start point of the latter half window, a preset threshold is set as a determination threshold according to the cumulative arrival probability of the whole segment travel time distribution in the urban gate window, and the one-way mapping in the window is checked by the time warping function; The earliest time that meets the condition that the cumulative arrival probability is greater than or equal to the determination threshold is searched in the urban gate window, and the time is taken as the determination time. When no time meeting the condition is found, the start point of the latter half window is taken as the determination time; The determination time is mapped to the arrival hour, the arrival hour is determined according to the natural hour in which the determination time is located, and the arrival hour is bound with the urban gate window, and the arrival ruling is output; When the urban gate window is the adjusted urban gate window, the determination is made using the adjusted window center and window width, and the arrival hour is limited not to be earlier than the natural hour in which the start point of the latter half window is located; When multiple determination time candidates are generated in the urban gate window by the time warping function and the whole segment travel time distribution, the determination time is selected according to the principle that the probability density of the whole segment travel time distribution in the window is maximum, and the one-way causal order with the upstream time profile is checked by the time warping function. If it does not meet the condition, it is reselected according to the second largest principle. 7.The valley pollution arrival determination method based on deep learning according to claim 1, wherein, S6 specifically is: Taking the arrival ruling, the stage schedule and the preparation time length parameter as inputs, the phase corresponding to the arrival ruling and the stage arrival window are located, and the arrival hour is extracted as the backtracking end point; The preparation time length parameter is used to backtrace and calculate the pre-control advance hour from the arrival hour. When the backtracking crosses the phase boundary, the remaining time length of the preparation time length parameter is used to continue backtracking according to the end time of the previous phase in the stage schedule, and the pre-control advance hour is limited not to be earlier than the start hour of the first phase in the stage schedule; When the arrival ruling is adjusted to change the arrival hour, the pre-control advance hour is recalculated and updated by repeating the backtracking to keep consistent with the stage schedule; The pre-control advance hour and the arrival hour are combined into a ruling pair, and the ruling pair is bound with the stage schedule. When the pre-control advance hour and the arrival hour are in the same natural hour and are not enough to complete the preparation duration parameter, the calculation is postponed to the end of the previous natural hour and retraced until the pre-control advance hour can meet the preparation duration parameter.

Images