Dynamic digital flight rule construction method based on standardized airspace grid

Abstract

The invention relates to the technical field of urban air traffic airspace management, and discloses a standardized airspace grid-based dynamic digital flight rule construction method, which comprises the following steps of: encoding novel aircraft types and existing various types of structured description documents by utilizing a pre-training text encoder; generating a new type semantic feature vector and an existing type semantic feature vector set; pairwise interaction feature vectors are calculated based on the interval values of the existing type pairs and the performance parameters, and pairwise interval inference functions are generated through multiple linear regression fitting. According to the method, the pairwise interaction feature vectors of the double aircrafts serve as inference input and are fitted into the pairwise interval inference function, so that the technical effects that the exclusive interval value of the type group can still be inferred when no new type item exists in the heterogeneous aircraft interval matrix, and classification conflict detection is prevented from being degraded into using a default conservative interval value are achieved.

Description

Technical Field

[0001] This invention relates to the field of urban air traffic airspace management technology, and more specifically, to a method for constructing dynamic digital flight rules based on a standardized airspace grid. Background Technology

[0002] In the field of urban air traffic airspace management, various heterogeneous aircraft, such as fixed-wing manned aircraft, multi-rotor unmanned delivery aircraft, eVTOL manned aircraft, and tethered drones, are already operating in the same gridded airspace. Existing management schemes determine the specific safety interval values ​​between each type pair based on a pre-set heterogeneous aircraft spacing matrix. Potential conflict pairs are identified through classification-based spatiotemporal collision detection. Then, a type-aware conflict severity score is calculated using a collision avoidance maneuverability weighted factor. Finally, differentiated flight rules are generated for each aircraft type group and aggregated along a multi-level grid structure.

[0003] When a completely new type of aircraft is introduced into the airspace for the first time, the existing technology has the following shortcomings: First, the heterogeneous aircraft spacing matrix does not contain spacing value entries between the new aircraft type group and any existing type group. Classification conflict detection cannot query a specific safe spacing value for aircraft pairs involving the new type, causing the detection of these conflict pairs to degenerate into using default conservative spacing values, which excessively restricts the operational density of the new aircraft. Second, the existing zero-sample transfer framework establishes a one-way mapping relationship from the operational characteristics of a single aircraft to the rule parameters, while the safe spacing standard depends on the performance interaction between the conflicting parties. The operational characteristics of a single aircraft cannot independently determine the pairwise spacing value. The existing zero-sample transfer framework does not have the ability to handle pairwise relationship inference, resulting in a structural defect in pairwise spacing inference due to the simple superposition of the two technologies. Summary of the Invention

[0004] This invention provides a method for constructing dynamic digital flight rules based on a standardized airspace grid, which solves the technical problems in related technologies such as the lack of spacing rules for heterogeneous aircraft when new aircraft are accessed, inaccurate conflict risk assessment, and difficulty in automatically generating differentiated flight rules.

[0005] This invention provides a method for constructing dynamic digital flight rules based on a standardized airspace grid, comprising the following steps: By using a pre-trained text encoder to encode the structured description documents of the new aircraft type and the existing types, a set of semantic feature vectors for the new type and the set of semantic feature vectors for the existing types is generated. Based on the interval values ​​and performance parameters of existing type pairs, the pairwise interaction feature vector is calculated, and the pairwise interval inference function is generated by fitting multiple linear regression. For each new aircraft type group and each existing type group, calculate the pairwise interactive feature vectors and substitute them into the pairwise interval inference function. Combine semantic similarity weighted correction to generate the final inference interval value and add uncertainty markers to fill the heterogeneous aircraft interval matrix. Acquire four-dimensional trajectory prediction data of all aircraft in the airspace, sample and map them to the basic layer grid cells according to the preset time step, and generate the grid occupancy time series of each aircraft. Iterate through the coexisting aircraft pairs in each basic layer grid cell within each time window, query the filled heterogeneous aircraft spacing matrix to obtain the exclusive safety spacing value, and mark aircraft pairs whose minimum expected distance value is lower than the exclusive safety spacing value as potential conflict pairs. For each potential conflict pair, a type-aware conflict severity score is calculated based on a weighted factor of the exclusive safety interval value, the minimum expected distance value, and the inverse of the collision avoidance maneuverability. Based on semantic similarity, a set of source concepts is selected from existing types, and a rule parameter template for a new type of aircraft is generated by weighted averaging of transfer function and semantic similarity. Differentiated flight rule parameters are generated by combining cumulative conflict risk values ​​and rule parameter templates according to aircraft type groups, and then aggregated and outputted layer by layer along a multi-level grid structure to produce a differentiated dynamic digital flight rule set for heterogeneous aircraft.

[0006] Furthermore, the calculation of the pairwise interaction feature vectors includes: The maximum speed, minimum speed, maximum turning angular velocity, and maximum rate of climb for each type group were standardized using Z-score to eliminate dimensions. For a pair of type groups consisting of type group A and type group B, the paired interactive feature vector contains four components: the first component is the absolute value of the difference between the maximum speeds after standardization of the two type groups; the second component is the absolute value of the difference between the minimum speeds after standardization of the two type groups; the third component is the reciprocal of the sum of the maximum turning angular velocities after standardization of the two type groups; and the fourth component is the reciprocal of the sum of the maximum climb rates after standardization of the two type groups. The form of the pairwise interval inference function is that the inferred safe interval value is equal to the product of the transpose of the weight vector and the pairwise interaction feature vector plus a bias term. The weight vector and the bias term are obtained by fitting the training sample set composed of the pairwise interaction feature vectors and corresponding interval values ​​of all existing type pairs using the least squares method, and the standard deviation of the difference between the predicted value and the actual interval value of all training samples is recorded as the standard deviation of the prediction residual.

[0007] Furthermore, the step of generating the final inference interval value by combining semantic similarity weighted correction includes: The cosine similarity is used to calculate the similarity between the semantic feature vector of the new type and the semantic feature vector of each existing type group. The cosine similarity between all existing type groups and the new type is normalized to obtain the normalized weight. For each existing type group, calculate the mean of the interval values ​​of the existing type group and other existing type groups in the heterogeneous aircraft interval matrix, and obtain the mean of the left-behind prediction by predicting the interval value of the existing type group through leave-behind verification. The leave-behind verification is to remove all type group pairs containing the existing type group from the training sample set, refit the pair interval inference function on the remaining samples, and take the mean of the predicted values. The final inferred interval value is equal to the inferred interval estimate plus a correction term, which is the sum of the products of the normalized weights of each of the existing type groups and the difference between the mean of the corresponding interval value and the mean of the leave-one prediction.

[0008] Furthermore, the uncertainty marker is calculated as follows: For each final inference interval value between the new aircraft type group and the existing type group, the uncertainty label is equal to the standard deviation of the prediction residual divided by the sum of the cosine similarity between the semantic feature vector of the new type and the semantic feature vector of the existing type group and a small positive number to prevent division by zero; the uncertainty label and the corresponding final inference interval value are written together into the filled heterogeneous aircraft interval matrix.

[0009] Furthermore, the description fields in the structured description document include aircraft category identifier, maximum speed, minimum speed, maximum turn rate, maximum rate of climb, maximum rate of descent, communication capability level, and operating mode description; the pre-trained text encoder is a language model based on the Transformer architecture, which processes numerical fields by converting the numerical values ​​into textual expressions containing units of measurement before encoding, and encodes each description field into a textual semantic vector and concatenates them in the order of the fields to generate a semantic feature vector.

[0010] Furthermore, the calculation of the type-aware conflict severity score includes: The baseline score is equal to the difference between the exclusive safety separation value and the minimum expected distance value divided by the exclusive safety separation value; the collision avoidance maneuverability score is equal to the product of the aircraft's maximum turn rate after Z-score standardization and the maximum climb rate after standardization; the type-aware conflict severity score is equal to the baseline score multiplied by the reciprocal of the sum of the collision avoidance maneuverability scores of the two aircraft in the conflict. For conflict pairs involving new aircraft and where the uncertainty mark of the inferred interval value exceeds a preset uncertainty threshold, the type-aware conflict severity score is multiplied by a preset uncertainty amplification factor to obtain an adjusted type-aware conflict severity score, wherein the preset uncertainty amplification factor is greater than one.

[0011] Furthermore, it also includes: Along the flight path of each potential conflict pair, the conflict influence domain is expanded to the adjacent grid cells in front and behind, with the conflict occurrence grid cell as the center. The expansion range is determined based on the speed of the two aircraft in the conflict pair and the preset conflict influence time window. The conflict risk value is accumulated for each grid cell covered by the conflict influence domain according to the aircraft type group, and the accumulated conflict risk value of each basic layer grid cell is generated according to the type group.

[0012] Furthermore, the rule parameter template for generating novel aircraft through a weighted average of transfer function and semantic similarity includes: For each source concept, several existing types belonging to the same operational mode as the source concept are obtained from the existing type group as local training samples. The operational feature parameter vector of the local training samples is used as input and the rule parameter vector is used as output. The least squares method is used to fit a linear regression model to generate the transfer function corresponding to the source concept. The operational feature parameters of the new aircraft are substituted into the transfer function corresponding to each source concept to output the predicted value of the rule parameters. The rule parameter template of the new aircraft is equal to the weighted average of the predicted value of the rule parameters corresponding to each source concept with the corresponding semantic similarity as the weight. The operational characteristic parameters include maximum speed, minimum speed, maximum turning angular velocity, maximum rate of ascent, maximum rate of descent, and communication capability level. Before being substituted into the transfer function, the numerical operational characteristic parameters are standardized using Z-score, and the communication capability level is converted into a numerical value using ordinal encoding. The rule parameters include upper speed limit, lower speed limit, minimum turning radius limit, maximum altitude layer change frequency, and communication reporting interval.

[0013] Furthermore, the layer-by-layer aggregation along the multi-level mesh structure includes: Rule aggregation is performed from bottom to top along the multi-level grid index structure according to type groups. The most stringent value of the flight rule parameters of the same type group in all the lower-level grid cells contained in the upper-level grid cell is taken, and the flight rule parameters of each level and type group are generated by pushing them upward layer by layer.

[0014] Furthermore, it also includes the following processing: encapsulating the flight rule parameters of each level and type group into a multi-level classification digital flight rule data structure, attaching the new aircraft type group identifier and uncertainty label of the inference interval value to the rule parameters corresponding to the new aircraft type group, and outputting a set of differentiated dynamic digital flight rules containing heterogeneous aircraft. The uncertainty label is the maximum value of the uncertainty labels of all type pairs involved in the new aircraft type group.

[0015] The beneficial effects of this invention are as follows: This invention solves the technical problem that the existing zero-sample transfer framework cannot handle the inference of pairwise interval relationships between two aircraft by fitting a pairwise interval inference function with the pairwise interaction feature vectors of two aircraft as inference input. It achieves the technical effect of inferring the type pair-specific interval value even when there are no new type entries in the interval matrix of heterogeneous aircraft, and avoiding the degradation of classification conflict detection to the use of default conservative interval values. By adjusting the estimated inference interval using semantic similarity weighted correction, the technical effect of reducing the influence of the pairwise interval inference function on the extrapolation bias of new type feature domains was achieved. By calculating uncertainty markers based on the standard deviation of predicted residuals and semantic similarity, and introducing an uncertainty amplification factor into the conflict severity score, a technical effect of maintaining the conservatism of safety margin was achieved in the early stage of introducing a new type of conflict, where historical verification data was lacking. By migrating rule parameter templates from the semantically most similar source concepts and combining them with a multi-level grid structure for layer-by-layer aggregation, the technical effect of incorporating new aircraft type groups into a multi-level flight rule unified management system was achieved under the condition of zero historical operational data. Attached Figure Description

[0016] Figure 1 This is a flowchart of the dynamic digital flight rule construction method based on a standardized airspace grid according to the present invention; Figure 2 This is a comparison chart of key performance parameters for each type of group in the examples of this invention; Figure 3 This is an example of the existing type group pair interaction feature vector and actual interval value diagram in the present invention; Figure 4 This is an example of the inference interval value and uncertainty diagram of the novel aircraft and each existing type group in the present invention; Figure 5 This is a comparison chart of potential conflict pair safety interval values ​​and minimum expected distances in an example of the present invention; Figure 6 This is a potential conflict severity rating chart for type-aware conflict in an example of the present invention; Figure 7 This is an example of the novel aircraft rule parameter template generation - source concept weighted average graph in this invention; Figure 8 This is a key numerical diagram of the entire data flow process for the novel aircraft access in an example of the present invention. Detailed Implementation

[0017] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0018] At least one embodiment of the present invention discloses a method for constructing dynamic digital flight rules based on a standardized airspace grid, such as... Figure 1 As shown, it includes the following steps: Step 1: Encode the new aircraft type and the existing structured description documents of each type to generate semantic feature vectors; Obtain structured description documents for the new aircraft type, use a pre-trained text encoder to encode each description field in the structured description document into a text semantic vector, and concatenate them in the order of the fields to generate a semantic feature vector for the new type; obtain structured description documents for each existing type group, use the same pre-trained text encoder to perform the same encoding and concatenation operations on the structured description documents of each existing type to generate a set of semantic feature vectors for the existing types.

[0019] It should be noted that the description fields in the above structured description document include aircraft category identifier, maximum speed, minimum speed, maximum turn rate, maximum rate of climb, maximum rate of descent, communication capability level, and operating mode description.

[0020] Among them, the aircraft category identifier and operating mode description are text fields, the maximum speed, minimum speed, maximum turn rate, maximum climb rate, and maximum descent rate are numerical fields, and the communication capability level is a discrete level field.

[0021] The pre-trained text encoder processes numerical fields by converting the numerical values ​​into textual representations that include units of measurement before encoding them.

[0022] It should be noted that the above-mentioned pre-trained text encoder is a language model based on the Transformer architecture. The input of the pre-trained text encoder is a text sequence, and the output is a semantic feature vector of fixed dimensions.

[0023] The pre-trained text encoder is pre-trained on a general corpus and does not require parameter updates during the application of this method.

[0024] Step 2: Fit a pairwise interval inference function based on the interval data of the existing type pairs; Obtain the interval values ​​of all existing type pairs in the existing heterogeneous aircraft interval matrix, as well as the performance parameters of the corresponding two type pairs. Calculate the pairwise interaction feature vector for each existing type pair. Before calculating the pairwise interaction feature vector, perform Z-score standardization on the maximum speed, minimum speed, maximum turn rate, and maximum climb rate of each type pair to eliminate the influence of differences in the dimensions of different physical quantities on the subsequent linear regression fitting. Use the pairwise interaction feature vectors of all existing type pairs as input samples and the corresponding interval values ​​as labels to form a training sample set. Use multiple linear regression to fit the training sample set, generating a pairwise interval inference function from the pairwise interaction feature vectors to the safe interval values, and record the standard deviation of the prediction residuals during the fitting process. ,in is the standard deviation of the difference between the predicted value and the actual margin value for all training samples.

[0025] It should be noted that the components of the above pairwise interactive feature vectors are defined as follows: For type groups and type group A pair of type groups, paired interactive feature vectors for:

[0026] in, and Type groups and type group The maximum speed after standardization and These are the standardized minimum speeds for the two types of groups, and These are the standardized maximum turning angular velocities for the two groups, respectively. and These represent the standardized maximum climb rates for the two groups. The paired interactive feature vectors reflect the relative speed characteristics of the two groups through the absolute value of the speed difference, and reflect the joint maneuver avoidance capability of the two groups through the sum of the turning angular velocities and the reciprocal of the sum of the climb rates.

[0027] It should be noted that the third component in the above pairwise interactive feature vectors and the fourth component During calculation, it is required that and This means that at least one of the two types of aircraft involved in the calculation has a non-zero maximum turning angular velocity and a non-zero maximum rate of climb. This condition is naturally true for actual aircraft types.

[0028] It should be noted that the above pairwise interval inference function has the following form:

[0029] in, For the inferred safety interval value, The weight vector obtained by fitting a multiple linear regression. For type group and type group Paired interactive feature vectors For bias terms, This indicates transpose. Multiple linear regression uses the least squares method to fit the training sample set, solving for the loss function using the mean squared error of the difference between the predicted and actual interval values ​​of all training samples. and , is the standard deviation of the difference between the predicted value and the actual margin value for all training samples.

[0030] Step 3: Infer the safety interval values ​​between the new type of aircraft and each existing type and fill in the heterogeneous aircraft interval matrix; Group of new aircraft types With each existing type group ( ,in (Given the total number of existing type groups), calculate the pairwise interaction feature vectors between the two groups respectively. The paired interactive feature vectors Substitute into the pairwise interval inference function and output the estimated inference interval. ; Calculate the semantic feature vector of the new type and each existing type group using cosine similarity. Similarity between semantic feature vectors The inference interval prediction is weighted and corrected using semantic similarity as the confidence weight, to generate the final inference interval value between the new type and each existing type. Fill all the final inferred interval values ​​into the new rows and columns of the heterogeneous aircraft interval matrix to generate the filled heterogeneous aircraft interval matrix.

[0031] It should be noted that the above weighted correction process is as follows: for new aircraft type groups With existing type groups Inference interval estimates between The cosine similarity between all existing type groups and the new type is used. Perform normalization to obtain normalized weights. ,in This represents the total number of existing type groups.

[0032] For existing type groups The average interval value of itself in the heterogeneous aircraft interval matrix with other existing type groups is denoted as . The final inferred interval value is:

[0033] in, Group of new aircraft types With existing type groups The estimated inference interval between them For existing type groups The corresponding normalized weights, For existing type groups The average of the interval values ​​between the heterogeneous aircraft and other existing type groups in the interval matrix. To use pairwise interval inference functions for existing type groups under leave-one-out verification The results of predicting the interval values, specifically, leave-one-out verification for existing type groups. The prediction method is as follows: remove all samples containing... from the training sample set. The type of sample groups are refitted with a pairwise interval inference function on the remaining samples, and then... The pairwise interaction feature vectors between the new and existing type groups are substituted into the refitted pairwise interval inference function to obtain predicted values, and the mean of these predicted values ​​is taken as the predicted value. This correction term makes the contribution of existing type groups with higher semantic similarity to the new type to the correction direction greater.

[0034] Furthermore, to quantify the reliability of the inference interval values ​​to ensure a safety margin in the initial stage of introducing the new type, for each final inference interval value... Additional uncertainty marker Uncertainty labels are based on the standard deviation of the predicted residuals of the pairwise inference function. Similarity to cosine calculate:

[0035] in, Let be the standard deviation of the difference between the predicted and actual margin values ​​for all training samples during the pairwise margin inference function fitting process. For the semantic feature vector of the new type and the existing type group The cosine similarity between semantic feature vectors To prevent division by zero of tiny positive numbers.

[0036] The uncertainty marker increases as semantic similarity decreases, reflecting the degree of deviation when extrapolating the inferred interval value to the new type of feature domain. The uncertainty marker and the final inferred interval value are written together into the filled heterogeneous aircraft interval matrix.

[0037] Step 4: Sample four-dimensional trajectory prediction data and generate grid occupancy time series for each aircraft; Obtain the four-dimensional trajectory prediction data and type group identifiers of all aircraft (including new aircraft) currently in flight and planned to take off in the current airspace. Sample each four-dimensional trajectory prediction data into a discrete trajectory point sequence according to a preset time step. Map the three-dimensional spatial coordinates of each discrete trajectory point to the spatial index code of the basic layer grid cell to generate the grid occupancy time series of each aircraft.

[0038] It should be noted that the aforementioned four-dimensional trajectory prediction data includes a sequence of the aircraft's longitude, latitude, altitude, and corresponding timestamps within a preset future time range. The basic layer grid cell is the finest-grained spatial partitioning unit in the airspace grid management structure, and each basic layer grid cell has a unique spatial index code. The above grid occupancy time series records the spatial index codes of the basic layer grid cells occupied by a particular aircraft at each time step.

[0039] Furthermore, the aforementioned preset time step and preset time range together determine the temporal resolution and coverage length of the grid occupancy time series: the preset time step determines the interval between adjacent sampling times, and the preset time range determines the future time span covered by the four-dimensional trajectory prediction data. Both are pre-configured according to the airspace management accuracy requirements during system deployment and are kept consistent for all aircraft in the same rule calculation process to ensure that the grid occupancy time series of each aircraft are aligned on the time axis, thereby supporting the traversal detection of coexisting aircraft pairs within the same time window in subsequent step 5.

[0040] Step 5: Perform classification-based spatiotemporal collision detection and mark potential conflict pairs; Perform categorized spatiotemporal collision detection on the grid occupancy time series of all aircraft: Traverse the set of occupant aircraft in each time window of all base layer grid cells. For each pair of aircraft coexisting in the same time window within the same base layer grid cell, query the filled heterogeneous aircraft interval matrix according to the type group identifiers of the two aircraft to obtain the exclusive safety interval value of the aircraft pair. Calculate the minimum expected distance value of the aircraft pair based on the discrete trajectory point sequence of the two aircraft within the time window. Mark aircraft pairs with minimum expected distance values ​​lower than the exclusive safety interval value as potential conflict pairs and record the conflict occurrence grid cell, conflict time window, exclusive safety interval value, and minimum expected distance value of the conflict pair.

[0041] It should be noted that the dedicated safety interval value used by the aforementioned aircraft involving new types of aircraft is the final inferred interval value generated in step 3. Finally, the interval value was inferred. The query method is the same as the query method for the interval value of the existing type group pair. Both are obtained by indexing the type group identifiers of the two aircraft in the filled heterogeneous aircraft interval matrix.

[0042] It should be noted that the calculation method for the minimum expected distance value is as follows: for two aircraft coexisting in the same time window, traverse all pairs of trajectory points at the same time step in the discrete trajectory point sequence of each aircraft in the time window, calculate the Euclidean distance between each pair of trajectory points at the same time, and take the minimum value of the Euclidean distance in all time steps as the minimum expected distance value of the aircraft pair in the time window.

[0043] Step 6: Calculate the type-perceived conflict severity score for each potential conflict pair; For each potential conflict pair, obtain the specific safety interval value for that conflict pair. and minimum expected distance value Calculate the base score; obtain the maximum turn rate and maximum climb rate of each of the two aircraft in the conflict pair, and calculate the inverse weighting factor of the collision avoidance maneuverability; multiply the base score by the inverse weighting factor to generate the type-aware conflict severity score of the conflict pair.

[0044] It should be noted that the above-mentioned types of perceived conflict severity scores The calculation method is as follows:

[0045] in, This is the specific safety interval value for this conflict pair. This represents the minimum expected distance value for the conflict pair. and These represent the collision avoidance maneuverability values ​​for the two aircraft in a conflict, and the collision avoidance maneuverability. Based on the aircraft's maximum turn rate after Z-score standardization and maximum climb rate calculate:

[0046] in, The maximum turning angular velocity of an aircraft after standardization. This refers to the aircraft's standardized maximum climb rate. In other words, collision avoidance maneuverability is the product of the standardized horizontal turning capability and vertical climb capability. The weaker the joint avoidance capability between the two aircraft (…), the better. The smaller the value, the higher the type-aware conflict severity score.

[0047] It should be noted that the above-mentioned formula for scoring the severity of perceived conflict requires... This means that at least one aircraft in the conflict pair possesses non-zero collision avoidance maneuverability, a condition that is inherently true for aircraft in actual operation; it also requires... That is, the exclusive safety interval value is positive, a condition guaranteed by the definition of the heterogeneous aircraft spacing matrix.

[0048] Furthermore, in order to provide a more conservative risk assessment for conflict pairs with low reliability of inferred interval values, uncertainty labels are applied to conflicts involving novel aircraft and inferred interval values. Exceeding the preset uncertainty threshold The conflict pairs will be scored based on type-aware conflict severity. Multiply by the preset uncertainty amplification factor ( ), to obtain the adjusted type-aware conflict severity score ,in The original type-perceived conflict severity score. This is a preset uncertainty amplification factor. This processing allows conflict pairs with low reliability of inference interval values ​​to automatically obtain a higher risk rating.

[0049] Furthermore, based on the aforementioned type-aware conflict severity scoring, the following processing is also included: Along the flight paths of each potential conflict pair, the conflict influence domain is expanded to adjacent grid cells centered on the conflict occurrence grid cell. The expansion range is determined based on the speeds of the two aircraft in the conflict pair and a preset conflict impact time window. Conflict risk values ​​are accumulated for each grid cell covered by the conflict influence domain according to aircraft type groups, generating a cumulative conflict risk value for each base-layer grid cell categorized by type group. For scenarios where expanding the conflict influence domain is not required, the cumulative conflict risk value for each base-layer grid cell categorized by type group is directly taken as the sum of the type-aware conflict severity scores of all conflict pairs involving the corresponding type group within that base-layer grid cell.

[0050] Step 7: Generate rule parameter templates for novel aircraft based on semantic similarity transfer; Calculate the cosine similarity between the new type of semantic feature vector and the existing semantic feature vectors of each type, and select the top vectors with the highest similarity. Each existing type serves as a source concept set, in which The set of source concepts is the preset number of source concepts. In , , These represent the numbers arranged from highest to lowest similarity. The, the The and the first The system extracts existing types; extracts rule parameter configuration records and operational feature parameters of each source concept from the source concept set; fits the mapping relationship between the operational feature parameters and rule parameters of each source concept using linear regression to generate a transfer function; substitutes the operational feature parameters of the new aircraft into the transfer function corresponding to each source concept to output the predicted value of the rule parameters corresponding to each source concept; and performs a weighted average of the predicted values ​​of rule parameters from different source concepts using the semantic similarity between each source concept and the new type as the weight to generate the rule parameter template of the new aircraft.

[0051] It should be noted that the aforementioned operational characteristic parameters include the aircraft's maximum speed, minimum speed, maximum turn rate, maximum rate of climb, maximum rate of descent, and communication capability level. Before substituting these operational characteristic parameters into the transfer function, Z-score standardization was used to eliminate dimensions for the maximum speed, minimum speed, maximum turn rate, maximum rate of climb, and maximum rate of descent. Ordinal encoding was used to convert the communication capability level to a numerical value to eliminate the influence of dimensional differences between different physical quantities and non-numerical fields on the linear regression fit. Rule parameters include the upper speed limit, lower speed limit, minimum turning radius limit, maximum altitude layer change frequency, and communication reporting interval.

[0052] In the above weighted average, for each source concept ( ,in The estimated value of each rule parameter for the total number of source concepts. and corresponding semantic similarity Rule parameter template for new aircraft for:

[0053] in, For the first Individual Source Concept The corresponding estimated value of the rule parameters, For the first Individual Source Concept Semantic similarity with new types, The total number of source concepts, For the index of the source concept, the value starts from... arrive .

[0054] It should be noted that the fitting method for the transfer functions of the above source concepts is as follows: for the source concept Obtain the source concept from the existing type group. Several existing types belonging to the same operating mode are used as local training samples. The operating feature parameter vectors of each local training sample are used as inputs and the corresponding rule parameter vectors are used as outputs. The least squares method is used to fit a linear regression model to generate source concepts. The corresponding transfer function; when the number of local training samples is insufficient for independent fitting, the running feature parameters and rule parameters of all existing type groups are used as supplementary training samples to participate in the fitting.

[0055] Step 8: Generate differentiated flight rules by aircraft type group and aggregate the output along a multi-level grid structure; For each basic layer grid cell, differentiated flight rule parameters are generated according to aircraft type groups: For existing type groups, the cumulative conflict risk value of the existing type group in the basic layer grid cell is obtained, the rule tightening level is determined according to the preset range of the cumulative conflict risk value, the baseline parameters are obtained from the exclusive rule parameter template of the existing type group, and the dynamic adjustment coefficient corresponding to the rule tightening level is superimposed to generate the flight rule parameters of the existing type group in the basic layer grid cell. For a new type of aircraft group, the cumulative conflict risk value and the rule parameter template generated by migration for the new type of aircraft group in the basic layer grid cell are obtained. The rule compaction level is determined according to the cumulative conflict risk value. The baseline parameters in the rule parameter template are superimposed with the corresponding dynamic adjustment coefficients to generate the flight rule parameters for the new type of aircraft group in the basic layer grid cell.

[0056] Furthermore, in order to manage the flight rule parameters of the basic layer grid cells in a unified manner to the upper layer grid structure, the following processing is also included: performing rule aggregation from bottom to top along the multi-level grid index structure according to type groups. Specifically, the most stringent value of the flight rule parameters of the same type group in all the lower layer grid cells contained in the upper layer grid cell is taken and pushed up layer by layer to generate the flight rule parameters of each level and type group.

[0057] Furthermore, based on the above rule aggregation, the following processing is also included: encapsulating the flight rule parameters of each level and type group into a multi-level classified digital flight rule data structure, and attaching the new aircraft type group identifier and uncertainty label of the inference interval value to the rule parameters corresponding to the new aircraft type group (the uncertainty label is taken from the uncertainty markers of all type group pairs involved in the new aircraft type group). (maximum value), outputting a differentiated dynamic digital flight rule set for heterogeneous aircraft, including new types.

[0058] The aforementioned uncertainty labels provide a reference for subsequent airspace operation management. Once sufficient operational data on new aircraft has been accumulated, the inferred interval values ​​can be replaced and the rule parameters updated using actual operational data.

[0059] This implementation fits a pairwise interval inference function from the interval values ​​and pairwise interaction features of existing type pairs. It uses the pairwise interaction feature vectors of two aircraft (including the absolute values ​​of the maximum and minimum speed differences between the two types, the reciprocal of the sum of the turn angular velocities, and the reciprocal of the sum of the climb rates) as the inference input, rather than the operational features of a single aircraft. Therefore, it overcomes the structural defect of existing zero-sample transfer frameworks, which only support unidirectional mapping from single aircraft features to rule parameters and cannot handle the inference of pairwise relationships between two aircraft. This allows the type pair-specific interval values ​​to be inferred based on the performance interaction relationships between the new aircraft and each existing type even when there are no new type entries in the heterogeneous aircraft interval matrix. This avoids the situation where classification conflict detection degenerates into using default conservative interval values, resulting in excessive restrictions on the operational density of new aircraft.

[0060] Furthermore, this implementation adjusts the estimated inference interval by using semantic similarity weighted correction. It applies a higher contribution weight to the correction direction by using existing types that are semantically closer to the new type. Therefore, when the pairwise interval inference function extrapolates the feature domain of the new type, the correction result is biased towards the interval distribution of existing types that are closer to the operating characteristics of the new type, thus reducing the impact of extrapolation bias.

[0061] In addition, this implementation calculates uncertainty markers based on the standard deviation of predicted residuals and semantic similarity, and introduces an uncertainty amplification factor in the type-aware conflict severity score. This allows conflict pairs with lower reliability of inference interval values ​​to automatically obtain higher risk ratings, thus maintaining the conservatism of safety margins in the early stages of introducing new types when historical verification data is lacking. As semantic similarity decreases, uncertainty markers increase, the trigger probability of the amplification factor increases, and risk assessment tends to be more stringent.

[0062] Finally, this implementation method migrates rule parameter templates from the semantically most similar source concepts and dynamically adjusts them by combining the migrated rule parameter templates with classification conflict risk values. Under the condition of zero historical operational data, differentiated flight rule parameters are generated for the new aircraft type group. Through the layer-by-layer aggregation of a multi-level grid structure, the new type is incorporated into a multi-level flight rule system that is uniformly managed with the existing type groups. Thus, the complete access of the new type is completed without relying on the actual operational data of the new aircraft.

[0063] Based on the above-mentioned method for constructing dynamic digital flight rules based on standardized airspace grids, it is applied to the following application scenario, with the following background: A city's airspace management center is responsible for managing a grid-managed airspace, which already has four types of aircraft operating long-term: fixed-wing manned aircraft (type group) Multi-rotor unmanned delivery aircraft (type group) eVTOL manned spacecraft (type group) ) and tethered drones (type group) ).

[0064] The airspace rule calculation server has maintained a heterogeneous aircraft spacing matrix for six type pairs among the above four types of aircraft, and has configured exclusive flight rule parameter templates for each type pair.

[0065] On March 15, 20XX, a composite-wing cargo drone (type group) was launched. This is the first application to enter this airspace for operation. The UAV has both fixed-wing cruise and vertical take-off and landing capabilities, with a maximum speed of 72 km / h, a minimum speed of 12 km / h, a maximum turn rate of 18 ° / s, a maximum climb rate of 4.2 m / s, a maximum descent rate of 3.8 m / s, a communication capability level of 3, and an autonomous cargo delivery operation mode.

[0066] The airspace rule calculation server needs to perform type group calculations without any new historical data. Infer the exclusive safety interval values ​​for each existing type group and generate differentiated flight rules, which are then incorporated into a unified multi-level flight rule system.

[0067] An example of the implementation process is as follows: The implementation example of core step 1 is as follows: Corresponding to step 1, the airspace rule calculation server obtains the compound-wing cargo drone (type group) The document contains structured descriptions of four existing type groups. Each description field is encoded separately using a pre-trained text encoder and then concatenated to generate semantic feature vectors for each type group. For example, the maximum speed field "72km / h" is text-based and input into the encoder, where it is concatenated with the encoded results of the other fields to form a 768-dimensional semantic feature vector. Four types of groups have already generated semantic feature vector sets through the same process. .

[0068] Table 1 Key Fields of Structured Description Documents for Each Group Type

[0069] The implementation example of core step 2 is as follows: Corresponding to step 2, the server extracts the interval values ​​of the existing 6 type pairs from the heterogeneous aircraft interval matrix, performs Z-score standardization on the speed, turn rate and climb rate of each type pair, eliminates the difference in dimensions, calculates the pairwise interaction feature vector of each pair, and fits the pairwise interval inference function using the least squares method.

[0070] By type group (Fixed-wing manned aircraft) and Taking the calculation of pairwise interactive feature vectors of a multi-rotor delivery aircraft as an example. Assume that after Z-score normalization, of , , , ; of , , , .

[0071] The first component of the pairwise interactive feature vector is:

[0072] The second component is:

[0073] The third component is:

[0074] The fourth component is:

[0075] Therefore .

[0076] Table 2. Paired interaction feature vectors and actual margin values ​​of existing type pairs.

[0077] Least squares fitting is performed on the above 6 samples to obtain the weight vector. Bias term residual standard deviation m.

[0078] The implementation example of core step 3 is as follows: Corresponding to step 3, the server computing type group The paired interaction feature vectors between each existing type group are substituted into the paired interval inference function to obtain the estimated inference interval. Then, cosine similarity is used for weighted correction to generate the final inference interval value and add an uncertainty label.

[0079] By type group and Taking the inference of (multi-rotor delivery aircraft) as an example. Assume that after standardization... of , , , ,but:

[0080]

[0081] Substitute into the inference function:

[0082] In terms of semantic similarity, new types The cosine similarity with each existing type is calculated by a pre-trained text encoder. , , , The normalized weights are as follows:

[0083] but , , , .

[0084] Leave one out for verification to calculate each existing type and matrix mean Afterwards, Apply weighted correction to obtain the final inferred interval value. .

[0085] Uncertainty is denoted as:

[0086] Table 3. Inference intervals and uncertainty symbols between new aircraft and existing type groups.

[0087] The above results are written into the filled heterogeneous aircraft spacing matrix, where - The uncertainty mark of 75.1m exceeds the preset threshold. m will trigger the amplification factor adjustment in step 6.

[0088] The implementation example of core step 5 is as follows: Corresponding to step 5, the server performs classified spatiotemporal collision detection on the grid occupancy time series of all aircraft in the current airspace.

[0089] At 09:12:00 on March 15, 20XX, in the basic layer mesh unit Within the same time window, a compound-wing cargo drone (aircraft) was detected. Type group ) and tethered drones (aircraft) Type group )coexist.

[0090] The server uses the two type group identifiers to query and populate the heterogeneous aircraft spacing matrix to obtain the specific safety spacing value. m; Iterate through the Euclidean distances between the discrete trajectory points of the two machines within the time window, and take the minimum value to obtain the minimum predicted distance value. m. Due to The aircraft pair was marked as a potential conflict pair and relevant information was recorded.

[0091] Table 4 Potential Conflict Detection Records

[0092] An implementation example of core step 6 is as follows: Corresponding to step 6, the server scores the severity of conflicts for each potential conflict pair based on the perceived computational type. (The sentence is incomplete and ends abruptly.) ( and For example, let's assume that after standardization... of , , of , The collision avoidance maneuverability is as follows:

[0093]

[0094] because The calculation conditions are met, and the basic score is:

[0095] because Involving new types of aircraft and Let the preset uncertainty amplification factor be set. The adjusted score is:

[0096] Table 5 Type-perceived conflict severity scores for each potential conflict pair

[0097] The implementation example of core step 7 is as follows: Corresponding to step 7, server computing type group The cosine similarity with the existing four groups is used to select the top group with the highest similarity. The existing types are used as the source concept set. Table 3 shows the similarity ranking as follows: Therefore, the set of source concepts is The semantic similarity is respectively , , .

[0098] After fitting transfer functions to each source concept, the type groups are... The standardized operational characteristic parameters are substituted into the output rule parameter prediction value.

[0099] Taking the speed limit cap as an example, the estimated values ​​corresponding to the three source concepts are as follows: , , The weighted average is:

[0100] Table 6. Process of Generating Novel Aircraft Rule Parameter Templates

[0101] In step 8, the aforementioned rule parameter template is combined with the cumulative conflict risk value of the grid cells to generate differentiated flight rules and encapsulate additional uncertainty labels (taking the maximum value of each pair of uncertainty labels). Output to a multi-level flight rules system.

[0102] The data flow in the application scenario is explained as follows: The data flow throughout the entire implementation process presents a clear, chain-like progression logic. Step 1 starts from the type group. Starting with the existing structured description documents of the four type groups, semantic feature vectors for each type group are generated, providing the basic input for subsequent similarity calculations. Step 2 utilizes the performance parameters and margin values ​​of the existing six type group pairs, constructs pairwise interaction feature vectors after Z-score standardization, fits an inference function, and outputs a weight vector. Bias terms and residual standard deviation The three parameters, m, are directly passed to step 3. Step 3 comprehensively utilizes the semantic feature vector from step 1 (to calculate cosine similarity). ) and the inference function of step 2 (calculation) After weighted correction, four final inference interval values ​​and uncertainty labels are generated, which are then used to fill the heterogeneous aircraft interval matrix. m and The two data points, m and m, flow directly into steps 5 and 6.

[0103] Step 5 uses the filled interval matrix as the query basis to perform collision detection on the grid occupancy time series and outputs potential conflict pairs (including...). and These two data points, along with the uncertainty marker from step 3, are entered into step 6 to calculate the type-aware conflict severity score. Because the uncertainty exceeds the threshold, it is amplified to Step 7 generates cumulative conflict risk values ​​for each grid cell, categorized by type group. Step 8 uses the semantic similarity from Step 1 to determine the source concept set and generate rule parameter templates in parallel. Step 9 merges the cumulative conflict risk values ​​from Step 6 with the rule parameter templates from Step 7, ultimately generating differentiated flight rules and attaching the uncertainty label 75.1 m output from Step 3. These rules are then aggregated and output along a multi-level grid structure, completing the full airspace access for the compound-wing cargo UAV.

[0104] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for constructing dynamic digital flight rules based on a standardized airspace grid, characterized in that, Includes the following steps: By using a pre-trained text encoder to encode the structured description documents of the new aircraft type and the existing types, a set of semantic feature vectors for the new type and the set of semantic feature vectors for the existing types is generated. Based on the interval values ​​and performance parameters of existing type pairs, the pairwise interaction feature vector is calculated, and the pairwise interval inference function is generated by fitting multiple linear regression. For each new aircraft type group and each existing type group, calculate the pairwise interactive feature vectors and substitute them into the pairwise interval inference function. Combine semantic similarity weighted correction to generate the final inference interval value and add uncertainty markers to fill the heterogeneous aircraft interval matrix. Acquire four-dimensional trajectory prediction data of all aircraft in the airspace, sample and map them to the basic layer grid cells according to the preset time step, and generate the grid occupancy time series of each aircraft. Iterate through the coexisting aircraft pairs in each basic layer grid cell within each time window, query the filled heterogeneous aircraft spacing matrix to obtain the exclusive safety spacing value, and mark aircraft pairs whose minimum expected distance value is lower than the exclusive safety spacing value as potential conflict pairs. For each potential conflict pair, a type-aware conflict severity score is calculated based on a weighted factor of the exclusive safety interval value, the minimum expected distance value, and the inverse of the collision avoidance maneuverability. Based on semantic similarity, a set of source concepts is selected from existing types, and a rule parameter template for a new type of aircraft is generated by weighted averaging of transfer function and semantic similarity. Differentiated flight rule parameters are generated by combining cumulative conflict risk values ​​and rule parameter templates according to aircraft type groups, and then aggregated and outputted layer by layer along a multi-level grid structure to produce a differentiated dynamic digital flight rule set for heterogeneous aircraft.

2. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 1, characterized in that, The calculation of the pairwise interactive feature vectors includes: The maximum speed, minimum speed, maximum turning angular velocity, and maximum rate of climb for each type group were standardized using Z-score to eliminate dimensions. For a pair of type groups consisting of type group A and type group B, the paired interactive feature vector contains four components: the first component is the absolute value of the difference between the maximum speeds after standardization of the two type groups; the second component is the absolute value of the difference between the minimum speeds after standardization of the two type groups; the third component is the reciprocal of the sum of the maximum turning angular velocities after standardization of the two type groups; and the fourth component is the reciprocal of the sum of the maximum climb rates after standardization of the two type groups. The form of the pairwise interval inference function is that the inferred safe interval value is equal to the product of the transpose of the weight vector and the pairwise interaction feature vector plus a bias term. The weight vector and the bias term are obtained by fitting the training sample set composed of the pairwise interaction feature vectors and corresponding interval values ​​of all existing type pairs using the least squares method, and the standard deviation of the difference between the predicted value and the actual interval value of all training samples is recorded as the standard deviation of the prediction residual.

3. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 2, characterized in that, The step of generating the final inference interval value by combining semantic similarity weighted correction includes: The cosine similarity is used to calculate the similarity between the semantic feature vector of the new type and the semantic feature vector of each existing type group. The cosine similarity between all existing type groups and the new type is normalized to obtain the normalized weight. For each existing type group, calculate the mean of the interval values ​​of the existing type group and other existing type groups in the heterogeneous aircraft interval matrix, and obtain the mean of the left-behind prediction by predicting the interval value of the existing type group through leave-behind verification. The leave-behind verification is to remove all type group pairs containing the existing type group from the training sample set, refit the pair interval inference function on the remaining samples, and take the mean of the predicted values. The final inferred interval value is equal to the inferred interval estimate plus a correction term, which is the sum of the products of the normalized weights of each of the existing type groups and the difference between the mean of the corresponding interval value and the mean of the leave-one prediction.

4. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 3, characterized in that, The uncertainty marker is calculated as follows: For each final inference interval value between the new aircraft type group and the existing type group, the uncertainty label is equal to the standard deviation of the prediction residual divided by the sum of the cosine similarity between the semantic feature vector of the new type and the semantic feature vector of the existing type group and a small positive number to prevent division by zero; the uncertainty label and the corresponding final inference interval value are written together into the filled heterogeneous aircraft interval matrix.

5. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 4, characterized in that, The structured description document includes description fields such as aircraft category identifier, maximum speed, minimum speed, maximum turn rate, maximum rate of climb, maximum rate of descent, communication capability level, and operating mode description. The pre-trained text encoder is a language model based on the Transformer architecture. The numerical fields are processed by converting the numerical values ​​into textual expressions containing units of measurement before encoding. Each description field is encoded into a textual semantic vector, and the semantic feature vector is generated by concatenating the fields in order.

6. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 1, characterized in that, The calculation of the type-aware conflict severity score includes: The baseline score is equal to the difference between the exclusive safety separation value and the minimum expected distance value divided by the exclusive safety separation value; the collision avoidance maneuverability score is equal to the product of the aircraft's maximum turn rate after Z-score standardization and the maximum climb rate after standardization; the type-aware conflict severity score is equal to the baseline score multiplied by the reciprocal of the sum of the collision avoidance maneuverability scores of the two aircraft in the conflict. For conflict pairs involving new aircraft and where the uncertainty mark of the inferred interval value exceeds a preset uncertainty threshold, the type-aware conflict severity score is multiplied by a preset uncertainty amplification factor to obtain an adjusted type-aware conflict severity score, wherein the preset uncertainty amplification factor is greater than one.

7. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 6, characterized in that, Also includes: Along the flight path of each potential conflict pair, the conflict influence domain is expanded to the adjacent grid cells in front and behind, with the conflict occurrence grid cell as the center. The expansion range is determined based on the speed of the two aircraft in the conflict pair and the preset conflict influence time window. The conflict risk value is accumulated for each grid cell covered by the conflict influence domain according to the aircraft type group, and the accumulated conflict risk value of each basic layer grid cell is generated according to the type group.

8. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 1, characterized in that, The rule parameter template for generating novel aircraft through a transfer function and a semantic similarity weighted average includes: For each source concept, several existing types belonging to the same operational mode as the source concept are obtained from the existing type group as local training samples. The operational feature parameter vector of the local training samples is used as input and the rule parameter vector is used as output. The least squares method is used to fit a linear regression model to generate the transfer function corresponding to the source concept. The operational feature parameters of the new aircraft are substituted into the transfer function corresponding to each source concept to output the predicted value of the rule parameters. The rule parameter template of the new aircraft is equal to the weighted average of the predicted value of the rule parameters corresponding to each source concept with the corresponding semantic similarity as the weight. The operational characteristic parameters include maximum speed, minimum speed, maximum turning angular velocity, maximum rate of ascent, maximum rate of descent, and communication capability level. Before being substituted into the transfer function, the numerical operational characteristic parameters are standardized using Z-score, and the communication capability level is converted into a numerical value using ordinal encoding. The rule parameters include upper speed limit, lower speed limit, minimum turning radius limit, maximum altitude layer change frequency, and communication reporting interval.

9. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 1, characterized in that, The layer-by-layer aggregation along the multi-level grid structure includes: Rule aggregation is performed from bottom to top along the multi-level grid index structure according to type groups. The most stringent value of the flight rule parameters of the same type group in all the lower-level grid cells contained in the upper-level grid cell is taken, and the flight rule parameters of each level and type group are generated by pushing them upward layer by layer.

10. The method for constructing dynamic digital flight rules based on a standardized airspace grid according to claim 9, characterized in that, It also includes the following processing: encapsulating the flight rule parameters of each level and type group into a multi-level classification digital flight rule data structure, adding the new aircraft type group identifier and uncertainty label of the inference interval value to the rule parameters corresponding to the new aircraft type group, and outputting a set of differentiated dynamic digital flight rules containing heterogeneous aircraft. The uncertainty label is the maximum value of the uncertainty labels of all type pairs involved in the new aircraft type group.

Images