A method for dynamic analysis and abnormal positioning of automobile CAN bus load rate

Abstract

The application belongs to the technical field of automobile message communication, and discloses a dynamic analysis and abnormal positioning method for a CAN bus load rate of an automobile; CAN message streams of an automobile to be analyzed are acquired and encoded to obtain an observed event chain after arbitration; message frames and message identifiers in the observed event chain after arbitration are subjected to frame length processing and arbitration sequence position analysis respectively to obtain a service occupation boundary chain and an arbitration sequence chain; a reverse arbitration queue is generated by analyzing the time sequence correlation between the observed event chain after arbitration and the service occupation boundary chain; the observed event is subjected to arbitration inversion to obtain a sending intention strength trajectory of the message identifier; the sending intention strength trajectory is decomposed into multiple path segments, the path segments are subjected to path reconstruction and same slot competition to obtain a sparse activation path; the message identifiers of the sparse activation path are aggregated into abnormal message identifiers, and an abnormal positioning result is output. The application realizes dynamic analysis and abnormal positioning of a bus load rate.

Description

Technical Field

[0001] This invention relates to the field of automotive message communication technology, and more specifically, to a method for dynamic analysis and anomaly localization of automotive CAN bus load rate. Background Technology

[0002] In automotive internal communication networks, the CAN bus load rate reflects the occupancy of bus resources over a certain period, serving as a crucial basis for evaluating in-vehicle communication status, identifying communication congestion, and detecting abnormal message activity. With the increasing complexity of in-vehicle electronic systems, the number of messages, communication frequency, and node interactions are constantly growing. Therefore, dynamically analyzing the changes in bus load rate and locating abnormal messages or nodes has become a critical requirement for in-vehicle communication safety monitoring and bus testing and diagnostics.

[0003] Existing methods for bus load analysis typically rely heavily on observations of messages actually sent after arbitration. This makes it difficult to further distinguish between the actual competitive pressure exerted on bus resources by each message before arbitration and the observation delay caused by arbitration queuing. Consequently, load source analysis tends to remain at the level of sent message performance. Furthermore, in anomaly localization, some methods rely more on single load changes, single message statistics, or phased anomaly results for judgment, lacking path-based extraction of the continuous transmission intent change process. This makes it difficult to reliably identify the source of persistent abnormal load, leaving room for improvement in the stability and interpretability of the anomaly message identifier and the location results of the corresponding electronic control unit. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: a method for dynamic analysis and anomaly localization of automotive CAN bus load rate, comprising: S1. Obtain the CAN message stream of the vehicle to be analyzed and perform event-based encoding to obtain the arbitrated observation event chain; S2. Perform frame length processing and arbitration sequence analysis on the message frames and message identifiers in the post-arbitration observation event chain to obtain the service occupancy boundary chain and arbitration sequence chain. S3. By analyzing the temporal correlation between the post-arbitration observation event chain and the service occupancy boundary chain, a reverse arbitration queue is generated; S4. By analyzing the correlation between the queue arrangement characteristics in the reverse arbitration queue, the arbitration advantage characteristics in the arbitration sequence chain, and the boundary continuity characteristics in the service occupancy boundary chain, arbitration inversion is performed on the observed events to obtain the sending intent strength trajectory corresponding to each message identifier. S5. Decompose the transmission intent strength trajectory into multiple path segments, and obtain sparse activation paths by reconstructing the path segments and competing for slots in the same slot. S6. Aggregate the message identifiers to which the sparse activation path belongs into abnormal message identifiers, map them to the corresponding ECU nodes, and output the abnormal location results.

[0005] Preferably, the method in S3 includes: S31. Using the end position of each service in the service occupancy boundary chain as the slot reference, propose a reverse competition slot; S32. Create a set of paired event numbers and initialize the set of paired event numbers to an empty set; S33. Traverse the inverse contention slots in the direction from the tail to the head of the service occupancy boundary chain, and mark the traversal target as the current inverse contention slot. S34. At the current inverse contention slot, traverse the observation event chain after arbitration and receive observation events whose time is earlier than the service end position corresponding to the current inverse contention slot. For each observed event, determine whether the event number already exists in the set of paired event numbers; If the event number of the observed event does not exist in the set of paired event numbers, then the corresponding observed event will be included in the candidate pool and the paired set corresponding to the current reverse competition slot; If the event number of the observed event already exists in the set of paired event numbers, then skip the observed event; S35. Each observation event in the candidate pool is designated as a candidate sending intention item. Different candidate sending intention items are arranged from high to low according to the arbitration order to obtain the reverse arbitration queue corresponding to the current reverse competition slot.

[0006] Preferably, the method in S4 includes: S41. Extract historical normal message segments, initialize the variable channel for the candidate sending intent items in the reverse arbitration queue, and obtain the initial value of the message strength variable corresponding to each message identifier. S42. By analyzing the winning compensation features and cross-slot retention features of historical candidate sending intent items, configure the initial values ​​of the slot incremental strength variable and the queue residual strength variable and merge them with the initial value of the message strength variable to obtain the sending intent strength variable; S43. Read the message identifier in the current candidate sending intent item, mark the corresponding sending intent strength variable as the arbitration scoring strength value, and accumulate the arbitration scoring strength value and arbitration sequence to obtain the arbitration scoring value of the corresponding candidate sending intent item. S44. Perform differentiable normalization on the arbitration scores of different candidate sending intention items to obtain the continuous selection probability of each candidate sending intention item. S45. Based on the winning attribute and continuous selection probability of each candidate sending intention item in the current reverse competition slot, construct the inversion error characteristics of the current reverse competition slot. S46. The inversion error characteristics are backed to the message strength variable, slot incremental strength variable and queue residual strength variable according to the variable type and arbitration inversion is performed until the inversion error converges. The converged transmission intention strength variables are connected according to the message identifier and the reverse competition slot time order to form the transmission intention strength trajectory.

[0007] Preferably, the method in S41 includes: S411. Obtain historical CAN message streams that belong to the same CAN bus channel as the CAN message stream to be analyzed, and extract the set of historical normal message segments. S412. Divide the messages in the set of historical normal message segments into channels according to the message identifier. Perform steps S2 and S3 for each historical variable channel to obtain the historical reverse competition slot and historical candidate transmission intention item. S413. Combine the historical candidate sending intent items in each historical reverse competition slot to obtain a historical candidate group. Generate a set of historical candidate groups based on the historical candidate groups of different historical reverse competition slots. S414. In each historical candidate group, read the actual sent observation events of the historical reverse competition slot and record them as the historical winners of the corresponding historical reverse competition slot. S415. Count the number of times different historical winners appear in the historical candidate group set according to the message identifier, and obtain the candidate exposure number of the corresponding message identifier. S416. Count the number of times each message identifier is counted as a historical winning item to obtain the historical winning count of the corresponding message identifier; S417. Perform data transformation processing on each candidate exposure count and historical winning count to obtain the initial value of the message strength variable for each message identifier; S418. Based on the initial values ​​of the message strength variables in the historical reverse competition slots, calculate the proportion of the initial value of the message strength variable for each message identifier, and record it as the message strength ratio value. S419. Read the message identifier of the historical candidate sending intent item and set the corresponding message strength ratio value as the initial continuous selection probability.

[0008] Preferably, the method in S42 includes: S421. Construct a single-point win value. Configure the single-point win value of the historical winning items in the historical reverse competition slot to one, and configure the single-point win value of the non-winning items to zero. S422. Calculate the difference between the initial continuous selection probability and the corresponding single-point winning value for each historical candidate sending intention item, and obtain the difference calculation result. S423. If the difference calculation result is a winning item, initialize the difference calculation result of the winning item to the initial value of the slot increment strength variable of the corresponding historical candidate sending intention item in the historical reverse competition slot; if the difference calculation result is a non-winning item, initialize the initial value of the slot increment strength variable of the corresponding historical candidate sending intention item in the historical reverse competition slot to zero. S424. Identify the number of slots in which non-winning items are continuously retained in adjacent historical reverse competition slots, and record it as the historical waiting slot number sample. S425. Count the number of historical waiting slots for all non-winning items and calculate the average value to obtain the residual decay coefficient of the queue corresponding to the non-winning item. S426. The non-selection probability of the historical candidate sending intention item corresponding to the non-winning item is recursively decayed using the queue residual decay coefficient to obtain the initial value of the queue residual strength variable. The initial value of the queue residual strength variable of the historical candidate sending intention item corresponding to the winning item is initialized to zero. S427. The initial values ​​of the message strength variable, the slot incremental strength variable, and the queue residual strength variable are weighted and summed, and the calculation result is used as the sending intention strength variable of the message identifier corresponding to the historical candidate sending intention item.

[0009] Preferably, in the initial inversion round, the arbitration score corresponding to each candidate sending intention item in the current reverse competition slot is divided by the preset first temperature parameter, and then input into a differentiable normalization function for probability mapping to obtain the continuous selection probability of each candidate sending intention item. S442. Reduce the first temperature parameter according to the preset temperature decrease rule to obtain the second temperature parameter; substitute the second temperature parameter back into the differentiable normalization function corresponding to the current reverse competition slot, and keep the arbitration score value corresponding to each candidate sending intention item unchanged. S443. Recalculate the continuous selection probability corresponding to each candidate transmission intention item based on the second temperature parameter, compare the recalculated continuous selection probability with the previous round of continuous selection probability, and obtain the probability change result. S444. If the arbitration score of a candidate sending intention item is higher than the average arbitration score of candidate sending intention items in the current reverse competition slot, the corresponding candidate sending intention item is marked as a high arbitration advantage candidate; if it is lower than the average arbitration score of candidate sending intention items in the current reverse competition slot, the corresponding candidate sending intention item is marked as a low arbitration advantage candidate. S445. Determine whether the probability of consecutive selection of the high arbitration advantage candidate and the low arbitration advantage candidate after recalculation is not less than the probability of consecutive selection in the previous round. If the probability of a candidate with a high arbitration advantage is not less than the probability of continuous selection in the previous round and the probability of continuous selection corresponding to a candidate with a low arbitration advantage decreases, then the current second temperature parameter is determined to meet the probability contraction requirement, and the recalculated probability of continuous selection is taken as the result of the probability of continuous selection for the current inversion round.

[0010] Preferably, the method in S46 includes: S461. For the message strength variable, the inversion error features corresponding to all candidate sending intention items of the same message identifier are numerically accumulated to obtain the cumulative error of the message strength variable. S462. For the incremental intensity variable within the slot, the inversion error value of the corresponding candidate transmission intention item within the current reverse competition slot is used as the error update amount. S463. For the queue residual strength variable, extract the inversion error features of the same candidate transmission intention item in the previous and next reverse competition slots, calculate the error change of the inversion error features, and use the error change as the error update amount of the queue residual strength variable. The error update amounts corresponding to the message strength variable, slot incremental strength variable, and queue residual strength variable are reduced and weighted according to the preset update step size to obtain the next round of transmission intention strength variable for the candidate transmission intention item. S464. Re-execute S43 to S46 using the next round's intention strength variable to perform continuous inversion rounds; S465. Obtain the decrease in inversion error of adjacent inversion rounds. When the decrease in inversion error enters the convergence range, stop the arbitration inversion iteration and obtain the final transmission intention strength variable of each candidate transmission intention item in each reverse competition slot. S466. Arrange the final transmission intent strength variables of the same message identifier in the time sequence of the reverse competition slot to obtain the transmission intent strength trajectory of the corresponding message identifier.

[0011] Preferably, the method in S5 includes: S51. Divide the transmission intent strength trajectory corresponding to each message identifier into multiple path segments with temporal continuity. S52. Set a path activation coefficient for each path segment. According to the slot position of each path segment in the original trajectory, multiply the path segment with the corresponding path activation coefficient and then superimpose them to obtain the trajectory reconstruction result. S53. Calculate the difference between the trajectory reconstruction result and the transmission intention strength trajectory to obtain the path reconstruction error. Sum the absolute values ​​of the activation coefficients of each path to obtain the initial sparsity penalty term. S54. Calculate the difference in path activation coefficients between adjacent path segments to obtain the path continuity penalty term. Combine the path reconstruction error, the initial sparse penalty term, and the path continuity penalty term to generate the path solution objective.

[0012] Preferably, the method in S5 further includes: S55. Perform iterative optimization on the path solving objective to obtain the first round of path activation coefficients, and use the first round of path activation coefficients to generate a reweighted sparse penalty term. S56. Perform iterative optimization again on the path solution objective after writing the reweighted sparse penalty term to obtain the second round of path activation coefficients. Keep the path segments with non-zero second round path activation coefficients as candidate activation paths.

[0013] Preferably, the method in S6 includes: S61. Candidate activation paths that occupy the same reverse competition slot and are used to interpret the same transmission intention strength increment are grouped into the same competition group. S62. Compare the second-round path activation coefficients of each candidate activation path in the same competitive group, select the candidate activation path with the largest second-round path activation coefficient in the same competitive group, and compress the second-round path activation coefficients of the remaining candidate activation paths to zero. S63. Repeat the path activation coefficient comparison and path activation coefficient compression for all competing groups to obtain sparse activation paths; S64. Merge sparse activation paths belonging to the same message identifier in time to obtain message-level activation paths, and sum the path activation coefficients in the message-level activation paths to obtain the message-level activation quantity. S65. Select the message-level activation path with the largest message-level activation volume, and write the message identifier to which the corresponding message-level activation path belongs into the abnormal message identifier set. S66. Map the message identifiers in the abnormal message identifier set to the corresponding ECU nodes to obtain the abnormal node set, and output the abnormal location result.

[0014] The technical effects and advantages of the present invention regarding a dynamic analysis and anomaly localization method for automotive CAN bus load rate are as follows: (1) This invention obtains the post-arbitration observation event chain by performing event-based encoding on the CAN message stream, and constructs a reverse contention slot by combining the service occupancy boundary chain, projects the observation events as candidate sending intention items, and generates a reverse arbitration queue; then, the candidate sending intention items are arranged by the arbitration sequence chain, so that the load rate analysis object is transformed from the performance of the sent messages to the sending intentions that may participate in bus contention before arbitration; it can more accurately reflect the actual contention of each message for bus resources in the dynamic load rate analysis stage, and improve the accuracy of abnormal load source analysis.

[0015] (2) By differentiable normalization and backpropagation of inversion error, the transmission intent intensity trajectory corresponding to each message identifier is obtained; the transmission intent intensity trajectory is further divided into time-continuous path segments, and by combining path reconstruction, initial sparsity penalty, path continuity penalty and reweighted sparsity solution, sparse activation paths that can explain continuous load changes are retained; finally, the sparse activation paths are aggregated to the message identifier and mapped to the ECU node; the stability and interpretability of abnormal message identifier and abnormal ECU node location are improved. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process of a dynamic analysis and anomaly localization method for automotive CAN bus load rate according to the present invention.

[0017] Figure 2 This is a flowchart illustrating step S4 of a method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to the present invention.

[0018] Figure 3 This is a flowchart illustrating step S5 of the method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This application provides a method for dynamic analysis and anomaly localization of CAN bus load rate in automobiles. By utilizing differentiable arbitration inversion to obtain the transmission intent strength trajectory of each message identifier, and then extracting the source of abnormal load through reweighted sparse activation paths, the abnormal message identifiers are finally mapped to ECU nodes. This invention achieves dynamic analysis and anomaly localization of CAN bus load rate.

[0021] Please see Figure 1 , Figure 2 and Figure 3 In this embodiment of the invention, the method for dynamic analysis and anomaly localization of automotive CAN bus load rate is implemented in detail through the following steps: S1. Obtain the CAN message stream of the vehicle to be analyzed and perform event-based encoding to obtain the arbitrated observation event chain; In one embodiment of the present invention, the method in S1 specifically includes: Using the reception time as the sorting index, each frame of the CAN message stream to be analyzed is sorted sequentially; a corresponding observation event number is constructed based on each sorted frame, and different observation events are connected according to the order of the event numbers to obtain the initial event chain; the reception time difference between adjacent observation events is calculated, and the reception time difference is written into the adjacent event connection edge of the initial event chain, and the initial event chain after writing the adjacent event connection edge is used as the arbitrated observation event chain.

[0022] S2. Perform frame length processing and arbitration sequence analysis on the message frames and message identifiers in the post-arbitration observation event chain to obtain the service occupancy boundary chain and arbitration sequence chain. In one embodiment of the present invention, the method in S2 includes: The message frame length corresponding to each observation event is converted into bus transmission bits. The bus transmission bits are divided by the CAN bus bit rate to obtain the service occupancy duration of the corresponding observation event. The service occupancy durations are arranged according to the event order of the post-arbitration observation event chain to obtain a service occupancy duration sequence. Starting from the first observation event in the post-arbitration observation event chain, the service occupancy duration sequence is sequentially accumulated to obtain the service end position corresponding to each observation event. The service end positions are connected according to the event order of the observation events to obtain a service occupancy boundary chain. The service end position in the service occupancy boundary chain represents the time boundary when the CAN bus completes one message transmission and re-enters arbitration contention. The message identifier carried by each observation event is extracted. Using the arbitration rule in the CAN bus that the smaller the message identifier value, the higher the arbitration priority, each message identifier is converted into a corresponding arbitration sequence number. The arbitration sequence numbers are arranged according to the event order of the post-arbitration observation event chain to obtain an arbitration sequence number chain.

[0023] S3. By analyzing the temporal correlation between the post-arbitration observation event chain and the service occupancy boundary chain, a reverse arbitration queue is generated; The methods in S3 include: S31. Using the time boundary of every two service end positions in the service occupation boundary chain as the slot reference, propose a reverse competition slot. S32. At the current reverse contention slot, the observation event that actually occupies the current message service area in the post-arbitration observation event chain is taken as the target observation event of the current reverse contention slot. S33. Starting from the event number of the target observed event in the post-arbitration observed event chain, traverse the observed events along the subsequent arrangement direction of the post-arbitration observed event chain. S34. Calculate the time interval between the reception time of the observed event being traversed and the reception time of the target observed event; S35. When the time interval is not greater than the preset waiting time limit, the observed events to be traversed are included in the candidate pool corresponding to the current reverse competition slot. When the time interval exceeds the preset waiting time limit, stop traversing subsequent observation events for the current inverse competition slot; S36. Each observation event in the candidate pool is defined as a candidate sending intention item. Different candidate sending intention items are arranged from high to low according to the arbitration order to obtain the reverse arbitration queue corresponding to the current reverse competition slot. In one embodiment of the present invention, for example, the post-arbitration observation event chain includes, in sequence: E1: reception time 10.000ms, message identifier 0x100; E2: reception time 10.260ms, message identifier 0x180; E3: reception time 10.510ms, message identifier 0x300; E4: reception time 11.200ms, message identifier 0x120; the service occupancy boundary chain includes B1, B2, and B3, and the current message service interval is obtained between B1 and B2, corresponding to the reverse contention slot C1; if E2 actually occupies the message service interval, then E2 is taken as the target observation event of C2; Starting from E2, traverse the observed events in the subsequent direction and calculate the reception time interval between the subsequent observed events and E2; the interval between E3 and E2 is 0.250ms, and the interval between E4 and E2 is 0.940ms; if the preset waiting time upper limit is 0.300ms, then E2 and E3 enter the candidate pool corresponding to C2, and E4 does not enter the candidate pool; then E2 and E3 are respectively designated as candidate transmission intention items, and sorted according to the arbitration sequence corresponding to the message identifier to obtain the reverse arbitration queue corresponding to C1.

[0024] The preset waiting time limit can be obtained from the historical normal CAN message stream: first, count the reception time interval of adjacent observation events in the historical observation event chain, then filter the short-term subsequent waiting interval, and finally select the sorted preset percentile value as the waiting time limit; for example, if the 90th percentile value of the short-term subsequent waiting interval is 0.300ms, then 0.300ms is set as the preset waiting time limit.

[0025] S4. By analyzing the correlation between the queue arrangement characteristics in the reverse arbitration queue, the arbitration advantage characteristics in the arbitration sequence chain, and the boundary continuity characteristics in the service occupancy boundary chain, arbitration inversion is performed on the observed events to obtain the sending intent strength trajectory corresponding to each message identifier. The methods in S4 include: S41. Extract historical normal message segments, initialize the variable channel for the candidate sending intent items in the reverse arbitration queue, and obtain the initial value of the message strength variable corresponding to each message identifier. The method of S41 includes: S411. Obtain historical CAN message streams that belong to the same CAN bus channel as the CAN message stream to be analyzed, and extract the set of historical normal message segments. Among them, the set of historical normal message segments includes historical message segments that have been tested and marked as having no message injection, no diagnostic scan, and no node out of control. S412. Divide the messages in the set of historical normal message segments into channels according to the message identifier. Perform steps S2 and S3 for each historical variable channel to obtain the historical reverse competition slot and historical candidate transmission intention item. S413. Combine the historical candidate sending intent items in each historical reverse competition slot to obtain a historical candidate group. Generate a set of historical candidate groups based on the historical candidate groups of different historical reverse competition slots. S414. In each historical candidate group, read the actual sent observation events of the historical reverse competition slot and record them as the historical winners of the corresponding historical reverse competition slot. S415. Count the number of times different historical winners appear in the historical candidate group set according to the message identifier, and obtain the candidate exposure number of the corresponding message identifier. In one embodiment of the present invention, for example, if message identifier 0x100 appears 100 times in the entire historical candidate set, then its candidate exposure count is 100; if message identifier 0x100 appears 100 times and corresponds to an actual successful observation event 80 times, then its historical winning count is 80.

[0026] S416. Count the number of times each message identifier is counted as a historical winning item to obtain the historical winning count of the corresponding message identifier; S417. Perform data transformation processing on each candidate exposure count and historical winning count to obtain the initial value of the message strength variable for each message identifier; In one embodiment of the present invention, data transformation includes a smooth scaling transformation of the number of candidate exposures and the number of historical wins; The smoothing ratio transformation is used to avoid extreme values ​​in the ratio when the number of historical wins is 0 or equal to the number of candidate exposures. For example, if a message identifier has 100 candidate exposures and 20 historical wins, the historical win ratio after smoothing ratio transformation is approximately 0.206; if the number of candidate exposures is 10 and the number of historical wins is 0, the non-zero historical win ratio after smoothing ratio transformation is approximately 0.083.

[0027] S418. Based on the initial values ​​of the message strength variables in the historical reverse competition slots, calculate the proportion of the initial value of the message strength variable for each message identifier, and record it as the message strength ratio value. S419. Read the message identifier of the historical candidate sending intent item and set the corresponding message strength ratio value as the initial continuous selection probability.

[0028] S42. By analyzing the winning compensation features and cross-slot retention features of historical candidate sending intent items, configure the initial values ​​of the slot incremental strength variable and the queue residual strength variable and merge them with the initial value of the message strength variable to obtain the sending intent strength variable; The method of S42 includes: S421. Construct a single-point win value. Configure the single-point win value of the historical winning items in the historical reverse competition slot to one, and configure the single-point win value of the non-winning items to zero. S422. Calculate the difference between the initial continuous selection probability and the corresponding single-point winning value for each historical candidate sending intention item, and obtain the difference calculation result. S423. If the difference calculation result is a winning item, the difference calculation result of the winning item is initialized to the initial value of the intra-slot incremental intensity variable of the corresponding historical candidate sending intention item in the historical reverse competition slot; if the difference calculation result is a non-winning item, the initial value of the intra-slot incremental intensity variable of the corresponding historical candidate sending intention item in the historical reverse competition slot is initialized to zero; wherein, by comparing the initial continuous selection probability with the single-point winning value, the deviation between the current arbitration result and the actual sending result is quantified, and the local compensation amount of the winning item is obtained based on the deviation, and the local compensation amount is used as the initial value of the intra-slot incremental intensity variable of the historical reverse competition slot where the winning item is located; S424. Identify the number of slots in which a non-winning item is continuously retained in adjacent historical reverse competition slots, and record it as the historical waiting slot number sample; for example, if a historical candidate sending intention item fails to obtain the sending right in the 5th historical reverse competition slot, and continues to appear in the 6th and 7th historical reverse competition slots, until it obtains the sending right in the 8th historical reverse competition slot, then the 6th, 7th, and 8th historical reverse competition slots are the historical waiting slot number samples corresponding to the historical candidate sending intention item; S425. Count the number of historical waiting slots for all non-winning items and calculate the average value to obtain the residual decay coefficient of the queue corresponding to the non-winning item. S426. The non-selection probability of the historical candidate sending intention item corresponding to the non-winning item is recursively decayed using the queue residual decay coefficient to obtain the initial value of the queue residual strength variable. The initial value of the queue residual strength variable of the historical candidate sending intention item corresponding to the winning item is initialized to zero. In one embodiment of the present invention, for example, when the probability of a historical candidate sending intention item not being selected in the first historical reverse contention slot is 0.8 and the queue residual attenuation coefficient is 0.7, the residual probability of the historical candidate sending intention item being passed to the second historical reverse contention slot is 0.8 × 0.7 = 0.56; if it is still not selected in the second historical reverse contention slot, the residual probability of being passed to the third historical reverse contention slot is obtained by calculating 0.56 × 0.7 = 0.392, and the residual probability is calculated recursively slot by slot in the same way to obtain the initial value of the queue residual strength variable.

[0029] S427. The initial values ​​of the message strength variable, the slot incremental strength variable, and the queue residual strength variable are weighted and summed, and the calculation result is used as the sending intention strength variable of the message identifier corresponding to the historical candidate sending intention item. Among them, the message-level reference strength variable is shared among all the anti-competition slots of the variable channel, and is used to characterize the basic bus contention pressure of the same message identifier in the CAN message stream to be analyzed. The slot-incremental strength variable only applies to the candidate transmission intention item in the corresponding reverse contention slot, and is used to characterize the local strength increase of the candidate transmission intention item relative to the message-level baseline strength variable in its respective reverse contention slot; The queue residual strength variable is used to characterize the residual competitive pressure of candidate sending intention items that were not selected by arbitration in the previous reverse competitive slot continuing to be retained in the next reverse competitive slot; In one embodiment of the present invention, the weights of the initial values ​​of the message strength variable, the slot increment strength variable, and the queue residual strength variable are set to 0.5, 0.3, and 0.2, respectively. The weights are determined by the historical normal CAN message stream, specifically by selecting historical normal message segments belonging to the same CAN bus channel as the CAN message stream to be analyzed, statistically analyzing the contribution of historical transmission intent strength to the actual arbitration result, and determining each weight value according to the weight combination corresponding to the minimum historical inversion error.

[0030] S43. Read the message identifier in the current candidate sending intent item, mark the corresponding sending intent strength variable as the arbitration scoring strength value, and accumulate the arbitration scoring strength value and arbitration sequence to obtain the arbitration scoring value of the corresponding candidate sending intent item. S44. Perform differentiable normalization on the arbitration scores of different candidate sending intention items to obtain the continuous selection probability of each candidate sending intention item. The methods in S44 include: S441. In the initial inversion round, the arbitration score corresponding to each candidate sending intention item in the current reverse competition slot is divided by the preset first temperature parameter, and then input into the differentiable normalization function for probability mapping to obtain the continuous selection probability of each candidate sending intention item. In one embodiment of the present invention, a differentiable normalization function is used to convert the arbitration scores corresponding to multiple candidate transmission intention items into a continuous probability distribution with a sum of 1, using the Softmax normalization function; the first temperature parameter is determined by selecting historical normal message segments belonging to the same CAN bus channel as the CAN message stream to be analyzed, statistically analyzing the arbitration score distribution of candidate transmission intention items in the historical reverse competition slot, and determining the temperature parameter that can ensure that multiple candidate transmission intention items in the initial inversion round can retain a non-zero continuous selection probability and have the smallest average score difference as the first temperature parameter; The significance of introducing the first temperature parameter lies in controlling the smoothness of the distribution when the arbitration score value is transformed into the continuous selection probability. This allows the reverse arbitration inversion to simultaneously observe the contribution relationship of multiple candidate transmission intention items to the target observation event in the initial stage, avoiding the premature concentration of the continuous selection probability on a single candidate transmission intention item due to excessively high individual arbitration scores. This improves the identifiability of the transmission intention intensity variable and the stability of the inversion. The temperature parameter is used instead of directly setting a fixed scaling factor or fixed bias parameter on the arbitration score value because the temperature parameter can adjust the relative differences between all candidate transmission intention items on a uniform scale. Under the premise of keeping the arbitration score ranking relationship unchanged, it continuously controls the probability distribution to transition from a smooth state to a concentrated state. This does not change the relative competition relationship corresponding to the CAN arbitration priority, and can gradually approach the actual CAN hard arbitration result through the subsequent temperature reduction process.

[0031] S442. Reduce the first temperature parameter according to the preset temperature decrease rule to obtain the second temperature parameter; substitute the second temperature parameter back into the differentiable normalization function corresponding to the current reverse competition slot, and keep the arbitration score value corresponding to each candidate sending intention item unchanged. In one embodiment of the present invention, the preset temperature reduction rule adopts a fixed ratio reduction method. For example, the first temperature parameter is set to 1.0 and reduced by 0.8 in each round. Then the second temperature parameter is 0.8, and the subsequent temperature parameters are 0.64 and 0.512 respectively, and the temperature parameters gradually decrease.

[0032] S443. Recalculate the continuous selection probability corresponding to each candidate transmission intention item based on the second temperature parameter, compare the recalculated continuous selection probability with the previous round of continuous selection probability, and obtain the probability change result. S444. If the arbitration score of a candidate sending intention item is higher than the average arbitration score of candidate sending intention items in the current reverse competition slot, the corresponding candidate sending intention item is marked as a high arbitration advantage candidate; if it is lower than the average arbitration score of candidate sending intention items in the current reverse competition slot, the corresponding candidate sending intention item is marked as a low arbitration advantage candidate. S445. Determine whether the probability of consecutive selection of the high arbitration advantage candidate and the low arbitration advantage candidate after recalculation is not less than the probability of consecutive selection in the previous round. If the probability of continuous selection of a candidate with high arbitration advantage is not less than that of the previous round and the probability of continuous selection of a candidate with low arbitration advantage decreases, then the current second temperature parameter is determined to meet the probability contraction requirement, and the recalculated probability of continuous selection is taken as the result of the probability of continuous selection of the current inversion round. Otherwise, if the current second temperature parameter does not meet the probability contraction requirement, the temperature parameter is further reduced and the continuous selection probability calculation and probability change judgment process is repeated until the probability contraction requirement is met, and the continuous selection probability of each candidate sending intention item in the current reverse competition slot is obtained.

[0033] S45. Based on the winning attribute and continuous selection probability of each candidate sending intention item in the current reverse competition slot, construct the inversion error characteristics of the current reverse competition slot. In one embodiment of the present invention, the method of S45 specifically includes: In the current reverse competition slot, if the candidate sending intent item is the winning item, the target vector dimension is assigned a value of one; if the candidate sending intent item is not the winning item, the target vector dimension is assigned a value of zero. The target vector dimension value and the successive selection probability of each candidate sending intent item in the current reverse competition slot are concatenated in the same order to obtain the target vector and the probability vector respectively. Calculate the difference between the continuous selection probability at different positions in the probability vector and the corresponding position value in the target vector, and combine the differences at different positions to obtain the inversion error characteristics of the current reverse competition slot; S46. The inversion error characteristics are backed to the message strength variable, slot incremental strength variable and queue residual strength variable according to the variable type and arbitration inversion is performed until the inversion error converges. The converged transmission intention strength variable is connected according to the message identifier and the reverse competition slot time order to form the transmission intention strength trajectory. The methods in S46 include: S461. For the message strength variable, the inversion error features corresponding to all candidate sending intention items of the same message identifier are numerically accumulated to obtain the cumulative error of the message strength variable. S462. For the incremental intensity variable within the slot, the inversion error value of the corresponding candidate transmission intention item within the current reverse competition slot is used as the error update amount. S463. For the queue residual strength variable, extract the inversion error features of the same candidate transmission intention item in the previous and next reverse competition slots, calculate the error change of the inversion error features, and use the error change as the error update amount of the queue residual strength variable. The error update amounts corresponding to the message strength variable, slot incremental strength variable, and queue residual strength variable are reduced according to the preset update step size and then weighted and fused to obtain the next round of transmission intention strength variable for the candidate transmission intention item; wherein, the same weights as above are used. S464. Re-execute S43 to S46 using the next round's intention strength variable to perform continuous inversion rounds; S465. Obtain the decrease in inversion error of adjacent inversion rounds. When the decrease in inversion error enters the convergence range, stop the arbitration inversion iteration and obtain the final transmission intention strength variable of each candidate transmission intention item in each reverse competition slot. The preset convergence threshold is determined through historical normal CAN message streams. Specifically, historical normal CAN message streams are selected to perform differentiable arbitration inversion, the decrease in inversion error corresponding to each iteration is recorded, the average decrease in error during the continuous stable phase is statistically analyzed, and the average value is used as the preset convergence threshold. For example, if the preset convergence threshold obtained from the historical normal CAN message stream statistics is 0.001, when the inversion errors corresponding to the 21st to 25th iterations are 0.0185, 0.0179, 0.0174, 0.0170, and 0.0167 respectively, the decrease in error between adjacent iterations is 0.0006, 0.0005, 0.0004, and 0.0003 respectively, all of which are less than 0.001. Therefore, the inversion error is determined to have entered the convergence range.

[0034] S466. Arrange the final transmission intent strength variables of the same message identifier in the time sequence of the reverse competition slot to obtain the transmission intent strength trajectory of the corresponding message identifier.

[0035] To address the issue that existing automotive CAN bus load analysis often focuses on post-arbitration message observations, making it difficult to distinguish between actual transmission pressure and the impact of arbitration queuing, this invention performs event-based encoding on the CAN message stream to obtain a post-arbitration observation event chain. This chain, combined with a service occupancy boundary chain, constructs a reverse contention slot, projecting the observed events as candidate transmission intentions and generating a reverse arbitration queue. Subsequently, the arbitration sequence chain is used to arrange the candidate transmission intentions, transforming the load rate analysis object from the performance of already transmitted messages to transmission intentions that may participate in bus contention before arbitration. By using the above methods, we can more accurately reflect the actual competition for bus resources among various messages during the load rate dynamic analysis phase, thereby improving the accuracy of abnormal load source analysis.

[0036] S5. Decompose the transmission intent strength trajectory into multiple path segments, and obtain sparse activation paths by reconstructing the path segments and competing for slots in the same slot. The methods in S5 include: S51. Divide the transmission intent strength trajectory corresponding to each message identifier into multiple path segments with temporal continuity. In one embodiment of the present invention, for example, the system obtains the transmission intention strength trajectory of message identifier 0x18A in 12 consecutive back contention slots: the system segments the transmission intention strength trajectory with a fixed length and a sliding step size. For example, if the path segment length is set to 3 back contention slots and the sliding step size is 1 back contention slot, then the transmission intention strength trajectory of message identifier 0x18A is segmented into the following path segments: P1: The strengths of slots 1 to 3 are 0.02, 0.03, and 0.04, respectively; P2: The strengths of slots 2 to 4 are 0.03, 0.04, and 0.26, respectively; P3: The strengths of slots 3 to 5 are 0.04, 0.18, and 0.35, respectively.

[0037] S52. Set a path activation coefficient for each path segment. According to the slot position of each path segment in the original trajectory, multiply the path segment with the corresponding path activation coefficient and then superimpose them to obtain the trajectory reconstruction result. In one embodiment of the present invention, the average intensity within a path segment is used as the initial value of the path activation coefficient, and the path activation coefficients of P1 to P3 are respectively set and denoted as A1 to A3; the path activation coefficient is used to represent the contribution of the corresponding path segment in reconstructing the transmission intention intensity trajectory; for example, the initial value of A1 is 0.03; when the average intensity of the segment is used for initialization, segments with higher intensity and continuity will have a higher chance of being retained in the first round of solving; P1 covers slots 1 through 3, P2 covers slots 2 through 4, and P3 covers slots 3 through 5. If the path activation coefficients of P4, P5, and P6 are high, then slots 2, 3, and 4 will be reconstructed by multiple path segments. For each competing slot, the system accumulates the contributions of all path segments covering that slot to obtain the reconstruction intensity of that slot. The reconstruction intensities of all slots are arranged in chronological order to obtain the trajectory reconstruction result. For example, after one initial superposition, the reconstruction result may be: 0.03, 0.04, 0.07; the closer the reconstruction result is to the original transmission intent intensity trajectory, the better the current path segment and its path activation coefficient can explain the original trajectory.

[0038] S53. Calculate the difference between the trajectory reconstruction result and the transmission intention strength trajectory to obtain the path reconstruction error. Sum the absolute values ​​of the activation coefficients of each path to obtain the initial sparsity penalty term. In one embodiment of the present invention, for example, the original transmission intent strength trajectory is: 0.02, 0.03, 0.04; the reconstruction result is: 0.03, 0.04, 0.07; Each slot then receives a reconstruction difference value, for example, the difference value for slot 1 is 0.01, the difference value for slot 2 is 0.01, and the difference value for slot 3 is 0.03. The system summarizes the reconstruction differences of all slots to obtain the path reconstruction error. Among them, the penalty term is used to compress unnecessary path segments so that only a small number of path segments that can explain the main transmission intent intensity changes are retained.

[0039] S54. Calculate the difference in path activation coefficients between adjacent path segments to obtain the path continuity penalty term. Combine the path reconstruction error, the initial sparse penalty term, and the path continuity penalty term to generate the path solution objective. S55. Perform iterative optimization on the path solving objective to obtain the first round of path activation coefficients, and use the first round of path activation coefficients to generate a reweighted sparse penalty term. In one embodiment of the present invention, a proximal gradient iteration method with non-negative constraints is used to perform iterative optimization on the path solving objective to obtain the first round of path activation coefficients, specifically including: The trajectory of the sent intent strength is denoted as a trajectory vector. Where y is the transmission intent strength trajectory vector; N is the number of reverse contention slots; is the transmission intent strength in the Nth reverse contention slot; T is the transpose symbol.

[0040] S551. Embed the path segments obtained in S51 into time positions of the same length as the trajectory vector to obtain the path segment matrix. Where D is the path segment matrix; M is the number of path segments; This is the column vector after the Mth path segment is embedded into the full time length; each column of the path segment matrix D corresponds to one path segment; S552. Set non-negative path activation coefficients for each path segment, and denote the path activation coefficient vector as follows: The difference matrix between adjacent path segments is constructed as follows: Let be the path activation coefficient of the i-th path segment; In the formula, This is the difference matrix between adjacent path segments; S553. Establish the path solving objective. The path solving formula is as follows: ; In the formula, Indicates path reconstruction error; This is represented as the initial sparsity penalty term; Indicates the path continuation penalty term; Sparse weights; Continuous weights; The trajectory is reconstructed from path segments; S554, Set Step Size Based on the formula The calculation yields the result; where, Represents the largest eigenvalue of a matrix; This is the transpose of the path segment matrix D; This is the transpose of the difference matrix R; S555. In the r-th iteration, the smooth gradient of the r-th iteration is calculated using the smooth gradient formula, as follows: ; In the formula, This represents the smoothed gradient in the r-th iteration; r is the iteration number. Let be the path activation coefficient vector at the r-th iteration; This represents the gradient of the reconstruction error term with respect to the path activation coefficient vector a; This represents the gradient of the path continuity penalty term with respect to the path activation coefficient vector a; S556. Update the path activation coefficient vector along the negative gradient direction to obtain the intermediate vector. The update formula is as follows: In the formula, Let be the intermediate vector in the r-th iteration; the formula means: first, perform a normal gradient update on the path activation coefficients along the direction of the fastest descent of the objective function to obtain the intermediate result before the application of sparse constraints and non-negativity constraints; S557. Perform non-negative soft thresholding on the intermediate vector to obtain the path activation coefficients for the next round. The formula for non-negative soft thresholding is: ; In the formula, This represents the path activation coefficient of the i-th path segment after the (r+1)-th iteration; Represents the intermediate vector The i-th component; Repeat steps S555 to S557 until the change in the path activation coefficient vector between two adjacent rounds is less than the convergence threshold, or the number of iterations reaches the maximum number of iterations. Use the path activation coefficient vector obtained in the last round as the path activation coefficient of the first round. The convergence threshold is the distribution of the change in the path activation coefficient between two adjacent rounds in the same iterative algorithm of the historical normal CAN message flow, and the 95th percentile of this change in the stable phase is taken. The maximum number of iterations is the upper quantile of the number of iterations required to achieve convergence for most historical samples under the convergence threshold.

[0041] S56. Perform iterative optimization again on the path solution objective after writing the reweighted sparse penalty term to obtain the second round of path activation coefficients. Keep the path segments with non-zero second round path activation coefficients as candidate activation paths.

[0042] S6. Aggregate the message identifiers belonging to the sparse activation path into abnormal message identifiers, map them to the corresponding ECU nodes, and output the abnormal location results. The methods in S6 include: S61. Candidate activation paths that occupy the same reverse competition slot and are used to interpret the same transmission intention strength increment are grouped into the same competition group. S62. Compare the second-round path activation coefficients of each candidate activation path in the same competitive group, select the candidate activation path with the largest second-round path activation coefficient in the same competitive group, and compress the second-round path activation coefficients of the remaining candidate activation paths to zero. S63. Repeat the path activation coefficient comparison and path activation coefficient compression for all competing groups to obtain sparse activation paths; S64. Merge sparse activation paths belonging to the same message identifier in time to obtain message-level activation paths, and sum the path activation coefficients in the message-level activation paths to obtain the message-level activation quantity. S65. Select the message-level activation path with the largest message-level activation volume, and write the message identifier to which the corresponding message-level activation path belongs into the abnormal message identifier set. S66. Map the message identifiers in the abnormal message identifier set to the corresponding ECU nodes to obtain the abnormal node set, and output the abnormal location result.

[0043] Based on the ability to invert the sending intent, and addressing the problem that existing anomaly localization methods rely heavily on single anomaly results or message-level statistical results, making it difficult to reliably extract the source of persistent anomalies, this invention obtains the sending intent strength trajectory corresponding to each message identifier through differentiable normalization processing and inversion error backpropagation. Furthermore, the sending intent strength trajectory is segmented into time-continuous path segments, and by combining path reconstruction, initial sparse penalty, path continuity penalty, and reweighted sparse solution, sparse active paths that can explain continuous load changes are retained. Finally, the sparse active paths are aggregated to the message identifier and mapped to the ECU node. The above methods show that the anomaly location results are not directly derived from load changes at a single moment, but are naturally obtained from the continuous transmission intent strength path, which improves the stability and interpretability of anomaly message identifiers and anomaly ECU node location.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0045] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0046] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0047] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for dynamic analysis and anomaly localization of automotive CAN bus load rate, characterized in that, include: S1. Obtain the CAN message stream of the vehicle to be analyzed and perform event-based encoding to obtain the arbitrated observation event chain; S2. Perform frame length processing and arbitration sequence analysis on the message frames and message identifiers in the post-arbitration observation event chain to obtain the service occupancy boundary chain and arbitration sequence chain. S3. By analyzing the temporal correlation between the post-arbitration observation event chain and the service occupancy boundary chain, a reverse arbitration queue is generated; S4. By analyzing the correlation between the queue arrangement characteristics in the reverse arbitration queue, the arbitration advantage characteristics in the arbitration sequence chain, and the boundary continuity characteristics in the service occupancy boundary chain, arbitration inversion is performed on the observed events to obtain the sending intent strength trajectory corresponding to each message identifier. S5. Decompose the transmission intent strength trajectory into multiple path segments, and obtain sparse activation paths by reconstructing the path segments and competing for slots in the same slot. S6. Aggregate the message identifiers to which the sparse activation path belongs into abnormal message identifiers, map them to the corresponding ECU nodes, and output the abnormal location results.

2. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 1, characterized in that, The method in S3 includes: S31. Using the end position of each service in the service occupancy boundary chain as the slot reference, propose a reverse competition slot; S32. Create a set of paired event numbers and initialize the set of paired event numbers to an empty set; S33. Traverse the inverse contention slots in the direction from the tail to the head of the service occupancy boundary chain, and mark the traversal target as the current inverse contention slot. S34. At the current inverse contention slot, traverse the observation event chain after arbitration and receive observation events whose time is earlier than the service end position corresponding to the current inverse contention slot. For each observed event, determine whether the event number already exists in the set of paired event numbers; If the event number of the observed event does not exist in the set of paired event numbers, then the corresponding observed event will be included in the candidate pool and the paired set corresponding to the current reverse competition slot; If the event number of the observed event already exists in the set of paired event numbers, then skip the observed event; S35. Each observation event in the candidate pool is designated as a candidate sending intention item. Different candidate sending intention items are arranged from high to low according to the arbitration order to obtain the reverse arbitration queue corresponding to the current reverse competition slot.

3. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 2, characterized in that, The method in S4 includes: S41. Extract historical normal message segments, initialize the variable channel for the candidate sending intent items in the reverse arbitration queue, and obtain the initial value of the message strength variable corresponding to each message identifier. S42. By analyzing the winning compensation features and cross-slot retention features of historical candidate sending intent items, configure the initial values ​​of the slot incremental strength variable and the queue residual strength variable and merge them with the initial value of the message strength variable to obtain the sending intent strength variable; S43. Read the message identifier in the current candidate sending intent item, mark the corresponding sending intent strength variable as the arbitration scoring strength value, and accumulate the arbitration scoring strength value and arbitration sequence to obtain the arbitration scoring value of the corresponding candidate sending intent item. S44. Perform differentiable normalization on the arbitration scores of different candidate sending intention items to obtain the continuous selection probability of each candidate sending intention item. S45. Based on the winning attribute and continuous selection probability of each candidate sending intention item in the current reverse competition slot, construct the inversion error characteristics of the current reverse competition slot. S46. The inversion error characteristics are backed to the message strength variable, slot incremental strength variable and queue residual strength variable according to the variable type and arbitration inversion is performed until the inversion error converges. The converged transmission intention strength variables are connected according to the message identifier and the reverse competition slot time order to form the transmission intention strength trajectory.

4. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 3, characterized in that, The method in S41 includes: S411. Obtain historical CAN message streams that belong to the same CAN bus channel as the CAN message stream to be analyzed, and extract the set of historical normal message segments. S412. Divide the messages in the set of historical normal message segments into channels according to the message identifier. Perform steps S2 and S3 for each historical variable channel to obtain the historical reverse competition slot and historical candidate transmission intention item. S413. Combine the historical candidate sending intent items in each historical reverse competition slot to obtain a historical candidate group. Generate a set of historical candidate groups based on the historical candidate groups of different historical reverse competition slots. S414. In each historical candidate group, read the actual sent observation events of the historical reverse competition slot and record them as the historical winners of the corresponding historical reverse competition slot. S415. Count the number of times different historical winners appear in the historical candidate group set according to the message identifier, and obtain the candidate exposure number of the corresponding message identifier. S416. Count the number of times each message identifier is counted as a historical winning item to obtain the historical winning count of the corresponding message identifier; S417. Perform data transformation processing on each candidate exposure count and historical winning count to obtain the initial value of the message strength variable for each message identifier; S418. Based on the initial values ​​of the message strength variables in the historical reverse competition slots, calculate the proportion of the initial value of the message strength variable for each message identifier, and record it as the message strength ratio value. S419. Read the message identifier of the historical candidate sending intent item and set the corresponding message strength ratio value as the initial continuous selection probability.

5. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 4, characterized in that, The method in S42 includes: S421. Construct a single-point win value. Configure the single-point win value of the historical winning items in the historical reverse competition slot to one, and configure the single-point win value of the non-winning items to zero. S422. Calculate the difference between the initial continuous selection probability and the corresponding single-point winning value for each historical candidate sending intention item, and obtain the difference calculation result. S423. If the difference calculation result is a winning item, initialize the difference calculation result of the winning item to the initial value of the slot increment strength variable of the corresponding historical candidate sending intention item in the historical reverse competition slot; if the difference calculation result is a non-winning item, initialize the initial value of the slot increment strength variable of the corresponding historical candidate sending intention item in the historical reverse competition slot to zero. S424. Identify the number of slots in which non-winning items are continuously retained in adjacent historical reverse competition slots, and record it as the historical waiting slot number sample. S425. Count the number of historical waiting slots for all non-winning items and calculate the average value to obtain the residual decay coefficient of the queue corresponding to the non-winning item. S426. The non-selection probability of the historical candidate sending intention item corresponding to the non-winning item is recursively decayed using the queue residual decay coefficient to obtain the initial value of the queue residual strength variable. The initial value of the queue residual strength variable of the historical candidate sending intention item corresponding to the winning item is initialized to zero. S427. The initial values ​​of the message strength variable, the slot incremental strength variable, and the queue residual strength variable are weighted and summed, and the calculation result is used as the sending intention strength variable of the message identifier corresponding to the historical candidate sending intention item.

6. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 5, characterized in that, The method in S44 includes: S441. In the initial inversion round, the arbitration score corresponding to each candidate sending intention item in the current reverse competition slot is divided by the preset first temperature parameter, and then input into the differentiable normalization function for probability mapping to obtain the continuous selection probability of each candidate sending intention item. S442. Reduce the first temperature parameter according to the preset temperature decrease rule to obtain the second temperature parameter; substitute the second temperature parameter back into the differentiable normalization function corresponding to the current reverse competition slot, and keep the arbitration score value corresponding to each candidate sending intention item unchanged. S443. Recalculate the continuous selection probability corresponding to each candidate transmission intention item based on the second temperature parameter, compare the recalculated continuous selection probability with the previous round of continuous selection probability, and obtain the probability change result. S444. If the arbitration score of a candidate sending intention item is higher than the average arbitration score of candidate sending intention items in the current reverse competition slot, the corresponding candidate sending intention item is marked as a high arbitration advantage candidate; if it is lower than the average arbitration score of candidate sending intention items in the current reverse competition slot, the corresponding candidate sending intention item is marked as a low arbitration advantage candidate. S445. Determine whether the probability of consecutive selection of the high arbitration advantage candidate and the low arbitration advantage candidate after recalculation is not less than the probability of consecutive selection in the previous round. If the probability of a candidate with a high arbitration advantage is not less than the probability of continuous selection in the previous round and the probability of continuous selection corresponding to a candidate with a low arbitration advantage decreases, then the current second temperature parameter is determined to meet the probability contraction requirement, and the recalculated probability of continuous selection is taken as the result of the probability of continuous selection for the current inversion round.

7. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 6, characterized in that, The method in S46 includes: S461. For the message strength variable, the inversion error features corresponding to all candidate sending intention items of the same message identifier are numerically accumulated to obtain the cumulative error of the message strength variable. S462. For the incremental intensity variable within the slot, the inversion error value of the corresponding candidate transmission intention item within the current reverse competition slot is used as the error update amount. S463. For the queue residual strength variable, extract the inversion error features of the same candidate transmission intention item in the previous and next reverse competition slots, calculate the error change of the inversion error features, and use the error change as the error update amount of the queue residual strength variable. The error update amounts corresponding to the message strength variable, slot incremental strength variable, and queue residual strength variable are reduced according to the preset update step size and then weighted and fused to obtain the next round of transmission intention strength variable for the candidate transmission intention item. S464. Re-execute S43 to S46 using the next round's intention strength variable to perform continuous inversion rounds; S465. Obtain the decrease in inversion error of adjacent inversion rounds. When the decrease in inversion error enters the convergence range, stop the arbitration inversion iteration and obtain the final transmission intention strength variable of each candidate transmission intention item in each reverse competition slot. S466. Arrange the final transmission intent strength variables of the same message identifier in the time sequence of the reverse competition slot to obtain the transmission intent strength trajectory of the corresponding message identifier.

8. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 7, characterized in that, The method in S5 includes: S51. Divide the transmission intent strength trajectory corresponding to each message identifier into multiple path segments with temporal continuity. S52. Set a path activation coefficient for each path segment. Multiply the path segment by the corresponding path activation coefficient according to the slot position of each path segment in the original trajectory and then add them together to obtain the trajectory reconstruction result. S53. Calculate the difference between the trajectory reconstruction result and the transmission intention strength trajectory to obtain the path reconstruction error. Sum the absolute values ​​of the activation coefficients of each path to obtain the initial sparsity penalty term. S54. Calculate the difference in path activation coefficients between adjacent path segments to obtain the path continuity penalty term. Combine the path reconstruction error, the initial sparse penalty term, and the path continuity penalty term to generate the path solution objective.

9. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 8, characterized in that, The method in S5 further includes: S55. Perform iterative optimization on the path solving objective to obtain the first round of path activation coefficients, and use the first round of path activation coefficients to generate a reweighted sparse penalty term. S56. Perform iterative optimization again on the path solution objective after writing the reweighted sparse penalty term to obtain the second round of path activation coefficients. Keep the path segments with non-zero second round path activation coefficients as candidate activation paths.

10. The method for dynamic analysis and anomaly localization of automotive CAN bus load rate according to claim 9, characterized in that, The method in S6 includes: S61. Candidate activation paths that occupy the same reverse competition slot and are used to interpret the same transmission intention strength increment are grouped into the same competition group. S62. Compare the second-round path activation coefficients of each candidate activation path in the same competitive group, select the candidate activation path with the largest second-round path activation coefficient in the same competitive group, and compress the second-round path activation coefficients of the remaining candidate activation paths to zero. S63. Repeat the path activation coefficient comparison and path activation coefficient compression for all competing groups to obtain sparse activation paths; S64. Merge sparse activation paths belonging to the same message identifier in time to obtain message-level activation paths, and sum the path activation coefficients in the message-level activation paths to obtain the message-level activation quantity. S65. Select the message-level activation path with the largest message-level activation volume, and write the message identifier to which the corresponding message-level activation path belongs into the abnormal message identifier set. S66. Map the message identifiers in the abnormal message identifier set to the corresponding ECU nodes to obtain the abnormal node set, and output the abnormal location result.

Images