Method for extracting vehicle fuel cell working condition and bipolar plate welding working condition based on dtw

Abstract

This invention discloses a method for extracting operating conditions of vehicle fuel cells and bipolar plate welding conditions based on Time-Series Data Welding (DTW). It relates to the field of time-series data processing technology, including: long-term data preprocessing, multi-stage feature partitioning, weight allocation, generation of candidate representative segments, optimal segment selection based on DTW, iterative verification, and result output. The algorithm was initially applied to the time-series analysis of bipolar plate laser welding speed, extracting representative welding speed segments to accelerate thermo-mechanical coupling simulation. Based on this, the algorithm is extended to extract operating conditions of commercial vehicle fuel cells, achieving typical condition extraction and providing standard input for bipolar plate corrosion simulation. This invention significantly reduces computational costs while ensuring simulation accuracy. The algorithm is highly versatile, adaptable to both welding and automotive fuel cell scenarios, aligns with engineering practice, and has high industrial transformation value.

Description

Technical Field

[0001] This invention relates to the field of time series data processing technology, and in particular to a method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW. Background Technology

[0002] Commercial vehicle fuel cell operating condition identification and bipolar plate laser welding operating condition analysis are both typical industrial process time-series data processing scenarios. The relevant objects typically exhibit long durations, distinct stage characteristics, and are non-stationary sequences constrained by process conditions. Existing technologies often employ methods such as full-cycle data acquisition, preprocessing, stage division, statistical feature extraction, sliding window truncation, and similarity comparison to characterize operating or welding conditions, thereby generating representative operating condition segments that can be used for testing, evaluation, optimization, or process reproduction.

[0003] However, conventional methods still have two limitations for time-series processes with significant stage transitions and physical constraints, such as laser welding of fuel cells and bipolar plates in commercial vehicles: First, they rely too much on local statistical similarity and cannot fully cover key stages such as start-up, steady state, peak, or acceleration / deceleration, resulting in insufficient structural representativeness of the extracted segments; Second, they do not fully incorporate the physical characteristics of the process, such as heat input, power changes, or velocity evolution, which can easily lead to situations where segments appear similar but lack physical equivalence, thus affecting the effectiveness of subsequent testing and process analysis. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW, which solves the problems of incomplete coverage of representative segments and insufficient physical equivalence in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW (Dynamic Time Warping). The method includes: collecting target long-term time-series data; preprocessing the target long-term time-series data to form effective time-series data; extracting statistical features, stage change features, and process physical features from the effective time-series data based on the application scenario, and dividing the data according to the degree of influence of different stages on the operating condition quality, and trimming multiple stage feature segments respectively; sequentially splicing the multiple stage feature segments according to the temporal sequence and weight allocation data of each stage in the target operating condition to generate initial feature segments characterizing the core evolution process of the target operating condition; generating multiple candidate segments from the effective time-series data using a sliding window, and jointly screening each candidate segment based on the initial feature segments, combined with dynamic time warping distance, feature similarity, and the equivalence of process physical features between the candidate segments and the full-cycle effective data, to determine the optimal segment; comparing the similarity of the optimal segment with a similarity threshold, adjusting the stage trimming length or splicing parameters and iteratively performing the screening when the similarity is lower than the similarity threshold, and outputting the optimal segment when the similarity is not lower than the similarity threshold.

[0007] As a preferred embodiment of the method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW described in this invention, the formation of effective time-series data includes: removing missing values, cleaning outliers, removing invalid data segments, normalizing the target long time-series data, and reconstructing the time sequence to obtain continuous effective time-series data.

[0008] As a preferred embodiment of the method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW according to the present invention, the outlier detection rule is the 3σ criterion; the normalization process is min-max normalization; and the invalid data segment includes the velocity zeroing data during the arc resetting stage of bipolar plate laser welding.

[0009] As a preferred embodiment of the DTW-based method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions according to the present invention, the acquisition of multiple stage feature segments includes: calling the corresponding stage determination rules according to the application scenario; when the application scenario is a commercial vehicle fuel cell operation scenario, based on power amplitude, power change rate, power fluctuation amplitude, and duration, sequentially locating the start-up segment, steady-state segment, peak segment, and stop segment in the effective time series data, and trimming according to the interval position corresponding to each stage to obtain the corresponding stage feature segments; when the application scenario is a bipolar plate laser welding scenario, based on welding speed statistical characteristics, speed change slope characteristics, and unit length heat input characteristics, locating the arc ignition preheating segment, main weld welding segment, and internal flow channel welding segment in the effective time series data, and trimming according to the location data to obtain the corresponding stage feature segments.

[0010] As a preferred embodiment of the DTW-based method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions according to the present invention, the initial feature fragment includes: sequentially splicing multiple stage feature fragments according to the time sequence of each stage in the actual operating conditions; determining the fragment length ratio of each stage feature fragment according to the degree of influence of each stage on the quality or state characterization of the target operating conditions; and sequentially splicing the stage feature fragments after determining the length ratio to generate an initial feature fragment that completely covers the core stage features of the target scenario.

[0011] As a preferred embodiment of the DTW-based method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions according to the present invention, the step of generating multiple candidate segments from valid time-series data using a sliding window includes: setting the length of the sliding window, performing sliding truncation on the valid time-series data to generate multiple candidate segments; and retaining candidate segments that can completely cover the characteristics of multiple stages as valid candidate segments for screening.

[0012] As a preferred embodiment of the DTW-based method for extracting vehicle fuel cell operating conditions and bipolar plate welding operating conditions according to the present invention, the determination of the optimal segment includes: constructing a dynamic time warping distance matrix between the initial feature segment and each candidate segment, and calculating the cumulative distance in the distance matrix according to the recursive relationship; taking the cumulative distance corresponding to the end position of the distance matrix as the dynamic time warping distance of each candidate segment; calculating the normalized similarity of each candidate segment according to the dynamic time warping distance, sequence length, and value range of the candidate segment; when the application scenario is a commercial vehicle fuel cell operation scenario, the candidate segment with the smallest dynamic time warping distance, the highest similarity, and complete coverage of the start-up segment, steady-state segment, peak segment, and stop segment is selected as the optimal segment; when the application scenario is a bipolar plate laser welding scenario, the candidate segment with the smallest dynamic time warping distance, the highest similarity, and the smallest thermal input deviation from the full-cycle effective data is selected as the optimal segment.

[0013] As a preferred embodiment of the DTW-based method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions according to the present invention, the similarity includes: obtaining the cumulative distance between the candidate segment and the initial feature segment at the end position of the dynamic time warp distance matrix; and normalizing the cumulative distance based on the length of the initial feature segment, the length of the candidate segment, and the power value range or welding speed value range of the candidate segment to obtain the similarity between the candidate segment and the initial feature segment.

[0014] As a preferred embodiment of the DTW-based method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions according to the present invention, the step of comparing the similarity of the optimal segment with a similarity threshold, and adjusting the stage trimming length or splicing parameters and iteratively performing the filtering when the similarity is lower than the similarity threshold, specifically involves: comparing the similarity of the optimal segment with a similarity threshold; when the similarity is lower than the similarity threshold, adjusting at least one splicing parameter and re-performing stage feature segment trimming, initial feature segment splicing, and candidate segment filtering; when the similarity is not lower than the similarity threshold, outputting the optimal segment as the target representative operating condition segment; the splicing parameters include stage trimming length, stage splicing ratio, sliding window length, and sliding step size.

[0015] As a preferred embodiment of the DTW-based method for extracting vehicle fuel cell operating conditions and bipolar plate welding operating conditions according to the present invention, the optimal segment output includes: when the application scenario is a bipolar plate laser welding scenario, the optimal segment is checked for consistency based on the relative deviation of heat input, and the optimal segment is output when the relative deviation is not greater than the deviation threshold; when the application scenario is a commercial vehicle fuel cell operation scenario, the optimal segment is checked for integrity based on the stage feature coverage, and the optimal segment is output when the optimal segment includes at least the start-up segment, steady-state segment, peak segment, and stop segment.

[0016] The beneficial effects of this invention are as follows: by locating, dividing, and pruning long-term operating conditions based on stage features, the completeness of stage coverage and structural representativeness of the extracted representative segments are improved; by splicing multiple stage feature segments according to the actual operating condition time sequence, the core evolutionary features of the entire cycle are preserved in shorter segments, and the effectiveness of operating condition reproduction is improved; by combining dynamic time warping distance, feature similarity, and physical equivalence to jointly screen candidate segments, both structural similarity and process equivalence are taken into account, and the accuracy of optimal segment extraction is improved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for extracting the operating conditions and bipolar plate welding conditions of a vehicle fuel cell based on DTW.

[0019] Figure 2 This is a comparison chart of spliced ​​working condition segments and optimal working condition segments.

[0020] Figure 3This is a spliced ​​test condition diagram.

[0021] Figure 4 This is a comparison chart of splicing welding speed and the optimal matching speed segment.

[0022] Figure 5 This is a diagram showing the welding speed of the spliced ​​parts. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0026] Reference Figures 1-5 This is one embodiment of the present invention, which provides a method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW, including the following steps: In this embodiment, taking the fuel cell operating condition of a commercial vehicle as an example, specifically: S1. Collect target long-term time-series data, preprocess the target long-term time-series data, and form effective time-series data.

[0027] Read the baseline operating condition data file, extract the time and power columns, and obtain the target long-term time series data.

[0028] It should be noted that, in this embodiment, the target long-time series data is the long-term operating power data of a commercial vehicle fuel cell.

[0029] Missing values ​​are cleaned up for the target long-term time series data by deleting missing values, while retaining valid data points.

[0030] Sort the target long-time series data in ascending order of time, and calculate the power difference using the following expression: ; in, Indicates the first The power change rate corresponding to each data point Indicates the first Power values ​​for each data point Indicates the first Power values ​​for each data point Indicates the data point number.

[0031] The power change rate is the ratio of the power difference to the step size of adjacent sampling points (e.g., 10 seconds).

[0032] The power change rate sequence is processed to achieve a uniform length, so that the power change rate sequence and the target long time series data can be used together as effective time series data.

[0033] For example, in this embodiment, an exemplary parameter is provided: the time column is 0~360000 seconds, the step size is 10 seconds, and the power column has a total of 36000 data points. After deleting missing values, 35903 valid data points are retained. After deleting missing values ​​and calculating the power change rate, 35902 power change rate data points are obtained. A zero is added to the end of the power change rate sequence to extend the length of the power change rate sequence to 35903 data points. The power change rate sequence and the target long time series data are used as valid time series data.

[0034] It should also be noted that, in an optional embodiment, when the application scenario is bipolar laser welding, the specific steps are as follows: Collect welding speed timing data, remove invalid data when the speed returns to zero during the arc termination and reset phase, and retain valid welding speed timing data.

[0035] The 3σ criterion is used to remove outliers such as power grid interference during the data acquisition process.

[0036] For missing data resulting from outlier removal, linear interpolation is performed to fill in the missing data based on welding speed data at adjacent times before and after the missing data.

[0037] When missing data is located at the beginning or end of the sequence, the welding speed data of the nearest valid moment is used to complete it.

[0038] The welding speed time series data is normalized by min-max to eliminate the influence of dimensions and obtain normalized welding speed time series data.

[0039] For example, in this embodiment, an exemplary parameter is provided to remove invalid data of zero speed during the arc-end reset segment at 246~355 seconds / time in 11 welding operations, and retain the valid welding speed timing data of 0~245 seconds / time, for a total of 2695 sets.

[0040] S2. Based on the application scenario, extract statistical features, stage change features, and process physical features from the effective time series data, and locate and divide them according to the degree of impact of different stages on the quality of the working condition, and cut and obtain multiple stage feature fragments respectively.

[0041] Furthermore, the effective time-series data is used to locate and trim the starting segment.

[0042] The criteria for determining the start-up segment location and trimming are that the power is less than the power threshold and the power change rate is greater than the change rate threshold.

[0043] It should be noted that the power threshold is obtained by statistically analyzing the power values ​​corresponding to the starting position of the starting section in historical benchmark operating conditions and taking their arithmetic mean, with a value range of [5kW, 15kW]; the rate of change threshold is obtained by statistically analyzing the power change rate data of the starting section in historical benchmark operating conditions and taking the arithmetic mean of the part greater than zero, with a value range of [0.10kW / s, 0.20kW / s].

[0044] By traversing the valid time series data, the first starting point that meets the conditions is located, and the data points adjacent to the first starting point that meets the conditions are pruned within the set time window.

[0045] The power of the starting segment is calculated based on the data points obtained from the cropping, and the average power change rate is obtained.

[0046] The average power change rate is effectively determined. If the average power change rate meets the determination criteria, the starting segment positioning and trimming are completed.

[0047] For example, in this embodiment, an exemplary parameter is provided: the determination condition for starting segment positioning and trimming is power <10kW and power change rate >0.15kW / S; by traversing the valid timing data, the first starting point that meets the condition is located as the 10th data point of the baseline operating condition, corresponding to a time of 90 seconds; the trimming range is 50 points before and after the 10th data point, that is, the 5th to 55th data points, corresponding to a time of 50~550 seconds, with a duration of 500 seconds; through calculation, the starting segment power increases from 0.8kW to 82.3kW, with an average power change rate of 0.163kW / S, which meets the determination condition.

[0048] Furthermore, the steady-state segment is located and pruned from the effective time-series data.

[0049] The criteria for determining the steady-state segment location and trimming are: the power is within the steady-state power range, the absolute value of the power change rate is less than the fluctuation threshold, and the continuous duration is not less than the continuous duration.

[0050] It should be noted that the steady-state power range is obtained by statistically analyzing the minimum and maximum power values ​​during the steady-state operation phase in historical benchmark operating conditions, with a value range of [50kW, 90kW]; the fluctuation threshold is obtained by statistically analyzing the absolute values ​​of the power change rates of adjacent sampling points during the steady-state operation phase in historical benchmark operating conditions and taking the maximum value, with a value range of [1kW, 5kW]; the continuous duration is obtained by statistically analyzing the duration of data held within the power range during the steady-state operation phase in historical benchmark operating conditions and taking the minimum value, with a value range of [100s, 400s].

[0051] After the start-up phase ends, locate the interval that meets the conditions.

[0052] After locating the interval that meets the conditions, the interval is clipped, and the average power, maximum power, minimum power, and power fluctuation value are calculated based on the clipped data points to effectively determine the steady-state segment.

[0053] For example, in this embodiment, an exemplary parameter is provided. The determination conditions for steady-state segment positioning and trimming are: power 60~80kW, absolute value of power change rate <3kW, and continuous duration ≥200 seconds. The interval that meets the conditions is located after the start-up segment ends, i.e., after the 55th data point. The trimming range is the 55th to 155th data points, corresponding to a time of 550~1550 seconds and a duration of 1000 seconds. Through calculation, the average power of the steady-state segment is 72.6kW, the maximum power is 78.9kW, the minimum power is 61.2kW, and the maximum power fluctuation is 2.7kW, which meets the determination conditions.

[0054] Furthermore, peak segments are located and cropped from the effective time-series data.

[0055] Among them, the criterion for determining peak segment location and clipping is that the power is greater than the peak power threshold.

[0056] It should be noted that the peak power threshold is obtained by statistically analyzing the minimum power during the peak output phase in historical benchmark operating conditions and using this minimum value as the lower limit for judgment, with a value range of [80kW, 100kW].

[0057] After the steady-state phase ends, locate the interval that meets the conditions.

[0058] After locating the interval that meets the conditions, the interval is clipped, and the average power, maximum power, and minimum power are calculated based on the clipped data points to effectively determine the peak segment.

[0059] For example, in this embodiment, an exemplary parameter is provided: the determination condition for peak segment positioning and clipping is power > 90kW, and the interval that meets the condition is located after the steady-state segment ends, i.e., after the 155th data point; the clipping range is the 156th to 186th data points, corresponding to a time of 1560 to 1860 seconds, with a duration of 300 seconds; through calculation, the average power of the peak segment is 96.8kW, the maximum power is 102.5kW, and the minimum power is 91.3kW, which meets the determination condition.

[0060] Furthermore, stop segments are located and trimmed from the valid time-series data.

[0061] The criteria for determining the stop segment location and trimming are that the power change rate meets the stop change rate condition and the power is less than the stop power threshold.

[0062] It should be noted that the stopping rate of change condition is obtained by statistically analyzing the power change rate data of adjacent sampling points in the stopping section of the historical benchmark operating condition and taking their arithmetic mean, with a value range of [-15kW / s, -5kW / s]; the stopping power threshold is obtained by statistically analyzing the power value corresponding to the end position of the stopping section in the historical benchmark operating condition and taking its maximum value, with a value range of [10kW, 20kW].

[0063] After the peak segment ends, locate the interval that meets the conditions.

[0064] After locating the interval that meets the conditions, the interval is clipped, and the power of the stopping segment and the average power change rate are calculated based on the data points obtained from the clipping, so as to effectively determine the stopping segment.

[0065] For example, in this embodiment, an exemplary parameter is provided: the determination condition for stopping segment location and clipping is that the power change rate is < -0.4kW / S and the power is < 15kW, and the interval that meets the condition is located after the peak segment ends, i.e., after the 186th data point; the clipping range is the 186th to 206th data points, corresponding to a time of 1860 to 2060 seconds, with a duration of 200 seconds; through calculation, the power of the stopping segment decreases from 100.2kW to 4.7kW, and the average power change rate is -0.4475kW / S, which meets the determination condition.

[0066] It should also be noted that, in an optional embodiment, when the application scenario is bipolar laser welding, the specific steps are as follows: Based on the physical logic of welding metallurgy, core features strongly correlated with weld quality are extracted from normalized welding speed time series data, and priority is assigned according to process weight.

[0067] Among them, core features that are strongly related to weld quality are extracted, and priority is assigned according to process weight to ensure that the extracted segments can reflect the core quality influencing factors of the entire welding cycle.

[0068] The extracted features include statistical features such as the mean, variance, peak / valley distribution, and fluctuation period of the welding speed throughout the entire cycle, as well as process features such as the second-level fluctuation characteristics of the uniform speed section of the main weld, the slope characteristics of acceleration and deceleration at corners, and the speed rise law of the arc initiation section.

[0069] The expression for the physical characteristic of heat input per unit length of weld seam is: ; in, This represents the heat input per unit length of weld. Indicates thermal efficiency. Indicates laser power. This indicates the real-time welding speed.

[0070] Furthermore, after feature extraction, weights are assigned.

[0071] Specifically, based on the degree of influence of different process stages on weld quality, different weights are assigned to the characteristics of the uniform speed section of the main weld, the acceleration and deceleration characteristics of the internal flow channel, and the characteristics of the arc initiation section.

[0072] Among them, the uniform speed section of the main weld has the greatest impact on the overall weld formation consistency and penetration stability, the acceleration and deceleration stage of the internal flow channel has the second greatest impact on local heat input fluctuations, and the arc initiation section mainly affects the establishment of the initial molten pool with a relatively small range of influence.

[0073] For example, in this embodiment, the characteristic weight of the uniform speed section of the main weld is 60%, the characteristic weight of the acceleration and deceleration of the internal flow channel is 25%, and the characteristic weight of the arc initiation section is 15%.

[0074] Furthermore, after completing feature extraction and weight allocation, feature segments from each core stage are cropped.

[0075] Specifically, the characteristic segments of the arc initiation and preheating section are cut out, and the segments with the most typical characteristics of the arc initiation section speed increase and the continuous change of heat input per unit length of weld are selected for general cutting.

[0076] Cut the characteristic segments of the main weld seam welding section, and select the segments with the most typical characteristics of uniform fluctuation and corner acceleration and deceleration, and whose heat input per unit length of weld seam remains stable for general cutting.

[0077] The characteristic segments of the internal flow channel welding section are cut out, and the segments with the most typical acceleration and deceleration characteristics of the internal flow channel and whose changes in heat input per unit length of weld can characterize the local thermal adjustment process are selected for general cutting.

[0078] For example, in this embodiment, an exemplary parameter is provided: the most typical segment of the arc-starting speed rise characteristic in 11 welding operations is selected and the segment is collected and cut to obtain a 45-second characteristic segment; the most typical segment of the main weld seam uniform speed fluctuation and corner acceleration and deceleration characteristics in 11 welding operations is selected and the segment is collected and cut to obtain a 210-second characteristic segment; the most typical segment of the internal flow channel acceleration and deceleration characteristics in 11 welding operations is selected and the segment is collected and cut to obtain a 45-second characteristic segment.

[0079] like Figure 5 As shown, Figure 5 The curve in the figure represents a 300-second target welding speed segment that can represent the full data, cropped and spliced ​​from the exemplary 2695 sets of data.

[0080] S3. Based on the sequence and weight allocation data of each stage in the target working condition, the feature segments of multiple stages are sequentially spliced ​​to generate the initial feature segments that characterize the core evolution process of the target working condition.

[0081] Furthermore, the start-up segment, steady-state segment, peak segment, and stop segment are spliced ​​together according to the actual operating conditions and timing sequence to form the target test condition.

[0082] After splicing, the spliced ​​working condition time series is normalized so that the spliced ​​target test working condition can represent the complete stage evolution process on a unified time scale, and ensure the integrity of the time sequence and the continuity of the data in each stage.

[0083] For example, in this embodiment, an exemplary parameter is provided, with the splicing order being 50 points for the start-up segment, 100 points for the steady-state segment, 30 points for the peak segment, and 20 points for the stop segment, resulting in a total of 200 data points after splicing; the spliced ​​operating condition time series is normalized to 0~2000 seconds with a step size of 10 seconds, corresponding to a 2000-second test condition; after splicing, the target test condition power range is 0.8~102.5kW, the four-stage timing is complete, there is no missing data, and it can be directly used for bench testing.

[0084] It should also be noted that, in an optional embodiment, when the application scenario is bipolar laser welding, the specific steps are as follows: The initial feature segments are formed by sequentially splicing together the characteristic segments of the arc-starting and preheating section, the main weld section, and the internal flow channel weld section.

[0085] After splicing is completed, the initial feature segments are made to fully cover all process features in the welding process, and the duration of the initial feature segments and the effective data of the whole cycle are proportional.

[0086] It should be noted that the ratio is obtained by statistically analyzing the ratio of the duration of historical qualified operating condition segments to the duration of corresponding full-cycle valid data, and the minimum and maximum values ​​among all qualified ratios are taken as the ratio boundaries.

[0087] For example, in this embodiment, an exemplary parameter is provided, which splices the characteristic segments of the arc preheating section, the main weld section, and the internal flow channel weld section in a time sequence to form an initial characteristic segment of 300 seconds. The ratio of this initial characteristic segment to the effective data duration of the entire cycle is approximately 1 / 9, ensuring that the segment completely covers all process features.

[0088] like Figure 3 As shown, Figure 3 The curve in the figure represents a target test condition of 2000 seconds that can represent the full data, which is cropped and spliced ​​from the exemplary 360,000 seconds of data.

[0089] S4. Multiple candidate segments are generated using a sliding window for the effective time series data. Based on the initial feature segments, the candidate segments are jointly screened by combining the dynamic time warping distance, feature similarity, and the equivalence of process physical characteristics between the candidate segments and the full-cycle effective data to determine the optimal segment.

[0090] Further, candidate fragment extraction is performed.

[0091] Specifically, the baseline operating conditions are traversed, and candidate segments are extracted from the valid time series data at fixed steps (e.g., 30 seconds).

[0092] Using the spliced ​​target test case as a baseline, the dynamic time warping distance between the target test case and each candidate segment is calculated, and the expression is as follows: ; in, Indicates the target test condition sequence number The point and the candidate fragment sequence The cumulative DTW distance corresponding to each point Indicates the target test condition sequence number The value of each data point. Indicates the candidate fragment sequence number 1 The value of each data point. This represents the cumulative distance from the previous point in the target test condition sequence to the current position. This represents the cumulative distance from the previous point in the candidate segment sequence to the current position. This represents the cumulative distance traveled from the previous point of each of the two sequences to the current position.

[0093] Furthermore, the candidate segment corresponding to the minimum dynamic time warping distance is selected as the best matching segment, and the similarity is calculated based on the maximum power value, minimum power value, and dynamic time warping distance of the end position of the best matching segment. The expression is: ; ; in, This represents the similarity between the target test case and the best matching candidate segment. This represents the cumulative DTW distance between the target test case and the best-matching candidate segment at the end position. This represents the total number of data points for the target test condition. This represents the total number of data points in the best-matching candidate segment sequence. This represents the maximum power value among the best-matching candidate segments. This represents the minimum power value among the best-matching candidate segments.

[0094] For example, in this embodiment, an exemplary parameter is provided: traversing a 360,000-second baseline condition, candidate segments are extracted at a step size of 30, resulting in 1130 valid candidate segments; the calculated minimum dynamic time warping distance is... The candidate segment corresponding to the minimum dynamic time warp distance was selected as the best matching segment. The maximum power of the best matching segment was 101.8kW and the minimum power was 0.9kW. The similarity was calculated as follows: .

[0095] like Figure 2 As shown, Figure 2 The blue curve represents the target test condition that can be represented by 2000 seconds of data that is cropped and spliced ​​from the exemplary 360,000 seconds of data. The red curve represents the best test condition segment of 223 seconds obtained from the exemplary 360,000 seconds of data using the DTW algorithm, which has a similarity of 0.9446 to the target test condition.

[0096] It should also be noted that, in an optional embodiment, when the application scenario is bipolar laser welding, the specific steps are as follows: Furthermore, a sliding window method is used to traverse the preprocessed effective welding speed data, sliding with a fixed window length (e.g., 300 seconds) and a fixed step size (e.g., 10 seconds) to generate multiple candidate representative segments.

[0097] Among them, each group of candidate segments must simultaneously meet the full process characteristics of complete coverage of arc preheating, main weld welding and internal flow channel welding, the similarity between the velocity statistical characteristics within the segment and the effective data of the whole cycle is greater than or equal to the similarity threshold, and the relative deviation between the total heat input per unit length within the segment and the effective data of the whole cycle is less than or equal to the deviation threshold.

[0098] It should be noted that the deviation threshold is obtained by statistically analyzing the relative deviation of the total heat input per unit length of historical qualified working condition segments. The maximum permissible deviation that can ensure the equivalence of segment heat input and does not cause false screening of representative segments is taken as the deviation threshold, with a value range of 3% to 8%.

[0099] After generating candidate representative segments, the objective function is to maximize the equivalence of hot input and the highest feature similarity. The similarity is calculated by combining the dynamic time warping algorithm, and the optimal representative segment is selected.

[0100] Furthermore, initialize the dynamic time-warped distance matrix. .

[0101] in, , ( ), ( ).

[0102] The DTW distance is calculated recursively, and the bottom right element of the distance matrix is ​​taken as the final dynamic time-warped distance.

[0103] Similarity is calculated based on the final dynamic time warp distance, sequence length, and candidate segment velocity range.

[0104] After obtaining the similarity score, the optimal representative segment is determined by combining it with the equivalence screening.

[0105] For example, in this embodiment, an exemplary parameter is provided: the sliding window method is used to traverse the preprocessed 2695 seconds of effective welding speed data, with a window length of 300 seconds and a step size of 10 seconds, to generate 240 sets of candidate representative segments; the lower right corner element of the distance matrix is ​​taken. As the final dynamic time warping distance, the minimum dynamic time warping distance in this embodiment is: The calculated similarity is: Combined with heat input equivalence screening, the optimal representative segment was finally determined to correspond to the original data time range of 910 to 1261 seconds, a total of 351 seconds. The optimal representative segment covers the core process characteristics of 11 welding operations. The average velocity within the segment is 9.92 mm / s, with a relative deviation of 1.3% from the average value of the whole cycle. The total heat input has a relative deviation of 2.8% from the average value of the whole cycle, and is completely equivalent to the core physical characteristics of the whole cycle working condition.

[0106] As shown in the figure Figure 4 The blue curve represents a 300-second target welding speed segment that can represent the full data, which is cut and spliced ​​from the exemplary 2695 sets of data. The red curve represents a 351-second optimal welding speed segment obtained from the exemplary 2695 sets of data using the DTW algorithm, which has a similarity of 0.9620 to the target welding speed condition.

[0107] S5. Compare the similarity of the best segment with the similarity threshold. If the similarity is lower than the similarity threshold, adjust the stage trimming length or splicing parameters and iteratively perform the filtering. If the similarity is not lower than the similarity threshold, output the best segment.

[0108] Furthermore, the optimal segment similarity is compared with the similarity threshold.

[0109] When the similarity of the best segment is greater than or equal to the similarity threshold, the current best segment is determined to meet the consistency requirement, and no iteration is required.

[0110] When the similarity of the optimal segment is less than the similarity threshold, the trimming length or splicing parameters at each stage are adjusted, and the segments are re-trimmed, spliced, and candidate segments are generated until the optimal segment with a similarity not lower than the similarity threshold is obtained.

[0111] The splicing parameters include stage trimming length, stage splicing ratio, sliding window length, and sliding step size.

[0112] It should be noted that the stage splicing ratio refers to the length ratio of each stage feature segment in the initial feature segment.

[0113] It should be noted that the stage clipping length refers to the length of the continuous data interval intercepted along the time axis from the start position to the end position of the stage after the corresponding stage positioning is completed.

[0114] It should be noted that the similarity threshold is obtained by statistically analyzing the DTW similarity of historically verified qualified operating condition samples. The difference between the mean similarity of all qualified samples and one standard deviation is taken as the similarity threshold. The similarity threshold range is [0.80, 0.95]. In this embodiment, 0.9 is preferred. This method of taking a value can simultaneously reflect the concentration level and fluctuation range of qualified samples, ensuring that most normal operating condition matching results are acceptable while excluding abnormal matching results with significantly low similarity.

[0115] In summary, this invention improves the completeness and structural representativeness of the extracted representative segments by locating, dividing, and pruning long-term operating conditions based on stage features; it preserves the core evolutionary features of the entire cycle in shorter segments and improves the effectiveness of operating condition reproduction by splicing multiple stage feature segments according to the actual operating condition time sequence; and it achieves simultaneous consideration of structural similarity and process equivalence and improves the accuracy of optimal segment extraction by combining dynamic time warping distance, feature similarity, and physical equivalence for joint screening of candidate segments.

[0116] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW, characterized in that, include: Collect long-term time-series data of the target, preprocess the target long-term time-series data to form effective time-series data; Based on the application scenario, statistical features, stage change features, and process physical features are extracted from effective time series data. The data are then located and divided according to the degree of impact of different stages on the quality of the working condition, and multiple stage feature fragments are obtained by trimming them. Based on the sequential timing and weight allocation data of each stage in the target working condition, the feature segments of multiple stages are sequentially spliced ​​to generate the initial feature segments that characterize the core evolution process of the target working condition. Multiple candidate segments are generated using a sliding window for the effective time series data. Based on the initial feature segments, the optimal segment is determined by combining the dynamic time warping distance, feature similarity, and the equivalence of process physical features between the candidate segments and the full-cycle effective data. The similarity of the optimal segment is compared with the similarity threshold. When the similarity is lower than the similarity threshold, the stage trimming length or splicing parameters are adjusted and the filtering is performed iteratively. When the similarity is not lower than the similarity threshold, the optimal segment is output.

2. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 1, characterized in that, The formation of valid time-series data includes: The target long-term time series data is processed by removing missing values, cleaning outliers, removing invalid data segments, normalizing, and reconstructing the time order to obtain continuous and valid time series data.

3. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 2, characterized in that, The outlier detection rule is the 3σ criterion; The normalization process is min-max normalization; The invalid data segment includes the speed zeroing data during the arc-reset phase of bipolar laser welding.

4. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 1 or 2, characterized in that, The acquisition of multiple stage feature fragments includes: The corresponding stage determination rules are invoked based on the application scenario; When the application scenario is the operation of fuel cells in commercial vehicles, based on the power amplitude, power change rate, power fluctuation amplitude and duration, the start-up segment, steady-state segment, peak segment and stop segment in the effective time series data are sequentially located, and the corresponding segment is cropped according to the interval position of each stage to obtain the corresponding stage feature segment. When the application scenario is bipolar laser welding, based on the statistical characteristics of welding speed, the slope characteristics of speed change, and the heat input characteristics per unit length, the arc ignition and preheating segment, the main weld welding segment, and the internal flow channel welding segment in the effective time sequence data are located, and the corresponding stage feature segments are obtained by trimming according to the location data.

5. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 4, characterized in that, The initial feature fragment includes: The feature segments of multiple stages are sequentially spliced ​​together according to the time sequence of each stage in the actual working conditions. The proportion of segment length of characteristic segments in each stage is determined based on the degree of influence of each stage on the quality or state characterization of the target working condition. After determining the length proportions, the feature fragments of each stage are spliced ​​together in sequence to generate an initial feature fragment that fully covers the core stage features of the target scene.

6. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 1, characterized in that, The process of generating multiple candidate segments from valid time-series data using a sliding window includes: Set the sliding window length, slide and truncate the valid time series data to generate multiple candidate segments; Candidate segments that can fully cover the characteristics of multiple stages are retained as valid candidate segments for screening.

7. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 6, characterized in that, The determination of the optimal segment includes: Construct a dynamic time-warped distance matrix between the initial feature fragment and each candidate fragment, and calculate the cumulative distance in the distance matrix according to the recursive relationship; The cumulative distance corresponding to the end position of the distance matrix is ​​taken as the dynamic time-normalized distance of each candidate segment; The normalized similarity of each candidate segment is calculated based on the dynamic time warp distance, sequence length, and the range of candidate segment values. When the application scenario is the operation of fuel cells in commercial vehicles, the candidate segment with the smallest dynamic time warp distance, the highest similarity, and complete coverage of the start-up segment, steady-state segment, peak segment, and stop segment is selected as the optimal segment. When the application scenario is bipolar laser welding, the candidate segment with the smallest dynamic time warp distance, the highest similarity, and the smallest thermal input deviation with the full-cycle effective data is selected as the optimal segment.

8. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 7, characterized in that, The similarity includes: Obtain the cumulative distance between the candidate fragment and the initial feature fragment at the end position of the dynamic time-warped distance matrix; Based on the initial feature fragment length, candidate fragment length, and the power value range or welding speed value range of the candidate fragment, the cumulative distance is normalized to obtain the similarity between the candidate fragment and the initial feature fragment.

9. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 7, characterized in that, The process of comparing the similarity of the optimal segment with a similarity threshold, adjusting the trimming length or splicing parameters when the similarity is lower than the threshold, and iteratively performing the filtering is as follows: The similarity of the best segment is compared with a similarity threshold; When the similarity is lower than the similarity threshold, at least one splicing parameter is adjusted and the stage feature fragment trimming, initial feature fragment splicing, and candidate fragment screening are re-executed. When the similarity is not lower than the similarity threshold, the optimal segment is output as the representative segment of the target working condition. The splicing parameters include stage trimming length, stage splicing ratio, sliding window length, and sliding step size.

10. The method for extracting vehicle fuel cell operating conditions and bipolar plate welding conditions based on DTW as described in claim 1, characterized in that, The optimal output segment includes: When the application scenario is bipolar laser welding, the optimal segment is checked for consistency based on the relative deviation of the heat input, and the optimal segment is output when the relative deviation is not greater than the deviation threshold. When the application scenario is a commercial vehicle fuel cell operation scenario, the integrity of the optimal segment is checked based on the stage feature coverage, and the optimal segment is output when the optimal segment contains at least the start-up segment, steady-state segment, peak segment and stop segment.

Images