Unmanned mine card abnormal data detection method and system

Abstract

The invention relates to an unmanned mine card abnormal data detection method and system. The method comprises the following steps: collecting real-time data of an unmanned mine card; the method comprises the following steps: collecting a plurality of streaming anomaly detection algorithms, performing a simulation test on an unmanned mine card scene, quantifying delay, accuracy, concept drift adaptability and calculation complexity corresponding to each anomaly detection algorithm, and generating a quantitative index corresponding to each anomaly detection algorithm; the quantitative indexes comprise a delay index, an accuracy index, a concept drift adaptability index and a calculation complexity index; calculating a comprehensive score corresponding to each anomaly detection algorithm based on the quantitative index and a preset index weight, and sorting the anomaly detection algorithms based on the comprehensive scores; on the basis that the comprehensive score sum is a preset constraint condition of the unmanned mine card, one or more combinations are screened out from all the anomaly detection algorithms to serve as a final target algorithm; and inputting the preprocessed real-time data into a target algorithm to obtain an abnormal result, and generating abnormal level information based on the abnormal result.

Description

Technical Field

[0001] This disclosure relates to the field of unmanned mining truck monitoring, specifically to a method and system for detecting abnormal data from unmanned mining trucks. Background Technology

[0002] With the acceleration of intelligent mining, unmanned mining trucks have become core equipment for efficient operation in open-pit mines, and their safe and stable operation depends on the accurate monitoring of massive amounts of real-time data. However, existing streaming anomaly detection technologies generally have the following shortcomings in unmanned mining truck scenarios: such as single evaluation indicators, ignoring the mine's special needs for real-time performance, resource constraints, and data drift, algorithm selection relying on empirical judgment, and the lack of a dynamic matching mechanism with actual constraints.

[0003] For example, the prior art disclosed in CN115758266A provides a method for detecting abnormal data in unmanned mining trucks. This method includes: collecting real-time streaming data generated during the interaction of the positioning module, local path planning module, and control module in the robot operating system of the unmanned mining truck based on a topic subscription mechanism; establishing a data structure tree based on a random forest to generate an unsupervised data stream anomaly detection model; calculating the anomaly score value of each sample point in the data structure tree of the real-time streaming data based on a reservoir sampling strategy; and comparing the anomaly score value with a preset threshold to determine abnormal sample points, thereby achieving anomaly detection in the real-time streaming data. However, this prior art also focuses on a relatively singular evaluation indicator, failing to consider the impact and needs of mines regarding real-time performance, resource constraints, and data drift, and does not dynamically match the actual constraints of the mine, making it difficult to effectively monitor anomaly detection in massive amounts of real-time data. Summary of the Invention

[0004] To address the problems existing in the prior art, this disclosure proposes a method and system for detecting abnormal data from unmanned mining trucks, thereby solving at least one of the aforementioned technical problems. The technical solution adopted in this disclosure is as follows: In a first aspect, this disclosure provides a method for detecting abnormal data from unmanned mining trucks, the method comprising: Collect multi-dimensional real-time data from unmanned mining trucks; Several streaming anomaly detection algorithms are collected, and simulation tests are conducted on unmanned mining truck scenarios based on each anomaly detection algorithm. The latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm are quantified, and quantitative indicators corresponding to each anomaly detection algorithm are generated. The quantitative indicators include latency indicators, accuracy indicators, concept drift adaptability indicators, and computational complexity indicators. The comprehensive score corresponding to each anomaly detection algorithm is calculated based on the quantitative indicators and the preset indicator weights, and the anomaly detection algorithms are ranked based on the comprehensive scores. Based on the comprehensive score and the preset constraints of the unmanned mining truck, one or more combinations of anomaly detection algorithms are selected as the final target algorithm. The preprocessed real-time data is input into the target algorithm to obtain abnormal results, and abnormality level information is generated based on the abnormal results.

[0005] Preferably, the data sources for the real-time data include multiple types of sensors and control systems.

[0006] Preferably, the real-time data is collected, transmitted, and processed in the form of a data stream. For ease of collection, processing, and handling, the real-time data can be streaming data.

[0007] Preferably, the process of aggregating several anomaly detection algorithms, conducting simulation tests on unmanned mining truck scenarios based on each anomaly detection algorithm, quantifying the latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm, and generating quantitative indicators for each anomaly detection algorithm includes the following steps: N anomaly detection algorithms are collected and numbered sequentially, with the number of the anomaly detection algorithm denoted as i; where 1 <= i <= N, N >= 3; The latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm i are quantitatively scored to obtain a quantitative index for each anomaly detection algorithm i; wherein the quantitative index includes a latency index, an accuracy index, a concept drift adaptability index, and a computational complexity index.

[0008] Preferably, the step of calculating the comprehensive score corresponding to each anomaly detection algorithm based on the quantitative index and the preset index weights, and ranking the anomaly detection algorithms based on the comprehensive scores, includes the following steps: Based on the quantitative indicators and preset indicator weights, the comprehensive score corresponding to each anomaly detection algorithm is calculated, as follows: Overall score = Latency index × a1 + Accuracy index × a2 + Concept drift adaptability index × a3 + Computational complexity index × a4; Where a1, a2, a3 and a4 are preset indicator weights, which respectively correspond to the latency, accuracy, concept drift adaptability and computational complexity of the anomaly detection algorithm; The latency metric, accuracy metric, concept drift adaptability metric, and computational complexity metric are the quantitative metrics corresponding to the anomaly detection algorithm. Sort all the anomaly detection algorithms in descending order of their corresponding comprehensive scores.

[0009] Preferably, the step of selecting one or more anomaly detection algorithms as target algorithms based on the comprehensive score and the preset constraints of the unmanned mining truck includes the following steps: S401. The preset constraints are defined as a set of actual constraints pre-constructed based on the actual business needs, data characteristics, and hardware resources of the unmanned mining truck. The set of actual constraints is {actual business needs constraints, data characteristics constraints, and hardware resource constraints}. S402. Analyze and derive the data constraints of the real-time data. The range of values ​​for the data constraints is subject to the constraints of the actual constraint set—that is, the data constraints correspond to one or more of the constraints of {actual business requirement constraints, data characteristic constraints, and hardware resource constraints}. S403. If the data constraint does not correspond to any of the constraints in the actual constraint set, it indicates that the analysis is incorrect or has failed, and the analysis should be repeated. Therefore, return to S402. S404. If the data constraint corresponds only to one type of constraint in the actual constraint set, then the constraint corresponding to the data constraint is used as the filtering condition to select the anomaly detection algorithm with the highest comprehensive score from each anomaly detection algorithm, which is then used as the final target algorithm. S405. If the data constraint corresponds to only two constraints in the actual constraint set, then the two constraints corresponding to the data constraint are used together as the first screening condition. Based on the first screening condition, the anomaly detection algorithm with the largest comprehensive score is selected from each anomaly detection algorithm and used as the first target algorithm. The other constraint that does not correspond to the data constraint is used as the second screening condition. Based on the second screening condition, the anomaly detection algorithm with the largest comprehensive score is selected from each anomaly detection algorithm and used as the second target algorithm. The first objective algorithm and the second objective algorithm are combined to form the final objective algorithm; S406. If the data constraint corresponds to all constraints in the actual constraint set, then all constraints in the actual constraint set are used as screening conditions to select the anomaly detection algorithm with the highest comprehensive score from each anomaly detection algorithm, which is then used as the final target algorithm.

[0010] Preferably, if the anomaly detection algorithm with the highest comprehensive score is selected from the various anomaly detection algorithms, and several parallel anomaly detection algorithms are selected, then the parallel anomaly detection algorithms are combined into a new anomaly detection algorithm.

[0011] Preferably, before inputting the preprocessed real-time data into the target algorithm, the method further includes: preprocessing the real-time data.

[0012] Preferably, the step of inputting the preprocessed real-time data into the target algorithm to obtain anomaly results and generating anomaly level information based on the anomaly results includes the following steps: The preprocessed real-time data is input into the target algorithm to obtain abnormal results; The impact range, severity, urgency, and historical frequency of the abnormal results are assigned scores to obtain the assigned indicators for the abnormal results; the assigned indicators include impact range indicators, severity indicators, urgency indicators, and historical frequency indicators. The weighted summation of the assigned indicators yields a comprehensive assigned value, calculated using the following formula: Overall value = Impact range index × b1 + Severity index × b2 + Urgency index × b3 + Historical frequency index × b4; Among them, b1, b2, b3 and b4 are all preset weight coefficients used to match the comprehensive value with the corresponding evaluation interval; the influence range index, severity index, urgency index and historical frequency index are all value assignment indicators. Based on the comprehensive assignment, the corresponding anomaly level information is generated.

[0013] A second aspect of this disclosure provides a system for detecting streaming anomalies in unmanned mining trucks, the system comprising: The real-time data acquisition module is used to collect multi-dimensional real-time data from unmanned mining trucks; The detection algorithm quantification module is used to aggregate several streaming anomaly detection algorithms, conduct simulated tests on unmanned mining truck scenarios based on each anomaly detection algorithm, quantify the latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm, and generate quantitative indicators for each anomaly detection algorithm; the quantitative indicators include latency indicators, accuracy indicators, concept drift adaptability indicators, and computational complexity indicators. The comprehensive score calculation module is used to calculate the comprehensive score corresponding to each anomaly detection algorithm based on the quantitative indicators and preset indicator weights, and to rank the anomaly detection algorithms based on the comprehensive scores. The final target algorithm screening module is used to select one or more combinations of anomaly detection algorithms as the final target algorithm based on the comprehensive score and the preset constraints of the unmanned mining truck. An anomaly information generation module is used to input the preprocessed real-time data into the target algorithm to obtain anomaly results and generate anomaly level information based on the anomaly results.

[0014] Preferably, the system further includes a data preprocessing module for preprocessing the real-time data.

[0015] In a third aspect, this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the unmanned mining truck abnormal data detection method described above.

[0016] In a fourth aspect, this disclosure provides an electronic device including a processor and a memory, the processor being configured to execute a computer program stored in the memory to implement the unmanned mining truck abnormal data detection method as described above.

[0017] The beneficial effects of this disclosure are as follows: This disclosure proposes a method and system for detecting abnormal data in unmanned mining trucks. It innovatively constructs a multi-dimensional evaluation system of "latency-accuracy-concept drift adaptability-computational complexity" to build quantitative indicators, and directly links the quantitative indicators to the core needs of the mine, avoiding the algorithm selection bias caused by a single indicator. Various abnormal data detection algorithms are screened through quantitative indicators. By using weighted summation of quantitative algorithm scores, and combining the comprehensive score with the preset constraints of the mining truck, one or more optimal abnormal data detection algorithms are dynamically selected as the final target algorithm, either individually or in combination. This solves the subjective problem of traditional experience-based selection and improves the efficiency of algorithm adaptation to the scenario.

[0018] In this disclosure, the preset constraints can be a set of actual constraints pre-constructed based on the actual business needs, data characteristics, and hardware resources of the unmanned mining truck. This set of actual constraints is matched with the data constraints of the analyzed real-time data, thereby filtering various anomaly detection algorithms and ensuring full consideration of the actual constraints of the unmanned mining truck. Subsequently, this disclosure assigns multi-dimensional scores to the anomaly results generated after the final target algorithm inputs real-time data, based on "impact scope, severity, urgency, and historical frequency," and calculates a weighted comprehensive risk value, matching it to a four-level rating. This achieves precise graded response, ensuring accurate and dynamic monitoring of anomaly registration information and anomaly results, balancing safety assurance and production efficiency.

[0019] Compared with existing technologies, this disclosure has the following advantages: (1) Data quality and validity protection: This disclosure uses multi-dimensional data processing technology to ensure the accuracy, integrity and consistency of input data, providing a reliable foundation for subsequent anomaly detection.

[0020] (2) Protection of algorithm reliability and adaptability: This disclosure ensures that the algorithm meets the requirements of mining card scenarios in terms of real-time performance, accuracy, environmental adaptability and resource consumption through full-dimensional quantitative evaluation, and avoids abnormal missed detection or system overload caused by algorithm defects.

[0021] (3) Algorithm and scenario adaptability protection: This disclosure ensures that the algorithm matches the actual business, data characteristics and hardware resources through constraint-driven algorithm selection logic, avoiding the problem of "excellent algorithm but unsuitable scenario".

[0022] (4) Abnormal risk classification and response protection: This disclosure achieves accurate assessment and graded response of abnormal risks through multi-dimensional quantitative classification, avoiding efficiency loss caused by small risks being ignored or over-handled. Attached Figure Description The accompanying drawings, which form part of this application, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0023] Figure 1 The flowchart of the abnormal data detection method for unmanned mining trucks described in Embodiment 1 of this disclosure is as follows. Figure 1 .

[0024] Figure 2 The flowchart of the abnormal data detection method for unmanned mining trucks described in Embodiment 1 of this disclosure is as follows. Figure 2 The dotted lines indicate optional items.

[0025] Figure 3 This disclosure provides an architecture for a streaming anomaly detection system for unmanned mining trucks, as shown in Embodiment 2. Figure 1 .

[0026] Figure 4 This disclosure provides an architecture for a streaming anomaly detection system for unmanned mining trucks, as shown in Embodiment 2. Figure 2 The dotted lines indicate optional items. Detailed Implementation

[0027] The present disclosure will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present application can be combined with each other.

[0028] The following detailed descriptions are exemplary and intended to provide further detailed explanation of this disclosure. Unless otherwise specified, all technical terms used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this disclosure.

[0029] Example 1: like Figure 1 As shown, this disclosure provides a method for detecting abnormal data from unmanned mining trucks, the method comprising: S100: Collects multi-dimensional real-time data from unmanned mining trucks; S200: Several streaming anomaly detection algorithms are collected, and simulation tests are conducted on unmanned mining truck scenarios based on each anomaly detection algorithm. The latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm are quantified, and quantitative indicators corresponding to each anomaly detection algorithm are generated. The quantitative indicators include latency indicators, accuracy indicators, concept drift adaptability indicators, and computational complexity indicators. S300. Calculate the comprehensive score corresponding to each anomaly detection algorithm based on the quantitative index and the preset index weight, and rank each anomaly detection algorithm based on the comprehensive score. S400. Based on the comprehensive score and the preset constraints of the unmanned mining truck, select one or more combinations of anomaly detection algorithms as the final target algorithm. S500: Input the preprocessed real-time data into the target algorithm to obtain abnormal results, and generate abnormality level information based on the abnormal results.

[0030] Furthermore, the data sources for the real-time data include multiple types of sensors and control systems.

[0031] Furthermore, the real-time data is collected, transmitted, and processed in the form of a data stream.

[0032] Furthermore, the method of acquiring the real-time data can be determined according to actual needs. For example, it can acquire three-dimensional point cloud data of the surrounding environment through millimeter-wave radar and lidar, output real-time position (latitude and longitude, altitude) and motion attitude (heading angle, pitch angle) from GPS and IMU (inertial measurement unit), transmit core mechanical parameters (driving speed, engine speed, braking pressure, hydraulic system pressure) through vehicle bus (CAN / LIN), and high-frequency monitoring data such as temperature sensor (engine water temperature, bearing temperature) and vibration sensor (vehicle vibration acceleration) to form a multi-dimensional real-time data stream that includes environmental perception, motion status, and equipment operating conditions (500KB to 2MB of data per second generated per card, sampling frequency 50-100Hz).

[0033] In a feasible implementation, step S200 involves aggregating several anomaly detection algorithms, conducting simulation tests on unmanned mining truck scenarios based on each algorithm, quantifying the latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm, and generating quantitative indicators for each anomaly detection algorithm. This includes the following steps: S201. Gather N anomaly detection algorithms and number them sequentially, with the number of the anomaly detection algorithm denoted as i; where 1 <= i <= N, N >= 3; S202. Quantify and score the latency, accuracy, concept drift adaptability, and computational complexity corresponding to each anomaly detection algorithm i to obtain the quantitative index corresponding to each anomaly detection algorithm i; wherein, the quantitative index includes latency index, accuracy index, concept drift adaptability index, and computational complexity index.

[0034] In step S202, the latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm i are quantitatively scored. This can be done using existing technologies or in the following ways.

[0035] Optionally, in step S202, the delay is scored. The delay refers to the time interval from the input of real-time data to the anomaly detection algorithm to the output result. In a simulated streaming scenario, continuous real-time data (such as sensor data from an unmanned mining truck) is input to the anomaly detection algorithm. The time interval from the generation time of each data point to the time when the algorithm outputs the anomaly judgment result is recorded. The average of the time intervals calculated after repeating this process multiple times is used as the average delay. At the same time, the corresponding maximum delay and 95th percentile delay are obtained. The delay is scored to obtain a delay score. The delay scores are summarized into a delay index, and the scoring range is 0-10 points.

[0036] Scoring the delay to obtain a delay score can be achieved using existing technology, or it can be configured as needed, or the following methods can be referenced: If the average delay is less than 10ms, the 95th percentile delay is less than 20ms, and the maximum delay is less than 50ms, the score is determined to be 10 points. The average latency is 10-20ms, the 95th percentile latency is 20-50ms, and the maximum latency is 50-100ms, which determines the score as 8 points; The average latency is 20-50ms, the 95th percentile latency is 50-100ms, and the maximum latency is 100-200ms, which determines the score as 6 points. If the above threshold is exceeded (e.g., average latency > 50ms), the score will be determined and points will be deducted proportionally (e.g., 1 point will be deducted for every 10ms exceeding the threshold, with a minimum of 0 points).

[0037] Optionally, in step S202, the accuracy is scored using a labeled real-time dataset (which may contain known anomalous samples, such as labeled data for "motor overload" and "communication interruption" in mining trucks). The F1 score and AUC value of the anomaly detection algorithm on the full real-time dataset are calculated, and the F1 score combines precision (P=TP / (TP+FP)) and recall (R=TP / (TP+FN)). TP represents the true number (the actual anomaly is also judged as anomaly by the algorithm, and the algorithm's judgment is correct). TP (true number) is calculated using the maximum matching window—if the algorithm's judgment coverage within the time interval of the anomalous event is ≥70%, it is counted as 1 TP. FP represents the false positive number (the actual anomaly is normal, but the algorithm misjudges it as anomaly), and FN represents the missing number (the actual anomaly is missed by the algorithm as normal). The accuracy is calculated according to the formula F1=2PR / (P+R). The accuracy is scored to obtain the accuracy score, and the accuracy scores are summarized into an accuracy index, with a scoring range of 0-10 points.

[0038] The accuracy rate is scored to obtain an accuracy score. This can be achieved using existing technology, or it can be set as needed, or the following methods can be referenced: F1 ≥ 0.9 and AUC ≥ 0.95 determines the score to be 10 points; The condition 0.8 ≤ F1 < 0.9 and 0.9 ≤ AUC < 0.95 determines the score to be 8 points; The condition 0.7 ≤ F1 < 0.8 and 0.8 ≤ AUC < 0.9 determines the score to be 6 points; 0.6≤F1<0.7 and 0.7≤AUC<0.8 determines the score to be 4 points; If F1 < 0.6 or AUC < 0.7, the score is determined and points are deducted proportionally (e.g., F1 = 0.5 determines a score of 3 points, with a minimum of 0 points).

[0039] Optionally, in S202, the concept drift adaptability is scored. Concept drift adaptability specifically represents the change in the distribution of real-time data over time (e.g., changes in sensor data distribution due to aging of mining trucks). The algorithm needs to adapt quickly to maintain performance. The algorithm's F1 score represents the long-term stability of the algorithm and is used to measure its performance. After the drift occurs, the decrease in the algorithm's F1 score relative to before the drift (e.g., from 0.9 to 0.7, a decrease of 22%) and the time required for the algorithm's performance (F1 score) to recover to 90% of its pre-drift level (e.g., recovery within 50 data points) are obtained. The concept drift adaptability is then scored to obtain a concept drift adaptability score. The concept drift adaptability scores are summarized into a concept drift adaptability index, with a scoring range of 0-10 points.

[0040] Scoring concept drift adaptability to obtain a concept drift adaptability score can be achieved using existing technologies, or it can be set as needed, or the following methods can be referenced: If the decline is less than 10%, the recovery rate is less than 100 data points, and the long-term stability F1 is greater than or equal to 0.85, then the score is determined to be 10 points. If the decline is 10%-20%, the recovery speed is 100-300 data points, and the long-term stability F1=0.8-0.85, then the score is determined to be 8 points; If the decline is 20%-30%, the recovery speed is 300-500 data points, and the long-term stability F1=0.7-0.8, then the score is determined to be 6 points; If the decline is greater than 30%, the recovery speed is greater than 500 data points, or the long-term stability F1 is less than 0.7, the score will be determined by deducting points proportionally (e.g., a decline of 40% will result in a score of 4 points, with a minimum of 0 points).

[0041] Optionally, in step S202, the computational complexity is scored to obtain a score. Computational complexity reflects the algorithm's resource requirements (time / space). Computational complexity includes time complexity and space complexity, determined by obtaining the algorithm's running time as the data volume n increases (e.g., O(n), O(nlogn), O(n...). 2 The algorithm's memory usage during runtime and the growth trend of data volume n (e.g., O(1), O(n)) are obtained. At the same time, the peak time and memory usage for processing 100,000 data points under a fixed hardware environment (e.g., mining terminal CPU) are obtained. The computational complexity is scored to obtain a computational complexity score. The computational complexity scores are summarized into a computational complexity index, and the scoring range is 0-10 points.

[0042] The computational complexity is scored to obtain a score, which can be achieved using existing technologies, set as needed, or by referring to the following methods: The time complexity is O(n) and the space complexity is O(1). In practice, processing 100,000 data entries takes less than 1 second and requires less than 100MB of memory. The score is 10 points. The time complexity is O(nlogn) and the space complexity is O(n). The actual time taken is 1-3 seconds and the memory usage is 100-500MB. The score is determined to be 8 points. Time complexity O(n 2 The actual time taken is 3-10 seconds, and the memory usage is 500MB-1GB, so the score is 5 points. A score is determined by whether the time complexity is O(2n), the actual execution time is greater than 10 seconds, or the memory usage is greater than 1GB. Points are deducted proportionally based on this score (e.g., if the time complexity reaches O(n...). 3If the score is 2 points, the minimum score is 0 points; of course, other proportional deduction standards can also be set. Where n represents the amount of data.

[0043] In a feasible implementation, step S300 involves calculating a comprehensive score for each anomaly detection algorithm based on the quantitative indicators and preset indicator weights, and then ranking the anomaly detection algorithms based on the comprehensive scores, including the following steps: S301. Calculate the comprehensive score corresponding to each anomaly detection algorithm based on the quantification index and the preset index weights, using the following formula: Overall score = Latency index × a1 + Accuracy index × a2 + Concept drift adaptability index × a3 + Computational complexity index × a4; Where a1, a2, a3 and a4 are preset indicator weights, which respectively correspond to the latency, accuracy, concept drift adaptability and computational complexity of the anomaly detection algorithm; The latency metric, accuracy metric, concept drift adaptability metric, and computational complexity metric are the quantitative metrics corresponding to the anomaly detection algorithm. S302. Sort all the anomaly detection algorithms in descending order of their corresponding comprehensive scores.

[0044] In one feasible implementation, step S400, based on the comprehensive score and the preset constraints of the unmanned mining truck, selects one or more anomaly detection algorithms as target algorithms, including the following steps S401 to S406.

[0045] S401. The preset constraints are defined as a set of actual constraints pre-constructed based on the actual business needs, data characteristics, and hardware resources of the unmanned mining truck. The set of actual constraints is {actual business needs constraints, data characteristics constraints, and hardware resource constraints}.

[0046] The actual business needs, data characteristics, and hardware resources corresponding to the unmanned mining trucks are representative of the available data on unmanned mining trucks and are suitable as constraints on real-time conditions.

[0047] S402. Analyze and derive the data constraints of the real-time data. The range of values ​​for the data constraints is subject to the constraints of the actual constraint set—that is, the data constraints correspond to one or more of the constraints of {actual business requirement constraints, data characteristic constraints, and hardware resource constraints}.

[0048] The data constraints obtained from the analysis of the real-time data are limited to the actual constraint set, which facilitates the establishment of a connection between the actual situation of the real-time data and the data of representative unmanned mining trucks, and makes it easier to select the anomaly detection algorithm that meets the conditions as the final target algorithm.

[0049] S403. If the data constraint does not correspond to any of the constraints in the actual constraint set, it indicates an analysis error or analysis failure, and the analysis should be repeated. Therefore, return to S402.

[0050] If the data constraints of the real-time data obtained from the analysis do not correspond to any of the constraints in the actual constraint set, it indicates that the connection establishment has failed. The real-time data should be re-analyzed until data constraints in the real-time data that can establish a correspondence with the actual business needs, data characteristics, and hardware resources of the unmanned mining truck are found. That is, the data constraints found need to correspond to one or more of the constraints in {actual business needs constraints, data characteristic constraints, and hardware resource constraints}.

[0051] S404. If the data constraint corresponds only to one type of constraint in the actual constraint set, then the constraint corresponding to the data constraint is used as the filtering condition to select the anomaly detection algorithm with the highest comprehensive score from each anomaly detection algorithm, which is then used as the final target algorithm.

[0052] For ease of understanding, assuming that the data constraints in S404 correspond to "actual business requirement constraints", then the "actual business requirement constraints" will be used as a filtering condition to select one or more anomaly detection algorithms with the highest comprehensive scores from among the various anomaly detection algorithms, which will be used as the final target algorithm.

[0053] S405. If the data constraint corresponds to only two constraints in the actual constraint set, then the two constraints corresponding to the data constraint are used together as the first screening condition. Based on the first screening condition, the anomaly detection algorithm with the largest comprehensive score is selected from each anomaly detection algorithm and used as the first target algorithm.

[0054] The other constraint that does not correspond to the data constraint is used as the second screening condition. Based on the second screening condition, the anomaly detection algorithm with the largest comprehensive score is selected from each anomaly detection algorithm and used as the second target algorithm. The first target algorithm and the second target algorithm are combined to form the final target algorithm.

[0055] For ease of understanding, assuming that the data constraints in S405 correspond to {actual business requirement constraints, hardware resource constraints}, then the "actual business requirement constraints" and "hardware resource constraints" are used together as the first screening condition. From each anomaly detection algorithm, one or more anomaly detection algorithms with the highest comprehensive score are selected as the first target algorithm.

[0056] Continuing, the "data characteristic constraints" that do not correspond to the data constraints are used as the second screening condition. From each anomaly detection algorithm, one or more anomaly detection algorithms with the highest comprehensive scores are selected as the second target algorithms.

[0057] Finally, the first target algorithm and the second target algorithm are combined to form the final target algorithm.

[0058] S406. If the data constraint corresponds to all constraints in the actual constraint set, then all constraints in the actual constraint set are used as screening conditions to select the anomaly detection algorithm with the highest comprehensive score from each anomaly detection algorithm, which is then used as the final target algorithm.

[0059] For ease of understanding, if the data constraints simultaneously correspond to {actual business requirement constraints, data characteristic constraints, and hardware resource constraints}—that is, all constraints in the actual constraint set are used together as filtering conditions to select one or more anomaly detection algorithms with the highest comprehensive score from each anomaly detection algorithm.

[0060] Furthermore, if the anomaly detection algorithm with the highest comprehensive score is selected from the various anomaly detection algorithms, and several parallel anomaly detection algorithms are selected, then the parallel anomaly detection algorithms are combined into a new anomaly detection algorithm.

[0061] If multiple anomaly detection algorithms that simultaneously meet the filtering criteria appear during the screening process, these multiple anomaly detection algorithms will be combined and processed to form a single anomaly detection algorithm.

[0062] In one feasible implementation, such as Figure 2 As shown, before inputting the preprocessed real-time data into the target algorithm in S500, the method further includes: preprocessing the real-time data.

[0063] It is important to note that Figure 2 This is just an example to show that "preprocessing the real-time data" is done before S500, but "preprocessing the real-time data" only needs to be done between S100 and S500, and does not necessarily have to be done after S200, S300, and S400.

[0064] Furthermore, the preprocessing of the real-time data includes: performing noise processing, missing value processing, and standardization on the real-time data to obtain a preprocessed data stream.

[0065] Furthermore, the preprocessing of the real-time data includes: extracting real-time features from the preprocessed data stream, wherein the real-time features include statistical features, time-series features, and multimodal correlation features.

[0066] Furthermore, noise processing can be performed as follows: for high-frequency noise from the sensor, sliding window filtering (such as mean filtering, median filtering) or Kalman filtering (suitable for time series data, combining predicted and observed values ​​to optimize the results) can be used.

[0067] Furthermore, missing value handling can be as follows: for short-term missing values ​​(such as 1-2 sampling points), fill them with linear interpolation (based on data fitting between previous and subsequent times) or historical values ​​from the same period (such as the historical average speed of the same road segment); for long-term missing values, trigger sensor fault warnings.

[0068] Furthermore, standardization can be achieved by normalizing different sensor data (such as speed km / h, temperature ℃, pressure kPa) to the range of [0, 1] or [-1, 1] (such as min-max normalization) to eliminate dimensional differences.

[0069] Furthermore, statistical characteristics may include: mean (such as average velocity), variance (the degree of data fluctuation), extreme values ​​(maximum acceleration), and kurtosis (the degree to which data deviates from a normal distribution; high kurtosis may contain anomalous impulses).

[0070] Furthermore, time-series characteristics may include: rate of change, such as rate of change of velocity (acceleration), rate of change of steering angle (reflecting whether steering is smooth), and time-series correlation, such as the position difference of three consecutive moments (to determine whether the speed is uniform or variable).

[0071] Furthermore, multimodal correlation features can include: spatial correlation (e.g., the positional deviation between GPS positioning and lidar odometer should be <0.5m, exceeding the deviation indicates a positioning system failure), physical constraints (e.g., the rate of change of speed should be negative when braking pressure increases, and the opposite indicates a risk of braking failure), and environmental adaptability (e.g., when the contrast of camera images decreases in rainy weather, the confidence of radar obstacle detection needs to be increased by 30%). By constructing a correlation matrix of "sensor-parameter-environment", isolated data can be transformed into collaborative features with physical meaning.

[0072] In one feasible implementation, step S500 involves inputting the preprocessed real-time data into the target algorithm to obtain anomaly results, and generating anomaly level information based on the anomaly results, including the following steps: S501. Input the preprocessed real-time data into the target algorithm to obtain abnormal results; S502. Assign scores to the impact range, severity, urgency, and historical frequency of the abnormal results to obtain the assigned indicators for the abnormal results; the assigned indicators include impact range indicators, severity indicators, urgency indicators, and historical frequency indicators. S503. The weighted summation of the assigned indicators is used to obtain a comprehensive assigned value. The formula for calculating the comprehensive assigned value is as follows: Overall value = Impact range index × b1 + Severity index × b2 + Urgency index × b3 + Historical frequency index × b4; Among them, b1, b2, b3 and b4 are all preset weight coefficients used to match the comprehensive value with the corresponding evaluation interval; the influence range index, severity index, urgency index and historical frequency index are all value assignment indicators. S504. Generate corresponding exception level information based on the comprehensive assignment.

[0073] Furthermore, in step S502, the scoring of the impact range can be performed based on the scenario description corresponding to the acquired abnormal situation, specifically as follows: If the scenario is described as an anomaly causing all mining trucks in the mining area to stop working (such as a failure of the central dispatch system), then the scope of impact will be assigned a value of 10 (involving more than 50 devices or triggering a safety incident). If the scenario is described as an anomaly causing a single fleet (10-50 units) of mining trucks to stop operating, or a critical system (such as power or autopilot) to fail and require manual intervention to restart, then the impact range will be assigned a score of 8-9 (if it affects 1 fleet / critical system, the score will be 8 points; if it affects 2 fleets, the score will be 9 points). If the scenario is described as an anomaly causing partial malfunction of a single mining truck (e.g., a sensor failure, resulting in a reduction to semi-autonomous driving), and it does not affect driving but requires a speed reduction, then the impact range will be assigned a score of 5-7 (if partial malfunction of a single truck is assigned a score of 5; if it affects operational efficiency (e.g., mining efficiency drops by 50%), then it will be assigned a score of 6-7). If the scenario description is that the anomaly only triggers a local alarm (such as abnormal data from a non-critical sensor (such as the temperature inside the carriage)) and does not affect operation, then the impact range will be assigned a score of 2-4 (if it is only a local alarm and has no functional impact, assign a score of 2; if a short shutdown for calibration is required, assign a score of 3-4). If the scenario description is an anomaly that is filtered out by the algorithm and has no actual impact (such as a false alarm triggered by instantaneous noise from the sensor that recovers automatically), then the impact range will be assigned a score of 1.

[0074] Furthermore, in step S502, the severity level is assigned a score, which can be done based on the scenario description corresponding to the acquired anomaly. Specifically, it can be done as follows: If the scenario description is that the abnormality directly leads to the scrapping of equipment (such as battery explosion or engine block rupture), or causes personal injury or major safety accident, then the severity will be assigned 10 points (equipment scrapping / personal injury, assigned 10 points). If the scenario description is that an anomaly causes equipment downtime of ≥24 hours (such as the core module of the autonomous driving system burning out and requiring return to the factory for repair), or causes an environmental accident (such as a diesel leak), then the severity will be assigned a score of 8-9 (downtime ≥24 hours / environmental accident, assigned a score of 8; downtime ≥48 hours / environmental accident, assigned a score of 9). If the scenario is described as an anomaly causing equipment downtime of 1-24 hours (e.g., sensor failure, replacement of spare parts takes 3 hours), or production efficiency drops by ≥50%, then the severity will be assigned a score of 5-7 (downtime of 1-24 hours, assigned 5 points; efficiency drops by 50-80%, assigned 6 points; efficiency drops by more than 80%, assigned 7 points). If the scenario is described as an anomaly causing the equipment to operate at reduced capacity (e.g., power system power limit, efficiency reduction of 10-50%), but no shutdown is required, then the severity will be assigned a score of 2-4 (reduced capacity operation (if efficiency reduction is 10-30%, assign 2 points; if efficiency reduction is 30-50%, assign 3-4 points)). If the scenario description is that the anomaly does not affect the operation and efficiency of the device (such as a failure of the log recording module, which only affects post-event traceability), then the severity will be assigned a score of 1 (only affecting non-core functions, assigned a score of 1).

[0075] Furthermore, in step S502, the urgency level is assigned a score, which can be done based on the scenario description corresponding to the acquired abnormal situation, specifically as follows: If the scenario is described as an anomaly that could cause a secondary accident within 1 minute (such as a brake system malfunction that could lead to loss of control and collision after 1 minute), requiring a millisecond-level response, then the urgency level is assigned a score of 10 (10 points if the handling window is ≤ 1 minute). If the scenario is described as an anomaly that could damage the device within 1-10 minutes (e.g., battery overheating, which could lead to a fire after 10 minutes) requiring a minute-level response, then the urgency level should be assigned a score of 8-9 (8 points for a 1-5 minute handling window, and 9 points for a 5-10 minute handling window). If the scenario is described as an anomaly affecting functionality within 10 minutes to 2 hours (e.g., hydraulic system leak, insufficient pressure after 2 hours), requiring an hourly response, then the urgency level should be assigned a score of 5-7 (5 points for a handling window of 10 minutes to 1 hour, and 6-7 points for 1-2 hours). If the scenario is described as an anomaly affecting efficiency within 2-24 hours (e.g., sensor drift, positioning error exceeding the threshold after 24 hours), requiring a daily response, then the urgency level should be assigned a score of 2-4 (2 points for a handling window of 2-12 hours, and 3-4 points for 12-24 hours). If the scenario is described as an anomaly that may only have an impact after 24 hours (such as slow firmware aging of the device, which may only report an error after 30 days), and can be handled through regular inspections, then the urgency level will be assigned a score of 1.

[0076] Furthermore, in step S502, the historical frequency is assigned a score, which can be done based on the scenario description corresponding to the acquired abnormal situation, specifically as follows: If the scenario is described as the same type of anomaly (such as a sensor failure) occurring ≥5 times within 30 days, and there are no effective corrective measures, then the historical frequency will be assigned a score of 10 (≥5 times within 30 days, assigned a score of 10). If the scenario description is that the same type of abnormality occurs 3-4 times within 30 days, or ≥12 times within 1 year, then the score is 8-9 points (3-4 times within 30 days, 8 points; ≥12 times within 1 year, 9 points based on the historical frequency (the score needs to be combined with the rectification effect; if the frequency is still high after rectification, add 1 point to the score)). If the scenario description is that the same type of anomaly occurs 1-2 times within 30 days, or 6-11 times within 1 year, then the score is 5-7 (5 points for 1-2 times within 30 days, 6-7 points for 6-11 times within 1 year); if the scenario description is that the same type of anomaly occurs 3-5 times within 1 year, or occurs for the first time within 3 years, then the score is 2-4 (2 points for 3-5 times within 1 year, 3-4 points for the first time (it is necessary to pay attention to whether it is a new risk)). If the scenario description indicates that the same type of exception has not occurred for more than 3 years, or is a completely new exception type (without historical records), then a score of 1 is assigned.

[0077] Furthermore, step S504, generating corresponding exception level information based on the comprehensive assignment, may include: If the overall score is between 0 and 3, the corresponding exception level is set to level 1; If the overall score is between 3 and 6, the corresponding exception level is set to level two; If the overall score is between 6 and 8, the corresponding exception level is set to level three. If the overall score is between 8 and 10, the corresponding exception level is set to level four. Generate corresponding anomaly level information based on the corresponding anomaly level.

[0078] Example 2: like Figure 3As shown, in a second aspect, this disclosure provides a system for detecting streaming anomaly data from unmanned mining trucks, the system comprising: The real-time data acquisition module 100 is used to collect multi-dimensional real-time data from unmanned mining trucks. The detection algorithm quantization module 200 is used to aggregate several streaming anomaly detection algorithms, conduct simulation tests on unmanned mining truck scenarios based on each anomaly detection algorithm, quantify the latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm, and generate quantitative indicators for each anomaly detection algorithm; the quantitative indicators include latency indicators, accuracy indicators, concept drift adaptability indicators, and computational complexity indicators. The comprehensive score calculation module 300 is used to calculate the comprehensive score corresponding to each anomaly detection algorithm based on the quantitative index and the preset index weight, and to rank each anomaly detection algorithm based on the comprehensive score. The final target algorithm screening module 400 is used to select one or more combinations of anomaly detection algorithms as the final target algorithm based on the comprehensive score and the preset constraints of the unmanned mining truck. The anomaly information generation module 500 is used to input the preprocessed real-time data into the target algorithm to obtain anomaly results and generate anomaly level information based on the anomaly results.

[0079] Furthermore, such as Figure 4 As shown, the system further includes a data preprocessing module for preprocessing the real-time data. It should be noted that... Figure 4 This is just an example, indicating that the "data preprocessing module" is before the anomaly information generation module 500. However, the "data preprocessing module" only needs to be between the real-time data acquisition module 100 and the anomaly information generation module 500, and does not necessarily have to be performed after the detection algorithm quantification module 200, the comprehensive score calculation module 300, and the final target algorithm screening module 400.

[0080] In the system described in Example 2, the real-time data acquisition module 100, the detection algorithm quantization module 200, the comprehensive score calculation module 300, the final target algorithm screening module 400, and the abnormal information generation module 500 correspond to S100, S200, S300, S400, and S500 in the method described in Example 1, respectively.

[0081] It is worth noting that the system described in Embodiment 2 is only one system implementation of the unmanned mining truck abnormal data detection method, and does not limit the unmanned mining truck abnormal data detection method to rely on the system described in Embodiment 2.

[0082] Example 3: Embodiment 3 of this disclosure provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the abnormal data detection method for unmanned mining trucks as described in Embodiment 1. Alternatively, a streaming anomaly detection system for unmanned mining trucks as described in Example 2 can be implemented.

[0083] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules, or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer.

[0084] Example 4: Embodiment 4 of this disclosure provides an electronic device, which includes a processor and a memory. The processor is used to execute a computer program stored in the memory to implement the unmanned mining truck abnormal data detection method described in Embodiment 1.

[0085] Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0086] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0091] In summary, the unmanned mining truck abnormal data detection method and system provided in embodiments 1-4 of this disclosure innovatively constructs a multi-dimensional evaluation system of "latency-accuracy-concept drift adaptability-computational complexity" to build quantitative indicators, and directly links the quantitative indicators to the core needs of the mine, avoiding the algorithm selection bias caused by a single indicator. Various abnormal data detection algorithms are screened through quantitative indicators, and the comprehensive score of the quantitative algorithm is obtained by weighted summation. Combined with the comprehensive score and the preset constraints of the mining truck, one or more optimal abnormal data detection algorithms are dynamically selected as the final target algorithm, either individually or in combination. This solves the subjective problem of traditional experience-based selection and improves the efficiency of algorithm adaptation to the scenario.

[0092] In this disclosure, the preset constraints can be a set of actual constraints pre-constructed based on the actual business needs, data characteristics, and hardware resources of the unmanned mining truck. This set of actual constraints is matched with the data constraints of the analyzed real-time data, thereby filtering various anomaly detection algorithms and ensuring full consideration of the actual constraints of the unmanned mining truck. Subsequently, this disclosure assigns multi-dimensional scores to the anomaly results generated after the final target algorithm inputs real-time data, based on "impact scope, severity, urgency, and historical frequency," and calculates a weighted comprehensive risk value, matching it to a four-level rating. This achieves precise graded response, ensuring accurate and dynamic monitoring of anomaly registration information and anomaly results, balancing safety assurance and production efficiency.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure and not to limit them. Although this disclosure has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this disclosure. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this disclosure should be covered within the protection scope of the claims of this disclosure.

Claims

1. A method for detecting abnormal data from unmanned mining trucks, characterized in that, The method includes: Collect multi-dimensional real-time data from unmanned mining trucks; Several streaming anomaly detection algorithms are collected, and simulation tests are conducted on unmanned mining truck scenarios based on each anomaly detection algorithm. The latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm are quantified, and quantitative indicators corresponding to each anomaly detection algorithm are generated. The quantitative indicators include latency indicators, accuracy indicators, concept drift adaptability indicators, and computational complexity indicators. The comprehensive score corresponding to each anomaly detection algorithm is calculated based on the quantitative indicators and the preset indicator weights, and the anomaly detection algorithms are ranked based on the comprehensive scores. Based on the comprehensive score and the preset constraints of the unmanned mining truck, one or more combinations of anomaly detection algorithms are selected as the final target algorithm. The preprocessed real-time data is input into the target algorithm to obtain abnormal results, and abnormality level information is generated based on the abnormal results.

2. The method for detecting abnormal data from unmanned mining trucks as described in claim 1, characterized in that, The real-time data sources include various types of sensors and control systems, and the real-time data is collected, transmitted, and processed in the form of a data stream.

3. The method for detecting abnormal data from unmanned mining trucks as described in claim 1, characterized in that, The process involves compiling several anomaly detection algorithms, simulating and testing unmanned mining truck scenarios based on each algorithm, quantifying the latency, accuracy, concept drift adaptability, and computational complexity of each algorithm, and generating quantitative indicators for each algorithm. This includes the following steps: N anomaly detection algorithms are collected and numbered sequentially, with the number of the anomaly detection algorithm denoted as i; where 1 <= i <= N, N >= 3; The latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm i are quantitatively scored to obtain a quantitative index for each anomaly detection algorithm i; wherein the quantitative index includes a latency index, an accuracy index, a concept drift adaptability index, and a computational complexity index.

4. The method for detecting abnormal data from unmanned mining trucks as described in claim 1, characterized in that, The process of calculating a comprehensive score for each anomaly detection algorithm based on the quantitative indicators and preset indicator weights, and then ranking the anomaly detection algorithms based on the comprehensive scores, includes the following steps: Based on the quantitative indicators and preset indicator weights, the comprehensive score corresponding to each anomaly detection algorithm is calculated, as follows: Overall score = Latency index × a1 + Accuracy index × a2 + Concept drift adaptability index × a3 + Computational complexity index × a4; Where a1, a2, a3 and a4 are preset indicator weights, which respectively correspond to the latency, accuracy, concept drift adaptability and computational complexity of the anomaly detection algorithm; The latency metric, accuracy metric, concept drift adaptability metric, and computational complexity metric are the quantitative metrics corresponding to the anomaly detection algorithm. Sort all the anomaly detection algorithms in descending order of their corresponding comprehensive scores.

5. The method for detecting abnormal data from unmanned mining trucks as described in claim 1, characterized in that, The selection of one or more anomaly detection algorithms as target algorithms based on the comprehensive score and preset constraints of the unmanned mining truck includes the following steps: S401. The preset constraints are defined as a set of actual constraints pre-constructed based on the actual business needs, data characteristics, and hardware resources of the unmanned mining truck. The set of actual constraints is {actual business needs constraints, data characteristics constraints, and hardware resource constraints}. S402. Analyze and derive the data constraints of the real-time data. The range of values ​​for the data constraints is subject to the constraints of the actual constraint set—that is, the data constraints correspond to one or more of the constraints of {actual business requirement constraints, data characteristic constraints, and hardware resource constraints}. S403. If the data constraint does not correspond to any of the constraints in the actual constraint set, it indicates that the analysis is incorrect or has failed, and the analysis should be repeated. Therefore, return to S402. S404. If the data constraint corresponds only to one type of constraint in the actual constraint set, then the constraint corresponding to the data constraint is used as the filtering condition to select the anomaly detection algorithm with the highest comprehensive score from each anomaly detection algorithm, which is then used as the final target algorithm. S405. If the data constraint corresponds to only two constraints in the actual constraint set, then the two constraints corresponding to the data constraint are used together as the first screening condition. Based on the first screening condition, the anomaly detection algorithm with the largest comprehensive score is selected from each anomaly detection algorithm and used as the first target algorithm. The other constraint that does not correspond to the data constraint is used as the second screening condition. Based on the second screening condition, the anomaly detection algorithm with the largest comprehensive score is selected from each anomaly detection algorithm and used as the second target algorithm. The first objective algorithm and the second objective algorithm are combined to form the final objective algorithm; S406. If the data constraint corresponds to all constraints in the actual constraint set, then all constraints in the actual constraint set are used as screening conditions to select the anomaly detection algorithm with the highest comprehensive score from each anomaly detection algorithm, which is then used as the final target algorithm.

6. The method for detecting abnormal data from unmanned mining trucks as described in claim 5, characterized in that, If, when selecting the anomaly detection algorithm with the highest overall score from among the various anomaly detection algorithms, several parallel anomaly detection algorithms are selected, then the parallel anomaly detection algorithms are combined into a new anomaly detection algorithm.

7. The method for detecting abnormal data from unmanned mining trucks as described in claim 1, characterized in that, Before inputting the preprocessed real-time data into the target algorithm, the method further includes: The real-time data is preprocessed.

8. The method for detecting abnormal data from unmanned mining trucks as described in claim 7, characterized in that, The preprocessing of the real-time data includes: The real-time data is subjected to noise processing, missing value processing, and standardization to obtain a preprocessed data stream; Real-time features are extracted from the preprocessed data stream, including statistical features, time-series features, and multimodal correlation features.

9. The method for detecting abnormal data from unmanned mining trucks as described in claim 1, characterized in that, The process of inputting the preprocessed real-time data into the target algorithm to obtain anomaly results and generating anomaly level information based on the anomaly results includes the following steps: The preprocessed real-time data is input into the target algorithm to obtain abnormal results; The impact range, severity, urgency, and historical frequency of the abnormal results are assigned scores to obtain the assigned indicators for the abnormal results; the assigned indicators include impact range indicators, severity indicators, urgency indicators, and historical frequency indicators. The weighted summation of the assigned indicators yields a comprehensive assigned value, calculated using the following formula: Overall value = Impact range index × b1 + Severity index × b2 + Urgency index × b3 + Historical frequency index × b4; Among them, b1, b2, b3 and b4 are all preset weight coefficients used to match the comprehensive value with the corresponding evaluation interval; the influence range index, severity index, urgency index and historical frequency index are all assigned values. Based on the comprehensive assignment, the corresponding anomaly level information is generated.

10. A system for detecting streaming anomaly data in unmanned mining trucks, characterized in that, The system includes: The real-time data acquisition module is used to collect multi-dimensional real-time data from unmanned mining trucks; The detection algorithm quantification module is used to aggregate several streaming anomaly detection algorithms, conduct simulated tests on unmanned mining truck scenarios based on each anomaly detection algorithm, quantify the latency, accuracy, concept drift adaptability, and computational complexity of each anomaly detection algorithm, and generate quantitative indicators for each anomaly detection algorithm; the quantitative indicators include latency indicators, accuracy indicators, concept drift adaptability indicators, and computational complexity indicators. The comprehensive score calculation module is used to calculate the comprehensive score corresponding to each anomaly detection algorithm based on the quantitative indicators and preset indicator weights, and to rank the anomaly detection algorithms based on the comprehensive scores. The final target algorithm screening module is used to select one or more combinations of anomaly detection algorithms as the final target algorithm based on the comprehensive score and the preset constraints of the unmanned mining truck. An anomaly information generation module is used to input the preprocessed real-time data into the target algorithm to obtain anomaly results and generate anomaly level information based on the anomaly results.

Images