An automated refrigerator transport identification system

Abstract

The application discloses an automatic refrigerator transportation identification system and relates to the technical field of transportation. According to the classified transportation task type, the system extracts feature distribution from a clustering result set, constructs a personalized identification feature threshold set Q corresponding to the task type, and thus breaks through the limitation of the traditional general threshold being fixed and unchangeable. In the second identification module, the system collects parameter features in the current transportation process through real-time sensors, combines the Euclidean distance algorithm to quickly complete the task type attribution judgment, so that the system can confirm the "behavior mode" of the transportation task at the initial stage of transportation. In the correction module, the system compares and analyzes the real-time parameters based on the personalized threshold set Q corresponding to the current task, and determines the abnormal dimension Fg.

Description

TECHNICAL FIELD

[0001] The present application relates to the transportation technology field, in particular to an automatic refrigerator transportation identification system. BACKGROUND

[0002] At present, the cold chain transportation system is gradually evolving towards data-driven intelligent analysis, and as the core terminal in cold chain logistics, the behavior judgment, state recognition and operation exception monitoring in the transportation process of the refrigerator have gradually become the key concern. The traditional identification method generally uses general rules and fixed thresholds to judge the transportation behavior, which cannot effectively adapt to the different characteristics of different transportation task scenarios.

[0003] Although the existing refrigerator transportation monitoring system can realize continuous collection of parameters such as temperature, vibration and power, there are still significant deficiencies in the "task behavior identification" aspect. On the one hand, most systems use preset thresholds or manually set rules to judge the transportation process, without considering the natural differences in route structure, item attributes, environmental conditions and other aspects of different transportation tasks, which may cause false positives or false negatives in different scenarios with the same threshold. On the other hand, the current identification mechanism is mainly based on single threshold comparison, lacks behavior trend modeling capability, and is difficult to dynamically adapt to state changes in the transportation process, and is also difficult to evolve its own judgment criteria combined with historical data. SUMMARY

[0004] In view of the above problems existing in the prior art, the present application provides an automatic refrigerator transportation identification system.

[0005] The present application provides an automatic refrigerator transportation identification system, which comprises a set generation module, a first identification module, a second identification module and a correction module.

[0006] The set generation module is used to collect the parameter vector S of the transportation task based on the refrigerator transportation situation in the historical period, and determine the target function L according to the K-means clustering algorithm, and output the clustering result set based on the target function L.

[0007] The first identification module constructs a personalized identification feature threshold set Q for each task category according to the obtained clustering result set.

[0008] The second identification module monitors the real-time features in the current transportation process in real time through the monitoring device, and seeks the current transportation task category combined with the Euclidean distance algorithm.

[0009] The correction module is used to extract the clustering result matched with the current transportation task category from the clustering result set, analyze the abnormal dimension Fg in the current transportation process according to the clustering result matched with the current transportation task category, and trigger the dynamic correction mechanism to identify the feature threshold dynamic backtracking correction operation.

[0010] Optionally, the set generation module comprises a parameter collection unit, a clustering unit and a category identification unit.

[0011] The parameter collection unit is configured to collect the refrigerator transportation conditions in the historical time period by using a plurality of groups of sensors, and determine the refrigerator transportation conditions in each transportation task in the historical time period by distinguishing, and obtain the parameter vector S of the transportation task by statistics, and the expression of the parameter vector S of the transportation task is: ; wherein, to represents the refrigerator transportation conditions from the first historical transportation task to the nth historical transportation task, and n represents the total number of historical transportation tasks.

[0012] Optionally, the clustering unit is configured to converge the objective function L, and the specific steps comprise:

[0013] S11: randomly set K clustering clusters to determine each clustering center , and calculate the Euclidean distance between each transportation task and each clustering center by combining the parameter vector S of the transportation task, so as to obtain the objective function L, and assign the nearest clustering cluster;

[0014] S12: and for each clustering cluster, calculate the mean of the refrigerator transportation conditions of all historical transportation tasks belonging to the corresponding clustering cluster as the clustering center of the new clustering cluster;

[0015] S13: repeat the step S11 and the step S12 until the objective function L converges, that is, the grouping of the historical transportation tasks no longer changes.

[0016] Optionally, the category identification unit is configured to determine the transportation label of each historical transportation task from the objective function L when the grouping of the historical transportation tasks no longer changes, and store the corresponding transportation label into the automatic refrigerator transportation identification background, determine and construct the clustering result set P by statistics, and the expression of the clustering result set P is: ; wherein, to represents the first transportation label p to the mth transportation label p, and m represents the total number of transportation labels; and determines the transportation task category of the corresponding historical transportation task according to the transportation label corresponding to the corresponding historical transportation task.

[0017] Optionally, the first identification module comprises a feature division unit and a personalized division unit.

[0018] The feature division unit is configured to divide the refrigerator transportation situation of the historical transportation task corresponding to each transportation task category based on the clustering result set P obtained in the category identification unit and the identified transportation task category, and generate a feature vector W corresponding to each transportation task category through statistics, and the expression of the feature vector W is: ; to represent the number set of the first feature w to the number set of the xth feature w, and x represents the total number of features.

[0019] Optionally, the individualization division unit is configured to set a corresponding feature threshold for each feature under the corresponding transportation task category according to the feature vector W corresponding to each transportation task category obtained in the feature division unit, so as to generate a set of individualized identification feature thresholds Q under the corresponding transportation task category, and the expression of the set of individualized identification feature thresholds Q is: ; to represent the feature threshold set for the first feature under the corresponding transportation task category to the feature threshold set for the xth feature under the corresponding transportation task category .

[0020] Optionally, the second identification module is configured to use a plurality of groups of sensors to monitor real-time features in the current transportation task process in real time again, wherein the real-time features are the same as the parameter categories collected in the historical time period, and the difference lies in the time difference.

[0021] Optionally, the second identification module is further configured to substitute the real-time features into the clustering centers of the K new clustering clusters that do not change in the clustering unit , and after calculation and matching by the Euclidean distance algorithm, determine the transportation label corresponding to the current transportation task, and identify the current transportation task category according to the transportation label corresponding to the current transportation task.

[0022] Optionally, the correction module includes an extraction unit, an anomaly identification unit, a mechanism triggering unit, and a dynamic correction unit.

[0023] The extraction unit is configured to extract a clustering result that is the same as the current transportation task category from the clustering result set according to the current transportation task category identified by the second identification module, and extract a set of individualized identification feature thresholds Q under the current transportation task category from the individualization division unit according to the extracted clustering result;

[0024] The anomaly identification unit is configured to analyze the abnormal dimension Fg in the current transportation process according to the set of individualized identification feature thresholds Q under the current transportation task category obtained by the extraction unit, and the specific analysis steps of the abnormal dimension Fg are:​

[0025] S21: Compare the real-time features during the current transportation task with the feature thresholds set for each feature in the personalized identification feature threshold set Q under the current transportation task category condition. A comparison is made to determine whether each real-time feature exceeds the feature threshold set for the corresponding feature. ;

[0026] S22: If the corresponding real-time feature exceeds the feature threshold set for the corresponding feature. Each time, an integral is added. After traversing all real-time features, the added integrals are counted to determine the anomaly dimension Fg in the current transportation process.

[0027] Optionally, the mechanism triggering unit is used to pre-set the historical average anomaly dimension based on the personalized identification feature threshold set Q under the corresponding transportation task category conditions in the personalized segmentation unit. The current anomaly dimension Fg in the transportation process exceeds the historical average anomaly dimension. At that time, the corresponding features are determined from the anomaly dimension Fg, and a dynamic correction mechanism is triggered;

[0028] The dynamic correction unit, upon receiving a trigger from the dynamic correction mechanism, records the final transportation result of the current transportation task, compares the final transportation result with the corresponding features determined in the anomaly dimension Fg to determine whether the final transportation result is an incorrect judgment. If it is an incorrect judgment, it combines the current corresponding feature threshold... Set the corresponding feature threshold. Dynamic backtracking and correction; if it is not an incorrect judgment, no corresponding feature threshold is applied. Dynamic backtracking and correction operations.

[0029] This invention provides an automated refrigerated container transportation identification system, which has the following beneficial effects:

[0030] (1) By introducing the K-means clustering algorithm, based on the feature parameter vector S collected from historical transportation tasks, the objective function L is constructed, and the clustering result set is output. The system can automatically identify the task types corresponding to different transportation behaviors, realize the "type management" of transportation tasks, and avoid the problem of poor adaptability of unified rules to different behaviors in traditional systems. In the first identification module, the system extracts the feature distribution from the clustering result set according to the classified transportation task type, and constructs a personalized identification feature threshold set Q corresponding to the task type, thereby breaking through the limitation of fixed and immutable traditional general threshold, significantly improving the precise adaptation ability to different transportation scenarios, and enhancing the judgment accuracy of behaviors such as door operation, energy consumption fluctuation, and temperature mutation. In the second identification module, the system collects parameter features in the current transportation process through real-time sensors, and combines the Euclidean distance algorithm to quickly complete the task type attribution judgment, so that the system can confirm the "behavior pattern" of the transportation task at the initial stage of transportation, greatly improving the real-time and forward-looking of the judgment. The system compares and analyzes the real-time parameters based on the personalized threshold set Q corresponding to the current task, determines the abnormal dimension Fg, and automatically triggers the dynamic correction mechanism when the historical abnormal reference value is exceeded, executes the backtracking update of the threshold model, realizes the weak supervision learning and long-term optimization of the identification model, and effectively reduces the misjudgment rate and the omission rate in long-term use.

[0031] (2) The system introduces the Euclidean distance objective function L through the K-means clustering algorithm based on the parameter vector S, and iteratively optimizes the clustering of historical task samples multiple times, finally divides all historical tasks into K task type clusters, so that the deviation of each task behavior and the center feature of the corresponding category is minimized. This method can fully exploit the potential behavior pattern correlation between refrigerator transportation tasks, improve the identification accuracy and behavior classification accuracy of transportation tasks. The system assigns a transportation label to each transportation task according to the final convergence result of the clustering process, and constructs a clustering result set P to clarify the task type label structure. With the continuous increase of historical data, the system can re-run the clustering process to optimize and evolve the task type structure, realize the "self-update" and "dynamic adaptation" of the task division structure, and avoid the problem of rigid label system that cannot adapt to business changes.

[0032] (3) The system classifies the feature data of historical transportation tasks according to the task categories based on the clustering result set P and the transportation task category recognition result, and then generates a feature vector W under each transportation task category to reflect the numerical distribution range of the task characteristics. Based on the feature vector W, the system generates corresponding personalized recognition feature thresholds for each feature dimension using different statistical mechanisms, and constructs a personalized recognition feature threshold set Q. This set is independently set under each task category, so that the recognition model can flexibly match the tolerance and fluctuation range of different features according to the differences in task types, avoiding the "one-size-fits-all" judgment errors under the general threshold model. Since each feature threshold is derived from the feature quantity set w under the task category to which it belongs, the system can dynamically reconstruct the feature threshold Q when facing sample updates. Through continuous clustering and feature statistics, the model has the ability to adapt and converge in different transportation scenarios, further improving the state recognition accuracy and error tolerance of refrigerator transportation tasks. The "task-feature-threshold" ternary judgment mechanism is constructed: through the complete process from "transportation task category" to "corresponding feature statistical distribution" to "personalized threshold generation", the system realizes a "structure controllable, rule transparent" task recognition system, which not only helps to improve the system's interpretability and safety, but also provides a structured input interface for subsequent real-time task matching, anomaly recognition and model correction.

[0033] (4) The system accurately extracts the corresponding clustering results and personalized recognition feature threshold set Q according to the task category recognized by the current transportation task, ensuring consistency between real-time task recognition and historical behavior modeling. The system compares the real-time features with the feature threshold Q item by item, identifies the abnormal dimensions Fg that exceed the threshold, and realizes multi-feature comprehensive judgment through the integral mechanism to avoid false positives caused by single-point disturbance. When the abnormal dimension Fg exceeds the historical average abnormal level, the dynamic correction mechanism is automatically triggered, and the abnormal feature is verified. If it is determined to be a false positive, the corresponding threshold is slightly corrected in proportion, improving the fault tolerance and accuracy of subsequent task recognition. If it is not a false positive, the original model remains unchanged to ensure model stability. In summary, the system has a closed-loop capability of "self-detection, self-feedback and self-correction", significantly reducing the accumulation of false positives and the risk of false recognition in long-term operation, and achieving high adaptability and robustness of refrigerator transportation task recognition. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application.

[0035] Figure 1 The block diagram of the system of the present application. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0037] Embodiment 1

[0038] Please refer to Figure 1 The present application provides an automatic refrigerator transportation identification system, comprising a set generation module, a first identification module, a second identification module and a correction module.

[0039] The set generation module is used for collecting a parameter vector S of a transportation task based on refrigerator transportation in a historical period, and determining a target function L according to a K-means clustering algorithm, and outputting a clustering result set based on the target function L.

[0040] The first identification module is used for constructing a personalized identification feature threshold set Q for each task category according to the obtained clustering result set.

[0041] The second identification module is used for real-time monitoring of real-time features in the current transportation process through a monitoring device, and seeking a current transportation task category in combination with a Euclidean distance algorithm.

[0042] The correction module is used for extracting a clustering result matched with the current transportation task category from the clustering result set, and analyzing an abnormal dimension Fg in the current transportation process according to the clustering result matched with the current transportation task category, and triggering a dynamic correction mechanism to identify a dynamic backtracking correction operation of the identification feature threshold.

[0043] In the embodiments of the present application, the present application realizes automatic classification of transportation task behavior types by constructing a historical task feature vector set and combining a K-means clustering algorithm, and then generates a personalized identification feature threshold set for each task category, and further realizes high-precision state judgment of the current transportation task by matching and identifying the real-time collected transportation behavior features with the clustering center. Compared with the traditional identification mode based on a unified threshold, the present application has the advantages of strong task behavior adaptability, high abnormality identification precision, dynamic correction of the threshold model and the like, reduces the false positive rate and the missed judgment rate, and improves the intelligent management and control level of the cold chain transportation process.

[0044] ​For example, in the case of a refrigerator used for both city fresh food delivery and vaccine special line transportation, the former has the characteristics of frequent door opening and variable route, while the latter emphasizes temperature stability and low disturbance. The traditional unified door operation frequency threshold may misjudge fresh food delivery as abnormal, while the system can distinguish between the two through historical behavior clustering and set their respective door operation thresholds, path deviation tolerance and energy consumption fluctuation upper limit. When the system detects that the current transportation task has a high door frequency but stable temperature control, it can accurately identify it as a fresh food delivery task and judge it as normal, avoiding false alarms. At the same time, when the identification deviates, the system can also correct the threshold according to the final transportation result, so that the model is closer to the real scene in subsequent tasks, improving the overall recognition stability and self-learning ability of the system.

[0045] Embodiment 2

[0046] Please refer to Figure 1 , specifically: the set generation module includes a parameter acquisition unit, a clustering unit and a category identification unit;

[0047] The parameter acquisition unit is used to collect the refrigerator transportation situation in the historical time period by using a plurality of groups of sensors, and to determine the refrigerator transportation situation of each transportation task in the historical time period by distinguishing, and to obtain the parameter vector S of the transportation task by statistical analysis, and the expression of the parameter vector S of the transportation task is: ; wherein, to represents the refrigerator transportation situation in the first historical transportation task to the n-th historical transportation task, and n represents the total number of historical transportation tasks.

[0048] The refrigerator transportation situation includes but is not limited to door operation frequency, transportation path direction deviation, energy consumption variation, temperature mutation, and vibration disturbance intensity during transportation.

[0049] The clustering unit is used to converge the objective function L, and the specific steps include:

[0050] S11: randomly set K clustering clusters to determine each clustering center , and calculate the Euclidean distance between each transportation task and each clustering center to obtain the objective function L, and then assign it to the nearest clustering cluster; the specific acquisition method is as follows:

[0051] ;

[0052] In the formula, represents the refrigerator transportation situation in the i-th historical transportation task, Let represent the cluster center of the j-th cluster, i represent the historical transportation task number, n represent the total number of historical transportation tasks, j represent the cluster number, and K represent the total number of clusters. Represents the squared Euclidean distance, used to measure the similarity between the refrigerated container transportation situation and the cluster center during the historical transportation mission; This represents finding the partitioning method that minimizes the total distance among all possible clustering methods; C is a clustering partitioning method, which is the overall scheme of assigning which tasks to which cluster centers; This is to find the nearest task category center in order to assign task samples (refrigerated container transportation during the transportation task).

[0053] This means finding the center closest to the refrigerated container transportation status in the i-th historical transportation task process (i.e., the center closest to the refrigerated container transportation status in the i-th historical transportation task process). Minimum Euclidean distance ), and then put To the cluster corresponding to this center, thereby achieving each task Assign to the nearest task category;

[0054] It should be noted that the objective function L aims to divide the refrigerated container transportation information of the n input historical transportation tasks into K clusters, such that the sum of the squared Euclidean distances of each historical transportation task's refrigerated container transportation information to the center of its respective cluster is minimized.

[0055] S12: For each cluster, calculate the mean of all historical refrigerated container transportation data belonging to the corresponding cluster and use it as the cluster center of the new cluster. ;

[0056] S13: Repeat steps S11 and S12 until the objective function L converges, meaning that the grouping of historical transportation tasks no longer changes.

[0057] Specifically, assuming that when the number of historical data increases from 50 to 100, the original K-class structure may no longer accurately represent the new task behavior; the clustering process can be restarted, and the system will form new cluster centers and new task divisions; some old task categories will be split (subdivided behavior), and new categories will also be merged or degenerated to form a new structure. Therefore, the transportation task category is a non-static, iteratively evolving clustering label system.

[0058] The category identification unit is used to determine the transportation tag for each historical transportation task from the objective function L when the grouping of historical transportation tasks no longer changes. The corresponding transportation tags are then stored in the automated refrigerated container transportation identification backend. Statistical analysis is performed to determine and construct a clustering result set P. The expression for the clustering result set P is: ;in, to representing the first transport label p to the mth transport label p, m represents the total number of transport labels; and determining the transport task category of the corresponding historical transport task according to the transport label corresponding to the corresponding historical transport task.

[0059] In the embodiment of the application, the system forms a transport task parameter vector set S by standardizing the refrigerator transportation conditions collected in the historical period, combines the K-means clustering algorithm, constructs an objective function L, dynamically determines the cluster center and cluster division, and automatically forms the task label. This mechanism does not need to preset rules and can adaptively mine the behavior pattern differences in the transport task to realize the automatic generation and continuous evolution of the transport task type.

[0060] For example: a certain transport system history contains vaccine transportation, fresh short transfer and meat long-distance cold chain and other tasks, and three typical transport type labels can be identified by clustering . Among them, has very low gate operation frequency and temperature mutation value, has frequent gate operation but stable energy consumption characteristics, has relatively high vibration disturbance and energy consumption variation. In the traditional way, these differences may be ignored, but the system can automatically classify and form a structured result set , which provides strong support for customizing the identification model for each type of task. This method improves the recognizability of the transport task and the intelligent adaptation ability of the system, ensures that the model still has dynamic evolution ability when new data is continuously added, and significantly enhances the robustness and adaptability of the system to complex cold chain tasks.

[0061] Embodiment 3

[0062] Please refer to Figure 1 , specifically: the first identification module includes a feature division unit and an individualized division unit;

[0063] The feature division unit is used to divide the refrigerator transportation conditions of the historical transport tasks corresponding to each transport task category based on the cluster result set P obtained in the category recognition unit and the identified transport task category, and generate a feature vector W corresponding to each transport task category through statistics. The expression of the feature vector W is: ; to representing the number set w of the first feature to the number set w of the xth feature under the corresponding transport task category condition, x representing the total number of features.

[0064] It should be noted that the number set w of each feature is used to reflect the range of the value of each feature under the corresponding transport task category condition, so as to prepare for the threshold setting of each feature in each transport task category subsequently.

[0065] The personalized segmentation unit is used to set a corresponding feature threshold for each feature under the corresponding transportation task category condition, based on the feature vector W corresponding to each transportation task category obtained from the feature segmentation unit. To generate a personalized identification feature threshold set Q under the corresponding transportation task category conditions, the expression for the personalized identification feature threshold set Q is: ; to This represents the feature threshold set for the first feature under the corresponding transportation task category condition. The feature threshold set up up to the xth feature .

[0066] Specifically, the personalized recognition feature threshold set Q is generated as follows:

[0067] ;

[0068] In the formula, Indicates the number of the shipping label. , , , and They represent the first The first feature setting corresponds to the feature threshold, the second feature setting corresponds to the feature threshold, the third feature setting corresponds to the feature threshold, the fourth feature setting corresponds to the feature threshold, and the fifth feature setting corresponds to the feature threshold; the aforementioned x=5; the features are door operation frequency, transport path direction deviation, energy consumption variation, temperature change, and vibration disturbance intensity during transport.

[0069] , , , and These are the first, second, third, fourth, and fifth features of the r-th transport tag, respectively.

[0070] Indicates the first The first feature threshold under each transport tag is the door operation frequency threshold. This feature has a concentrated trend and stable fluctuations in certain tasks (e.g., doors are rarely opened in the cold chain of pharmaceuticals). The standard deviation is used to measure its natural deviation range in order to avoid misjudgment due to small fluctuations. The threshold is used to distinguish between "normal task behavior fluctuations" and "abnormally frequent door openings", and is strongly adapted to "operationally stable tasks" (such as high-end biological product delivery).

[0071] Indicates the first The second feature threshold under each transport tag is the transport path direction offset threshold. The median and interquartile range (IQR) are used instead of the mean and standard deviation to resist the interference of outliers on the model structure. Since the transport path direction offset contains a large number of outliers (GPS drift, signal loss, etc.), the median + IQR can better capture the core path fluctuation pattern.

[0072] Indicates the first The third feature threshold under each transport tag is the energy consumption variability threshold; the energy consumption level of this type of task is represented by the mean, and... By performing an exponential compression mapping of the standard deviation of energy consumption, the ability to elastically control abnormal fluctuations is enhanced. Since abnormal energy consumption is often caused by frequent start-stop of compressors and abnormal temperature control logic, the logarithmic scale is used to reduce the interference of "extreme values", but it still amplifies the impact of "frequent fluctuations".

[0073] This represents the fourth feature threshold under the r-th transport tag, namely the threshold for temperature mutation; the maximum mutation amplitude is taken as the high-risk indicator, and the coefficient is... The threshold decreases as N increases, thus reducing the overall threshold. When the number of task samples is small, our understanding of that task category is insufficient → allowing for a more lenient threshold → The threshold is relatively large, resulting in a larger overall threshold. When the number of task samples is very large, the system has a better grasp of the feature distribution of this type of task → the threshold should be tightened to improve the sensitivity of anomaly detection → As the threshold decreases, the overall threshold approaches the actual limit. Among these, the coefficient... It is a dynamic tolerance control factor used to introduce a sample size-driven automatic adjustment mechanism when constructing temperature mutation thresholds, realizing an evolutionary path from "coarse identification" to "precise identification". The setting of 0.8 is an empirical setting, representing the maximum tolerable adjustment range (80% weight as compensation when the sample size is small);

[0074] Indicates the first The fifth feature threshold under each transportation label is the vibration disturbance intensity threshold during transportation. Since vibration data has multiple frequencies and amplitudes, it is not suitable to use the mean or median. The square root of the variance (standard deviation) can be used to represent the fluctuation of vibration behavior under this category. Vibration often has noise interference, but variance can reflect the overall disturbance degree. Variance is a relatively appropriate way to characterize the "disturbance distribution range" and can achieve the effect of distinguishing normal vibration (road conditions) from sudden impact (sudden braking, collision).

[0075] Specifically, The average level of door opening frequency, used to provide the benchmark value of door operation behavior for this type of task, reflecting the overall "operation frequency", for example, high frequency of fresh food delivery, low frequency of vaccine transportation;

[0076] The standard deviation of the frequency of door operation in the rth transport label, used to measure whether the door opening behavior is concentrated and stable in the same type of task; if some tasks have frequent door opening and others have few, the standard deviation will be large, indicating "behavior dispersion".

[0077] The median of the path direction deviation, used to avoid the influence of extreme turning or abnormal path drift on the overall judgment, reflecting the typical level of "regular path fluctuation" in the task category;

[0078] The interquartile range (75%-25% difference) of path direction deviation, to measure the width of the concentrated interval of path change;

[0079] The average value of energy consumption variation, used to judge the average stability of the refrigeration system in this type of task; if the refrigerator compressor is frequently started and stopped, the energy consumption variation will be large, and the average value will increase.

[0080] The logarithmic scaling of the standard deviation, to prevent the abnormal amplification of the recognition threshold when the energy consumption variation standard deviation is too large. After taking the logarithm of the energy consumption variation, the system will gradually increase the tolerance rather than linearly, maintaining the controllability of abnormal judgment. Among them, The fluctuation range of energy consumption anomaly, used to reflect the violent fluctuation of energy consumption variation, which is the basis for evaluating the energy consumption jitter risk;

[0081] The maximum value of the temperature mutation rate in the corresponding task category, used as an upper limit indicator of the "extreme reaction ability" of the temperature control system;

[0082] The vibration intensity variance in the corresponding task category, indicating the degree of vibration energy distribution in the overall transportation process.

[0083] It should be noted that the corresponding transport label corresponds to a task category.

[0084] In the embodiment of the application, the system extracts the refrigerator transportation behavior parameters corresponding to each type of task from the clustering result set P according to the completed clustering grouping of the transportation task types, and respectively counts the feature quantity set formed by the behavior parameters in each type of task as the division basis, and further organizes the task type feature vector. The feature vector not only retains the original distribution information of the parameters, but also provides a structured representation of the behavior indicators within the task type, providing high-quality feature input for subsequent modeling.

[0085] The system generates personalized thresholds on five feature dimensions for each feature quantity set in the task type using mean, standard deviation, median, interquartile range, extreme value scaling, and other statistical methods, forming a recognition feature threshold set. Through this method, the system can automatically build a decision boundary that better reflects the behavior characteristics of the task type based on historical data distribution, significantly improving the sensitivity and adaptability of the recognition model to individual type behavior abnormalities.

[0086] The traditional recognition mechanism uses a unified fixed threshold for abnormal judgment, which often cannot adapt to the natural differences in behavior rhythm, operation frequency, and energy consumption mode between task types, leading to frequent misjudgment and missed judgment. Through the application, the system provides an "independent recognition baseline" for each type of transportation task based on the task clustering structure, and accurately matches the current task behavior characteristics through the personalized threshold set, thereby significantly reducing the false positive rate of atypical tasks, especially suitable for complex transportation scenarios in cold chain logistics with multiple concurrent tasks and multiple product categories. The feature vector and threshold set constructed for each transportation task category are not only used for current task recognition, but also serve as a pre-modeling result for dynamic recognition correction, ensuring the model has structural stability and data traceability, providing a decision boundary basis for subsequent recognition error correction and model self-adaptive evolution.

[0087] In summary, through the feature vector statistics and multi-dimensional personalized threshold modeling mechanism, the application effectively establishes a parameter structure support model in the refrigerator transportation recognition system, making the entire recognition system have strong adaptability, scalability, and high decision accuracy in the "historical modeling-real-time recognition-feedback correction" chain.

[0088] Embodiment 4

[0089] Please refer to Figure 1 , specifically: the second recognition module is used to monitor real-time features in the current transportation task process in real time using a plurality of groups of sensors, wherein the real-time features and the parameter categories collected in the historical time period (for example, the parameter category is the door operation frequency) are the same, and the difference lies in the time difference, one is the parameters collected in the historical time period of the transportation task, and the other is the parameters collected in the current transportation task.

[0090] The real-time features are substituted into the cluster centers of the K groups of new cluster clusters that do not change any more in the clustering unit After the calculation and matching of the Euclidean distance algorithm, the transport label corresponding to the current transport task is judged, and the current transport task category is identified according to the transport label corresponding to the current transport task.

[0091] In the embodiment of the application, first, the current transport process is monitored in real time by a plurality of groups of sensors corresponding to historical stage parameter categories, so that the system can synchronously acquire real-time feature parameters such as door operation frequency, path deviation degree, energy consumption variation, temperature mutation, vibration intensity, which are consistent with historical data structures, ensuring the direct comparability between current data and the clustering model, and laying a consistent foundation for feature dimension for real-time identification. Secondly, the system performs Euclidean distance comparison between the real-time feature vector of the current task and the K groups of stable converged cluster centers, and judges the historical task type closest to the current task according to the principle of minimum distance, thereby identifying the transport label and transport task category corresponding to the transport task. This mechanism has the characteristics of high efficiency, fast response, and clear determination, and is particularly suitable for real-time requirements such as early identification and accurate classification in cold chain transportation. In summary, the application effectively improves the real-time and accuracy of task classification judgment through the structured comparison of real-time data and historical clustering centers, providing key input support for subsequent abnormality judgment and threshold model calling.

[0092] Embodiment 5

[0093] Please refer to Figure 1 Specifically, the correction module includes an extraction unit, an anomaly identification unit, a mechanism triggering unit, and a dynamic correction unit.

[0094] The extraction unit is configured to extract a clustering result identical to the current transport task category from the clustering result set according to the current transport task category identified by the second identification module, and extract a set of personalized identification feature thresholds Q under the current transport task category from the individualized division unit according to the extracted clustering result.

[0095] The anomaly identification unit is configured to analyze the abnormal dimension Fg in the current transport process according to the set of personalized identification feature thresholds Q under the current transport task category obtained by the extraction unit.

[0096] S21: Compare the real-time features in the current transport task process with the feature thresholds set for each feature in the set of personalized identification feature thresholds Q under the current transport task category, respectively, to determine whether each real-time feature exceeds the feature threshold set for the corresponding feature. S21: Compare the real-time features in the current transport task process with the feature thresholds set for each feature in the set of personalized identification feature thresholds Q under the current transport task category, respectively, to determine whether each real-time feature exceeds the feature threshold set for the corresponding feature. ;

[0097] S22: If the corresponding real-time feature exceeds the feature threshold set for the corresponding feature. At any given time, an integral is calculated. After traversing all real-time features, the calculated integral is tallied to determine the anomaly dimension Fg in the current transportation process. The specific method is as follows:

[0098] ;

[0099] In the formula, y represents the feature number within the feature vector W in the feature partitioning unit, and W represents the feature vector W in the feature partitioning unit. This represents the y-th real-time feature collected during the current transportation task. This represents the feature threshold corresponding to the y-th real-time feature under the r-th transport tag. Indicates an indicator function, when Each time, it is counted as one point;

[0100] The mechanism triggering unit is used to pre-set the historical average anomaly dimension based on the personalized identification feature threshold set Q under the corresponding transportation task category conditions in the personalized segmentation unit. The current anomaly dimension Fg in the transportation process exceeds the historical average anomaly dimension. At that time, the corresponding features are determined from the anomaly dimension Fg, and a dynamic correction mechanism is triggered;

[0101] The dynamic correction unit is used to record the final transportation result of the current transportation task after receiving the trigger of the dynamic correction mechanism. It compares the final transportation result with the corresponding feature determined in the anomaly dimension Fg to determine whether the final transportation result is an incorrect judgment. For example, if there is no transportation failure due to abnormal door operation frequency, it will indicate that the current corresponding threshold setting has deviated and correction work is required.

[0102] If it is an incorrect judgment, then combine it with the current corresponding feature threshold. Set the corresponding feature threshold. The dynamic backtracking correction operation is performed, and the specific correction method is as follows:

[0103] ;

[0104] In the formula, This represents the feature threshold corresponding to the y-th real-time feature under the r-th transport tag after correction. This represents the identification error correction factor (false positive penalty value, ranging from 0.1 to 0.5), where T represents the number of historical samples for the corresponding task category, i.e., the number of refrigerated container transportation situations for historical transportation tasks in the corresponding transportation tags, used to control the update range; in the denominator... It is a smoothing factor used to non-linearly control the number of samples. The more samples there are, the smaller the learning rate, preventing overfitting. When T is very small... It is also small, but the overall correction range is large; when T is very large, Increasing the value of the denominator makes the overall correction approach zero, preventing the model from being frequently rewritten and maintaining stability. At the same time, to avoid illegal calculations (such as logarithm 0 or negative infinity) when T=0 or T=1, and to ensure that initial corrections can be made even when there are few task-type samples (even when T=0), adding 2 to the denominator is a smooth starting value, providing protective control for the initial learning process.

[0105] This is to ensure that the threshold correction is proportional to the expansion / contraction, so as to keep the original threshold dimension unchanged and facilitate a unified correction rule across multiple task dimensions;

[0106] The revised number The feature threshold corresponding to the y-th real-time feature under each transport tag The original judgment boundary will be slightly relaxed to enhance the fault tolerance of the next round of judgment.

[0107] Among them, recording the final transportation result of the current transportation task refers to the transportation anomalies that actually occur during the transportation process, including whether any of the characteristics are abnormal.

[0108] If it is not an incorrect judgment, then no corresponding feature threshold is applied. Dynamic backtracking and correction operations.

[0109] In this embodiment of the invention, firstly, the system extracts a set of personalized feature thresholds Q from the clustering results based on the identified current transportation task category, ensuring that anomaly judgment is strictly based on the task type context and improving the targeting of anomaly identification. Secondly, the system calculates the anomaly dimension Fg by comparing real-time features with each feature threshold in Q item by item, achieving integral identification of multi-dimensional abnormal behavior and forming a quantifiable anomaly expression. Combined with a preset historical average anomaly dimension baseline, the system can determine whether a correction mechanism is triggered. Furthermore, when an "incorrect judgment" is identified, the system automatically calls a dynamic backtracking correction formula with a sample size adjustment factor to slightly update the feature thresholds. This correction method can converge when data is sufficient and adapt quickly when data is sparse, thereby preventing the accumulation of misjudgments and model rigidity. In summary, this system constructs a closed-loop logic of "anomaly judgment - judgment error - threshold backtracking," significantly enhancing the adaptability, fault tolerance, and long-term stability of the refrigerated container transportation task identification system in actual operation.

[0110] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the same; although the present application has been described in detail with reference to the foregoing examples, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An automated freezer transport identification system, characterized by: The set generation module, the first identification module, the second identification module and the correction module are included. The set generation module is configured to collect a parameter vector S of a transport task based on a refrigerator transport condition in a historical period, determine a target function L according to a K-means clustering algorithm, and output a clustering result set based on the target function L. The first identification module is configured to construct a personalized identification feature threshold set Q for a task category of each transport task according to the obtained clustering result set. The second identification module is configured to monitor real-time features in a current transport process in real time through a monitoring device, and seek a current transport task category in combination with a Euclidean distance algorithm. The correction module is used to extract the clustering result matched with the current transportation task category from the clustering result set, analyze the abnormal dimension Fg in the current transportation process according to the clustering result matched with the current transportation task category, trigger the dynamic correction mechanism according to the abnormal dimension Fg, and identify the feature threshold of the dynamic backtracking correction operation. The set generation module includes a parameter collection unit, a clustering unit and a category identification unit. The parameter collection unit is used to collect the refrigerator transportation conditions in the historical time period by using a plurality of groups of sensors, and to determine the refrigerator transportation conditions in each transportation task in the historical time period by distinguishing, so as to obtain the parameter vector S of the transportation task through statistics, and the expression of the parameter vector S of the transportation task is: ; wherein, to represents the refrigerator transportation conditions in the process from the first historical transportation task to the nth historical transportation task, and n represents the total number of the historical transportation tasks. The clustering unit is configured to converge the target function L, and the specific steps include: S11: randomly set K clustering clusters to determine each clustering center And combined with the parameter vector S of the transportation task, calculate the Euclidean distance of each transportation task with each clustering center to obtain the objective function L, so as to be assigned to the nearest clustering cluster; S12: and for each clustering cluster, calculate the mean of the refrigerator transportation situation of all historical transportation tasks belonging to the corresponding clustering cluster as the clustering center of the new clustering cluster ; S13: repeatedly performing the step S11 and the step S12 until the target function L converges, i.e., the grouping of the historical transport tasks no longer changes.

2. The automatic refrigerator transport identification system according to claim 1, wherein: The category recognition unit is configured to determine the transport label of each historical transport task from the target function L when the grouping of the historical transport tasks is no longer changed, and store the corresponding transport label into an automated cold cabinet transport recognition background. Through statistics, a clustering result set P is determined and constructed, and an expression of the clustering result set P is: ; wherein, to denotes the first transport label p to the mth transport label p, and m denotes the total number of transport labels; and the transport task category of the corresponding historical transport task is determined according to the transport label corresponding to the corresponding historical transport task.

3. The automatic refrigerator transport identification system according to claim 2, wherein: The first identification module includes a feature division unit and a personalized division unit. The feature division unit is configured to divide the refrigerator transportation situation of the historical transportation task corresponding to each transportation task category based on the clustering result set P obtained in the category identification unit and the identified transportation task category, and generate a feature vector W corresponding to each transportation task category through statistics. The expression of the feature vector W is as follows: ; to represent the number set w of the first feature to the number set w of the xth feature under the corresponding transportation task category. x represents the total number of features.

4. The automatic refrigerator transport identification system according to claim 3, wherein: The personalized division unit is configured to set a corresponding feature threshold for each feature under the corresponding transportation task category according to the feature vector W corresponding to each transportation task category obtained in the feature division unit , to generate a personalized identification feature threshold set Q under the corresponding transportation task category, and the expression of the personalized identification feature threshold set Q is: ; to , which represents the feature threshold set for the first feature under the corresponding transportation task category to the feature threshold set for the xth feature .

5. The automatic refrigerator transport identification system according to claim 4, wherein: The second identification module is configured to monitor real-time features in a current transport task process in real time again by using a plurality of groups of sensors, wherein the real-time features are the same as the collected parameter categories in the historical period, and the difference lies in the time difference.

6. The automatic refrigerator transport identification system according to claim 5, wherein: The second identification module is further configured to substitute the real-time features into the cluster centers of the K groups of new cluster clusters in the clustering unit that are no longer changed After calculation and matching by the Euclidean distance algorithm, the current transport task corresponding transport label is determined, and the current transport task category is identified according to the current transport task corresponding transport label.

7. The automatic refrigerator transport identification system according to claim 6, wherein: The correction module includes an extraction unit, an abnormality identification unit, a mechanism triggering unit and a dynamic correction unit. The extraction unit is configured to extract a clustering result that is the same as the current transport task category from the clustering result set according to the current transport task category identified by the second identification module, and extract a personalized identification feature threshold set Q under the current transport task category from the personalized division unit according to the extracted clustering result. The abnormality identification unit is configured to analyze an abnormal dimension Fg in the current transport process according to the personalized identification feature threshold set Q under the current transport task category obtained by the extraction unit, and the specific analysis steps of the abnormal dimension Fg are as follows: S21: compare the real-time features in the current transportation task process with the feature threshold set by each feature in the personalized identification feature threshold set Q under the current transportation task category condition, respectively, to determine whether each real-time feature exceeds the feature threshold set by the corresponding feature ;​ S22: If the corresponding real-time feature exceeds the feature threshold set for the corresponding feature. Each time, an integral is added. After traversing all real-time features, the added integrals are counted to determine the anomaly dimension Fg in the current transportation process.

8. The automatic refrigerator transport identification system according to claim 7, wherein: The mechanism triggering unit is used to set the historical average abnormal dimension Fg according to the individualized identification feature threshold set Q under the corresponding transportation task category in the individualized division unit in advance , and determine the corresponding features from the abnormal dimension Fg when the abnormal dimension Fg in the current transportation process exceeds the historical average abnormal dimension , and trigger the dynamic correction mechanism. The dynamic correction unit is used to record the final transportation result of the current transportation task after receiving the trigger of the dynamic correction mechanism, compare the final transportation result with the corresponding feature determined in the abnormal dimension Fg, judge whether the final transportation result is a false judgment, if it is a false judgment, combine the current corresponding feature threshold , make dynamic backtracking correction work of corresponding feature threshold ; if it is not a false judgment, do not make dynamic backtracking correction work of corresponding feature threshold .

Images