An automated system for artificial intelligence models based on self-learning

Abstract

This invention relates to the field of data processing technology, and more particularly to an automated system based on a self-learning artificial intelligence model. The system includes: an acquisition module, a judgment module, a focus determination module, a correction module, an anomaly determination module, an adjustment module, and an early warning module. This invention acquires key data from the sorting area in real time; determines temporary areas based on spatial accumulation area; then determines focus areas by combining queue growth rate and sorting rate; next, obtains correction areas based on the location and spatial turnover rate of the focus areas; determines anomaly areas based on the artificial intelligence model and the correction areas; then dynamically optimizes thresholds using sorting accuracy as feedback; and finally issues early warnings for high-risk areas. This effectively solves the problems of low accuracy and slow response speed in identifying high-risk areas due to insufficient iteration and generalization capabilities, which prevent adaptation to complex and changing working environments, and the rigidity of automation mechanisms.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an automated system based on a self-learning artificial intelligence model. Background Technology

[0002] With the rapid development of the logistics industry, automated sorting systems in logistics centers play a crucial role in improving efficiency and reducing costs. However, most existing automated sorting systems rely on pre-set static thresholds and rules for monitoring and decision-making, a model that proves inadequate in the face of complex dynamic environments. Specifically, existing systems suffer from blind spots due to their reliance on static thresholds, failing to dynamically predict potential bottlenecks and proactively prevent congestion. Furthermore, their single-point monitoring model struggles to address sorting congestion and anomalies caused by multiple factors in modern sorting operations, resulting in delayed warnings and passive decision-making, failing to fundamentally solve the problem. Therefore, there is an urgent need for a self-learning and adaptive intelligent decision-making system to overcome the limitations of existing systems and achieve comprehensive monitoring, accurate prediction, and efficient management of the logistics sorting process.

[0003] Chinese Patent Application Publication No. CN117114524A discloses a logistics sorting method based on reinforcement learning and digital twins. The method includes: collecting historical cargo data from the logistics sorting system; collecting historical sorting data from the sorting slots of the sorting machine in the logistics sorting system and fitting a slot processing efficiency function; integrating tag information through a clustering algorithm to obtain a tag category similarity matrix and a transition probability matrix; designing a reinforcement learning strategy and a value network based on the tag category similarity and transition probability matrix, and constructing the leaf nodes of a Monte Carlo tree; expanding the leaf nodes of the Monte Carlo tree to obtain the optimal slot sorting strategy; for logistics sorting systems in different logistics transfer centers, constructing a digital twin, simulating historical cargo data and historical sorting data in the digital twin, and dynamically adjusting the Monte Carlo tree to obtain the optimal slot sorting strategy for the current logistics sorting system in the logistics transfer center.

[0004] Therefore, the logistics sorting method based on reinforcement learning and digital twins has the following problems: it requires a large amount of historical cargo data and sorting data; low data quality or insufficient data volume will lead to inaccurate fitting of the grid processing efficiency function, thus affecting subsequent strategy optimization; high computational complexity can easily lead to long system response time, making it difficult to adjust the sorting strategy in real time, especially in large-scale logistics sorting systems; it performs well on training data, but the situation is complex in actual applications, and the generalization ability is insufficient, resulting in inefficiency or incorrect sorting in the actual sorting process; and it cannot reflect the status of the actual logistics sorting system in real time. Summary of the Invention

[0005] To address this, the present invention provides an automated system based on a self-learning artificial intelligence model, which overcomes the problems of insufficient iteration and generalization capabilities in existing technologies, leading to an inability to adapt to complex and ever-changing working environments, and the low accuracy and slow response speed in identifying high-risk areas due to the rigidity of automation mechanisms. This is achieved through multi-dimensional data analysis and dynamic adjustment mechanisms.

[0006] To achieve the above objectives, the present invention provides an automated system for self-learning artificial intelligence models, comprising:

[0007] The acquisition module is used to obtain the spatial stacking area, queue growth rate, sorting rate, space turnover rate, and sorting accuracy rate of each sorting area to be tested in the logistics center within the past preset sorting cycle.

[0008] The determination module is used to determine several temporary areas based on the spatial accumulation area and a preset area threshold.

[0009] The attention determination module is used to determine several attention areas based on the queue growth rate, the sorting rate, and a preset collaboration threshold of each of the temporary areas.

[0010] The correction module is used to correct the location of each of the interest areas, the spatial turnover rate, and the spatial turnover rate of each of the sorting areas to be tested, so as to obtain a number of correction areas.

[0011] An anomaly determination module is used to determine several anomaly regions based on a preset artificial intelligence model, the queue growth rate of each of the sorting regions to be tested, and each of the correction regions.

[0012] An adjustment module is used to adjust the preset area threshold or the preset coordination threshold according to the sorting accuracy of each abnormal area within a preset adjustment period.

[0013] The early warning module is used to issue early warnings for several high-risk areas determined based on the spatial accumulation area, spatial turnover rate, and sorting accuracy of each of the abnormal areas, which are obtained again based on the adjusted preset area threshold or the preset collaborative threshold.

[0014] Furthermore, the attention determination module includes:

[0015] Focus on the fluctuation calculation unit, which is used to calculate the growth fluctuation value based on all the queue growth rates within a preset time period, and obtain several growth fluctuation values; and calculate the sorting fluctuation value based on all the sorting rates within a preset time period, and obtain several sorting fluctuation values.

[0016] A focus determination unit, which is connected to the focus fluctuation calculation unit, is used to determine several focus areas based on all the growth fluctuation values ​​and all the sorting fluctuation values.

[0017] Furthermore, the attention determination unit includes:

[0018] The coordination degree calculation subunit is used to calculate the change coordination degree based on all the growth fluctuation values ​​and all the sorting fluctuation values;

[0019] A focus determination subunit is connected to the coordination degree calculation subunit to determine the temporary region as the focus region based on the comparison result of the change coordination degree and the preset coordination threshold, so as to determine several focus regions.

[0020] Furthermore, the correction module includes:

[0021] The dispersion calculation unit is used to calculate the dispersion based on the location coordinates of each of the regions of interest.

[0022] A correction unit, connected to the dispersion calculation unit, is used to make corrections based on the comparison results of the dispersion and the preset dispersion threshold, according to the spatial turnover rate of each of the areas of interest and the spatial turnover rate of each of the sorting areas to be tested, to obtain several correction areas.

[0023] Furthermore, the correction unit includes:

[0024] The test cluster is determined by a sub-unit, which is used to cluster according to the location of each region of interest to obtain several test clusters;

[0025] A turnover fluctuation calculation subunit, which is connected to the test cluster determination subunit, is used to calculate the average turnover fluctuation of the focus based on the total spatial turnover rate of each focus area in the test cluster within a preset correction time, and to calculate the test turnover fluctuation value based on the spatial turnover rate of each test sorting area in the same test cluster within a preset correction time.

[0026] A correction subunit, connected to the turnover fluctuation calculation subunit, is used to calculate the turnover fluctuation deviation based on the turnover fluctuation value to be measured and the average turnover fluctuation value of concern, and to determine the sorting area to be measured and all the concern areas as correction areas based on a first comparison result of the turnover fluctuation deviation and a preset turnover fluctuation deviation threshold, and to determine all the concern areas as correction areas based on a second comparison result of the turnover fluctuation deviation and the preset turnover fluctuation deviation threshold.

[0027] Furthermore, the anomaly determination module includes:

[0028] The key area determination unit is used to determine several key areas based on the preset artificial intelligence model and the queue growth rate of each of the sorting areas to be tested.

[0029] A ratio calculation unit, connected to the key point determination unit, is used to calculate the ratio of the number of key areas to the number of correction areas to obtain a quantity ratio.

[0030] An overlap calculation unit, connected to the ratio calculation unit, is used to count the number of overlapping areas in all the key areas and all the correction areas when the quantity ratio is within a preset ratio range, to obtain the overlap quantity; when the number of key areas is greater than the number of correction areas, to calculate the ratio of the overlap quantity to the number of key areas, to obtain the overlap degree; or, when the number of key areas is less than the number of correction areas, to calculate the ratio of the overlap quantity to the number of correction areas, to obtain the overlap degree.

[0031] An anomaly determination unit, which is connected to the overlap calculation unit, is used to determine several of the anomaly regions based on the overlap degree.

[0032] Furthermore, the anomaly determination unit includes:

[0033] The overlap fluctuation calculation subunit is used to calculate the overlap fluctuation value based on all the overlap within a preset period.

[0034] An anomaly determination subunit, connected to the overlap fluctuation calculation subunit, is used to determine all the correction regions as the abnormal regions based on a first comparison result of the overlap fluctuation value and a preset overlap fluctuation threshold, and to determine all the key regions and all the correction regions as abnormal regions based on a second comparison result of the overlap fluctuation value and the preset overlap fluctuation threshold, thereby identifying several abnormal regions.

[0035] Furthermore, the adjustment module includes:

[0036] An accuracy fluctuation calculation unit is used to calculate an accuracy fluctuation value based on the total sorting accuracy of all the abnormal areas.

[0037] An area adjustment unit, which is connected to the accuracy fluctuation calculation unit, is used to adjust the preset area threshold according to the first comparison result of the accuracy fluctuation value and the preset accuracy fluctuation threshold.

[0038] An accuracy change rate calculation unit, which is connected to the accuracy fluctuation calculation unit, is used to calculate the accuracy change rate based on the second comparison result of the accuracy fluctuation value and the preset accuracy fluctuation threshold, according to the average accuracy of all the abnormal regions within the preset adjustment time.

[0039] A collaborative adjustment unit, connected to the accuracy change rate calculation unit, is used to adjust the preset collaborative threshold based on the comparison result between the accuracy change rate and the preset change rate threshold.

[0040] Furthermore, the early warning module includes:

[0041] The risk index calculation unit is used to calculate the risk index based on the space stacking area, the space turnover rate, and the sorting accuracy rate.

[0042] A high-risk determination unit, which is connected to the risk index calculation unit, is used to determine the abnormal area as the high-risk area based on the comparison result between the risk index and the preset risk index threshold, thereby obtaining several high-risk areas;

[0043] An early warning unit, which is connected to the high-risk determination unit, is used to issue early warnings for all the high-risk areas.

[0044] Furthermore, the temporary area is determined based on the comparison result between the spatial accumulation area and the preset area threshold.

[0045] Compared with existing technologies, the advantages of this invention lie in its ability to acquire real-time data on the spatial accumulation area, queue growth rate, sorting rate, space turnover rate, and sorting accuracy of each sorting area in a logistics center; to determine temporary areas based on the spatial accumulation area, which is a direct indicator of congestion; then, to identify areas of concern by combining the queue growth rate and sorting rate of these temporary areas. The synergy between the fluctuations in queue growth and sorting rate reveals the inherent stable state of the system, while a lack of coordination indicates that the system is heading towards imbalance; finally, adjustments are made based on the location and space turnover rate of the areas of concern to obtain corrected areas. This aims to identify bottleneck transmission and systemic risks caused by spatial correlation. The system does not view problems in isolation; it identifies key areas based on artificial intelligence models and queue growth rates, and comprehensively corrects these areas to identify abnormal areas, enabling correction of basic AI recognition and detection under business fluctuations. Then, through a self-learning mechanism that uses actual sorting accuracy as feedback, it dynamically optimizes the judgment threshold, allowing the system to transcend fixed rules and adapt to business fluctuations. Finally, it issues early warnings for high-risk areas, achieving accurate early warning and continuous optimization of logistics sorting efficiency and stability. This effectively solves the problems of low accuracy and slow response speed in identifying high-risk areas due to insufficient iteration and generalization capabilities, which prevent the system from adapting to complex and ever-changing work environments, and the rigidity of automation mechanisms.

[0046] Furthermore, by adopting a capacity-based coarse screening strategy, the most intuitive and stable physical indicator of spatial accumulation area is used as the primary signal of system load. By quickly comparing it with its critical pre-set area threshold, a one-time screening of all visible congestion areas is achieved with high efficiency and low computational cost. It is acknowledged that physical space is an insurmountable hard constraint for sorting operations. When the accumulation area exceeds the threshold, regardless of its dynamic performance, the area has reached its carrying capacity limit and there is a clear risk of operational difficulties and efficiency decline, thus it must be included in the scope of subsequent in-depth analysis.

[0047] Furthermore, by introducing the judgment logic of dynamic stability diagnosis, risk identification is elevated from static threshold judgment to the level of system behavior analysis. The real risk of a region lies not only in its instantaneous load, but also in the loss of its control capability. The volatility of the two key processes, queue growth and sorting rate, is amplified together, indicating that the system is losing balance and entering an unstable state. The volatility values ​​of the two in time series are calculated separately to capture the degree of disorder in the dynamic process. High-risk areas are the concentrated manifestation of this disorder. When the volatility intensifies synergistically, it indicates that the region is sliding from controllable operation to the edge of being out of control, thus being accurately located as an object that needs high attention, providing a set of targets with real potential failure tendencies for subsequent analysis.

[0048] Furthermore, by calculating the cosine similarity between the fluctuation values ​​of queue growth and sorting rate, the degree of synergy between the input and output perspectives is reflected. When the fluctuation patterns of the two are highly similar, i.e., the degree of synergy is high, it means that any input disturbance will be directly transmitted to the output end, the system loses its buffering and adjustment capabilities, and is in a critical state of resonance instability. First, the interference of absolute dimensions is eliminated by normalization to focus on the fluctuation pattern, and then the similarity of the two fluctuation sequences in direction is accurately quantified by cosine similarity to capture the inherent runaway trend of the system. When it exceeds the preset threshold, even if a single indicator does not reach the limit, the area is determined to be a high-risk area, thus realizing the early and accurate identification of potential operational crises.

[0049] Furthermore, by introducing the judgment logic of spatial correlation risk diagnosis, a leap from identifying isolated problem points to early warning of systemic regional risks has been achieved. When multiple areas of concern are highly clustered in space, their risks are not simply superimposed, but will generate a resonance amplification effect through the continuity of the work process, forming a bottleneck cluster that paralyzes the entire sorting subsystem. First, the spatial clustering degree is quantified by calculating the dispersion of location coordinates. When the dispersion is less than the preset dispersion threshold, that is, when the areas of concern are highly concentrated, the spatial turnover rate, a core indicator that characterizes the efficiency of regional collaborative operation, is introduced for correlation analysis. By comparing the turnover rate fluctuations of the problem area and the surrounding areas, the potential risk points that are affected by the cascading effect are accurately identified, thereby correcting the originally isolated areas of concern into a set of areas that better reflect the actual scope of the failure, realizing the early identification and precise intervention of systemic bottlenecks.

[0050] Furthermore, by clustering, spatial location associations are transformed into logical clusters to be tested. Physically adjacent areas form a community of shared destiny due to the continuity of the work process, and their performance fluctuations are transmissible. By capturing these potential risk units, the dynamic fluctuation of spatial turnover rate is used as a quantitative indicator of performance association. When the turnover fluctuation of the sorting area under test within the cluster is highly consistent with the average fluctuation of the core problem areas in the area of ​​concern, that is, the deviation is less than the preset turnover fluctuation deviation threshold, it indicates that they are in the same unstable system, and the risk has spread throughout the entire cluster, so the entire cluster needs to be corrected together. If the deviation is greater than the preset turnover fluctuation deviation threshold, it indicates that the risk is still limited to the original problem point, realizing the upgrade from point-based early warning to cluster-based risk prevention and control, and accurately depicting the true impact range of the bottleneck effect.

[0051] Furthermore, through dual-source verification of prediction and diagnosis, intelligent trade-offs and precise decision-making in anomaly area identification are achieved. The key areas predicted by AI from a single source or the correction areas diagnosed by the system both have uncertainties. However, the quantitative relationship and spatial overlap between the two reflect the degree of consensus on the system state. When the two are equal, overlap analysis is initiated, reflecting the principle of equal trade-offs. Using the minimum value as the denominator to calculate the overlap ensures that the indicator always reflects the proportion of the overlapping part in a smaller set, thereby keenly capturing the core consensus area. The comparability of the two-source data is quickly judged by the quantitative ratio, and the consistency of the two in spatial judgment is accurately quantified by the geometric overlap. Finally, based on the consensus strength, the truly abnormal areas that require intervention measures are intelligently determined, significantly improving the accuracy and reliability of the early warning results.

[0052] Furthermore, the reliability of the consensus of the entire judgment system is measured by the historical volatility of the overlap between the two judgment sources. When the volatility is high, it indicates that the prediction and diagnosis conclusions are inconsistent for a long time, and the system is in a state of cognitive divergence. In this case, it is more prudent and reliable to select the correction area based on real-time systematic diagnosis as the final abnormal area. When the volatility is low, it indicates that the two judgment sources are consistent for a long time, and the system is in a stable state of cognitive consensus. In this case, it is certain that the judgments of the two sources are combined to expand the warning range. Time series analysis is introduced into the decision arbitration process, and the consistency is judged by the overlap volatility value to dynamically select the most reliable anomaly judgment strategy, thereby maintaining the high credibility of the warning results in a complex and ever-changing production environment.

[0053] Furthermore, by constructing a two-layer threshold adaptive mechanism based on performance feedback, the system's judgment criteria and dynamic operating environment are precisely matched. The sorting accuracy rate, as the final performance indicator, is used as the gold standard to verify the effectiveness of the threshold. Precise adjustments are made for two typical failure modes. When the accuracy rate fluctuates drastically, it indicates that the current area threshold is too sensitive, causing the system to frequently misjudge between normal fluctuations and real risks. Therefore, the preset area threshold needs to be increased to improve system stability. When the accuracy rate fluctuates steadily but continues to deteriorate, it indicates that there are unidentified hidden bottlenecks in the system. The problem lies in the dynamic instability diagnosis stage. Therefore, the collaborative threshold needs to be increased to capture more subtle signs of loss of control. By using two different indicators, the accuracy rate fluctuation value and the accuracy rate change rate, the system can intelligently distinguish the types of interference faced by the system, forming a closed-loop control strategy that can self-optimize according to changes in the operating environment.

[0054] Furthermore, by constructing a multi-dimensional risk fusion decision-making model, accurate mapping and efficient decision-making from multi-source heterogeneous data to a unified risk level are achieved. The operational risk of the sorting area is a comprehensive result of the combined effects of static load, dynamic efficiency, and final quality. Any single indicator can only reflect one aspect of the risk. First, the influence of different dimensions is eliminated through max-min normalization, and then a comprehensive and quantitative risk index is generated through weighted summation with preset weights. This ensures that the system can accurately identify those high-risk areas that perform poorly in all three dimensions of load, efficiency, and quality from the abnormal areas selected through the aforementioned complex process, based on unified and objective standards, and that most urgently require intervention. This allows for the issuance of accurate and effective early warnings, realizing the intelligent transformation from complex information to concise decision-making. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the self-learning artificial intelligence model automation system in this embodiment;

[0056] Figure 2 This is a logic diagram for determining the temporary region in the determination module of this embodiment;

[0057] Figure 3 This is a logic diagram for determining the region of interest by the attention determination unit in this embodiment;

[0058] Figure 4 This is a logic diagram for determining the correction region by the correction unit in this embodiment. Detailed Implementation

[0059] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0060] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0061] Please see Figure 1 As shown, this is a schematic diagram of the self-learning-based artificial intelligence model automation system of this embodiment. This embodiment provides a self-learning-based artificial intelligence model automation system, including:

[0062] The acquisition module is used to obtain the spatial stacking area, queue growth rate, sorting rate, space turnover rate, and sorting accuracy rate of each sorting area to be tested in the logistics center within the past preset sorting cycle.

[0063] A determination module, which is connected to the acquisition module, is used to determine several temporary areas based on the spatial accumulation area and a preset area threshold.

[0064] The attention determination module is connected to the acquisition module and the determination module respectively, and is used to determine several attention areas based on the queue growth rate, the sorting rate and the preset collaborative threshold of each temporary area.

[0065] A correction module, which is connected to the acquisition module and the attention determination module respectively, is used to correct according to the position of each attention area, the space turnover rate and the space turnover rate of each sorting area to be tested, to obtain a number of correction areas.

[0066] An anomaly determination module, which is connected to the acquisition module and the correction module respectively, is used to determine a number of abnormal areas based on a preset artificial intelligence model, the queue growth rate of each of the sorting areas to be tested, and each of the key areas and each of the correction areas.

[0067] An adjustment module, which is connected to the acquisition module and the anomaly determination module respectively, is used to adjust the preset area threshold or the preset coordination threshold according to the sorting accuracy of each of the abnormal areas within a preset adjustment time.

[0068] The early warning module is connected to the acquisition module and the anomaly determination module respectively, and is used to issue early warnings for a number of high-risk areas determined based on the spatial accumulation area, spatial turnover rate and sorting accuracy of each of the abnormal areas obtained again based on the adjusted preset area threshold or the preset collaborative threshold.

[0069] In this embodiment, during peak business hours at the logistics center, the parcel volume is several times that of normal days, and the entire sorting network is under constant pressure. By acquiring key parameters of each sorting area during the sorting process, areas with congestion or malfunctions can be identified in a timely manner, and warnings can be sent to staff. This helps staff to grasp the main contradictions and take solutions in complex situations, ensuring the resilience of the logistics center under extreme pressure and improving sorting efficiency.

[0070] In this embodiment, sorting accuracy refers to the percentage of packages correctly sorted to their corresponding destinations within a preset sorting cycle out of the total number of packages sorted. This accuracy is collected by RFID readers located at the end of the sorting line or at the entrances of the chutes leading to each destination. The preset sorting cycle refers to the length of time used to statistically evaluate sorting accuracy. It depends on the volatility of business volume, the requirement for timely feedback, and the statistical significance of the data, and is typically set between 10 and 20 minutes. In this embodiment, it is set to 15 minutes, which allows for timely detection of problems, reflects operational trends, and provides the system with stable and reliable learning signals.

[0071] In this embodiment, the spatial accumulation area is the effective planar area actually occupied by packages within the sorting area, reflecting the congestion level of the sorting area. It is obtained by separating the cargo area from the background in the image using an image semantic segmentation algorithm and calculating the cargo area area. The queue growth rate refers to the rate at which the number of goods in the sorting queue increases per unit time, reflecting the backlog of sorting tasks. It is obtained by identifying the number of goods in the queue in real-time images and calculating its rate of change over time. The sorting rate refers to the number of goods processed by the sorting system per unit time, reflecting the efficiency of the sorting system. It is obtained by using a target detection algorithm to identify goods during the sorting process and then calculating the sorting rate through time series analysis. The space turnover rate refers to the speed at which goods move within the sorting area per unit time, reflecting the space utilization efficiency of the sorting area. It is obtained by analyzing real-time images, identifying the movement trajectory of goods, and calculating their dwell time and turnover speed within the sorting area.

[0072] In this embodiment, the preset artificial intelligence model is a pre-trained Long Short-Term Memory (LSTM) network model, used to accurately predict key areas with congestion risk in the logistics sorting system based on the queue growth rate. In this embodiment, the initial parameters of the model use the queue growth rate and timestamps as input features. After receiving data through the input layer, multiple LSTM units in the hidden layer capture time-series features. The input gate, forget gate, and output gate of each unit effectively handle long-term dependency issues. Finally, the output layer generates key area identifiers. A learning rate of 0.001, a batch size of 32, and 100 training epochs are used to lay the foundation for model training.

[0073] In this embodiment, the data preparation stage before model training collects historical logistics sorting data, covering queue growth rates and corresponding key area markers. The latter can be determined through expert evaluation or calculated based on historical data. In the data preprocessing stage, the data is normalized to scale it to a range of 0 to 1 to accelerate model convergence. During model training, the training data drives the LSTM model learning, and optimization algorithms such as Adam are used to iteratively update model parameters. Mean squared error (MSE) is used as the loss function to minimize the deviation between predicted and actual values. The validation set is used to fine-tune hyperparameters, and an early stopping mechanism is used to monitor model performance on the validation set in real time to prevent overfitting. Simultaneously, the test set comprehensively evaluates model performance, using metrics such as MSE, accuracy, and F1 score to measure its generalization ability, ensuring excellent prediction accuracy and reliability in complex and ever-changing real-world scenarios. Once the model training is complete, its parameters and structure will be carefully saved so that it can be seamlessly applied to actual logistics sorting scenarios. The model's performance metrics on the training set, validation set, and test set will be recorded in detail, including mean squared error, accuracy, and F1 score, to evaluate the model's accuracy and reliability.

[0074] The preset area threshold is a standard value used to determine whether there is a risk of congestion in the sorting area under test. It depends on the total area of ​​the sorting area, the average size of the logistics packages, the frequency and urgency of sorting tasks, and the fault tolerance of the system. It is usually set between 10% and 30% of the total area of ​​the sorting area. In this embodiment, it is set to 20% of the total area of ​​the sorting area, which can effectively balance sorting efficiency and space utilization, avoid the decline in sorting efficiency due to excessive stacking area, and promptly identify the risk of congestion in the sorting area.

[0075] The preset coordination threshold is a standard value used to determine whether the disorder of a temporary area is sufficiently significant. It depends on the system's fault tolerance, the complexity of the sorting task, and its sensitivity to anomalies, and is typically set between 0.7 and 0.9. In this embodiment, it is set to 0.8, which effectively distinguishes between normal and abnormal fluctuations, ensuring that an area is only identified as an area of ​​interest when the queue growth and sorting rate fluctuations are highly coordinated. This improves the system's accuracy and reliability and reduces false alarms.

[0076] The preset adjustment time is the length of time used to calculate the rate of change of the average accuracy of abnormal areas. It depends on the sorting task cycle of the logistics center, the parcel processing frequency, and the system's response speed requirements, and is usually set between 10 and 30 minutes. In this embodiment, it is set to 20 minutes, which can capture changes in sorting accuracy in a timely manner, while avoiding frequent misjudgments caused by too short a time, thus improving the stability and reliability of the system.

[0077] By acquiring real-time data on the spatial accumulation area, queue growth rate, sorting rate, space turnover rate, and sorting accuracy of each sorting area in the logistics center, temporary areas are identified based on the spatial accumulation area, which is a direct indicator of congestion. Then, the queue growth rate and sorting rate of these temporary areas are combined to determine areas of concern. The synergy between the fluctuations in queue growth and sorting rate reveals the system's inherent stable state; a lack of coordination between the two indicates that the system is heading towards imbalance. Finally, adjustments are made based on the location and space turnover rate of these areas of concern to obtain corrected areas. This approach aims to identify bottleneck transmission and systemic risks caused by spatial correlations, rather than viewing the problem in isolation. By using artificial intelligence models and queue growth rates to identify key areas and comprehensively correcting these areas to identify abnormal areas, the system can correct basic AI recognition and detection under business fluctuations. Then, through a self-learning mechanism that uses actual sorting accuracy as feedback, the system dynamically optimizes the judgment threshold, enabling it to transcend fixed rules and adapt to business fluctuations. Finally, it issues warnings for high-risk areas, achieving accurate early warning and continuous optimization of logistics sorting efficiency and stability. This effectively solves the problems of low accuracy and slow response speed in identifying high-risk areas due to insufficient iteration and generalization capabilities, which prevent the system from adapting to complex and ever-changing work environments, and the rigidity of automation mechanisms.

[0078] Please see Figure 2 As shown, it is the determination logic diagram of the determination module for temporary areas in this embodiment. In this embodiment, when the spatial accumulation area is greater than the preset area threshold, the area to be tested for sorting is determined to be the temporary area, and several temporary areas are obtained.

[0079] By adopting a capacity-based coarse screening strategy, the most intuitive and stable physical indicator of spatial accumulation area is used as the primary signal of system load. Through rapid comparison with its critical pre-set area threshold, all obviously congested areas are screened efficiently and with low computational cost. It is acknowledged that physical space is an insurmountable hard constraint for sorting operations. When the accumulation area exceeds the threshold, regardless of its dynamic performance, the area has reached its carrying capacity limit and there is a clear risk of operational difficulties and efficiency decline. Therefore, it must be included in the scope of subsequent in-depth analysis.

[0080] Specifically, the attention determination module includes:

[0081] Focus on the fluctuation calculation unit, which is used to calculate the standard deviation of all queue growth rates from the initial time to each time within a preset determined time period, to obtain several growth fluctuation values, and to obtain several sorting fluctuation values ​​based on the standard deviation calculated from all sorting rates from the initial time to each time within a preset determined time period.

[0082] A focus determination unit, which is connected to the focus fluctuation calculation unit, is used to determine several focus areas based on all the growth fluctuation values ​​and all the sorting fluctuation values.

[0083] The preset time limit is the length of time used for analyzing data when selecting and determining the area of ​​interest. It depends on the sorting task cycle, package processing frequency, and system response speed requirements of the logistics center, and is usually set between 10 and 30 minutes. In this embodiment, it is set to 15 minutes, which can capture queue growth and sorting rate fluctuations in a timely manner, while avoiding frequent misjudgments caused by too short a time, thus improving the stability and reliability of the system.

[0084] By introducing dynamic stability diagnostics, risk identification is elevated from static threshold judgment to system behavior analysis. The real risk of a region lies not only in its instantaneous load but also in its loss of control. The volatility of the two key processes, queue growth and sorting rate, is amplified together, indicating that the system is losing balance and entering an unstable state. The temporal volatility values ​​of these two processes are calculated separately to capture the degree of disorder in the dynamic process. High-risk areas are the concentrated manifestation of this disorder. When volatility intensifies synergistically, it indicates that the region is sliding from controllable operation to the edge of being out of control, thus accurately locating it as an object requiring high attention and providing a set of targets with real potential failure tendencies for subsequent analysis.

[0085] Please see Figure 3 As shown, this is a logic diagram for determining the region of interest by the attention determination unit in this embodiment. In this embodiment, the attention determination unit includes:

[0086] The coordination degree calculation subunit is used to perform maximum-min normalization on all the growth fluctuation values ​​to obtain several growth fluctuation normalization values, and to perform maximum-min normalization on all the sorting fluctuation values ​​to obtain several sorting fluctuation normalization values, and to calculate the cosine similarity between all the growth fluctuation normalization values ​​and all the sorting fluctuation normalization values ​​to obtain the change coordination degree.

[0087] A focus determination subunit, which is connected to the coordination degree calculation subunit, is used to determine the temporary region as the focus region when the changed coordination degree is greater than the preset coordination threshold, so as to determine several focus regions.

[0088] By calculating the cosine similarity between the fluctuation values ​​of queue growth and sorting rate, the degree of synergy between the input and output perspectives is reflected. When the fluctuation patterns of the two are highly similar, i.e., the degree of synergy is high, it means that any input disturbance will be directly transmitted to the output end. The system loses its buffering and adjustment capabilities and is in a critical state of resonance instability. First, the interference of absolute dimensions is eliminated by normalization to focus on the fluctuation pattern. Then, the cosine similarity is used to accurately quantify the similarity of the two fluctuation sequences in direction, capturing the inherent runaway trend of the system. When it exceeds the preset threshold, even if a single indicator does not reach the limit, the area is determined to be a high-risk area, realizing the early and accurate identification of potential operational crises.

[0089] Specifically, the correction module includes:

[0090] The dispersion calculation unit is used to obtain the Euclidean distance between the position coordinates of each region of interest and the coordinates of a preset reference point, obtain several reference distances, and calculate the standard deviation of all reference distances to obtain the dispersion.

[0091] A correction unit, connected to the dispersion calculation unit, is used to correct the dispersion based on the spatial turnover rate of each of the areas of interest and the spatial turnover rate of each of the sorting areas to be tested when the dispersion is less than a preset dispersion threshold, thereby obtaining several correction areas.

[0092] The preset reference point is a fixed coordinate point used to evaluate the spatial distribution dispersion of the area of ​​interest. It depends on the layout of the sorting area and the central location of the logistics operation. It is usually set at the geometric center of the sorting area or near the main operation point. In this embodiment, it is set at the geometric center of the sorting area, which can more intuitively evaluate the distribution of the area of ​​interest, facilitate the calculation of dispersion, and improve the overall coordination and management efficiency of the system.

[0093] The preset dispersion threshold is a standard value used to determine whether the location distribution of the areas of interest is concentrated. It depends on the layout of the sorting area, the distribution characteristics of logistics packages, and the system's space management strategy, and is usually set between 0.3 and 0.7. In this embodiment, it is set to 0.5, which can effectively balance space utilization and operational efficiency, avoid local congestion caused by excessive concentration of areas of interest, and ensure that the corrected area distribution is more reasonable, thereby improving sorting efficiency and the overall performance of the system.

[0094] By introducing the judgment logic of spatial correlation risk diagnosis, a leap from identifying isolated problem points to early warning of systemic regional risks has been achieved. When multiple areas of concern are highly clustered in space, their risks are not simply superimposed, but will generate a resonance amplification effect through the continuity of the operation process, forming a bottleneck cluster that paralyzes the entire sorting subsystem. First, the spatial clustering degree is quantified by calculating the dispersion of location coordinates. When the dispersion is less than the preset dispersion threshold, that is, when the areas of concern are highly concentrated, the spatial turnover rate, a core indicator that characterizes the efficiency of regional collaborative operation, is introduced for correlation analysis. By comparing the turnover rate fluctuations of the problem area and the surrounding areas, the potential risk points that are affected by the cascading effect are accurately identified. Thus, the originally isolated areas of concern are corrected into a set of areas that better reflect the actual scope of the failure, realizing the early identification and precise intervention of systemic bottlenecks.

[0095] Please see Figure 4 As shown, this is a logic diagram for determining the correction region by the correction unit in this embodiment. In this embodiment, the correction unit includes:

[0096] The test cluster is determined by a sub-unit, which is used to cluster according to the location of each region of interest to obtain several test clusters;

[0097] A turnover fluctuation calculation subunit, which is connected to the test cluster determination subunit, is used to calculate the average value of all the spatial turnover rates of all the interest turnover fluctuation values ​​of all the interest areas in the test cluster from the initial time to each time within a preset correction time, to obtain the average value of interest turnover fluctuation, and to calculate the test turnover fluctuation value by calculating all the spatial turnover rates of all the test sorting areas in the same test cluster within a preset correction time.

[0098] A correction subunit, connected to the turnover fluctuation calculation subunit, is used to calculate the relative deviation between the turnover fluctuation value to be measured and the average turnover fluctuation value of concern, to obtain the turnover fluctuation deviation; and, when the turnover fluctuation deviation is less than a preset turnover fluctuation deviation threshold, the sorting area to be measured and all the concern areas are determined to be correction areas; and, when the turnover fluctuation deviation is greater than or equal to the preset turnover fluctuation deviation threshold, all the concern areas are determined to be correction areas.

[0099] The preset correction time is the length of time used to calculate the fluctuation of spatial turnover rate in each area of ​​interest within the cluster under test. It depends on the sorting task cycle of the logistics center, the parcel processing frequency, and the system's response speed requirements, and is usually set between 10 and 30 minutes. In this embodiment, it is set to 20 minutes, which can capture the fluctuation of spatial turnover rate in a timely manner, while avoiding frequent misjudgments caused by too short a time, thereby improving the stability and reliability of the system.

[0100] The preset turnover fluctuation deviation threshold is a standard value used to determine whether the turnover fluctuation of the sorting area under test is synchronized with the area of ​​interest. It depends on the system's fault tolerance, the complexity of the sorting task, and the sensitivity to abnormal situations, and is usually set between 0.1 and 0.3. In this embodiment, it is set to 0.2 to ensure that the sorting area under test and the area of ​​interest are highly consistent before being identified as a correction area, thereby improving the accuracy of the system and reducing false alarms.

[0101] By clustering, spatially related areas are transformed into logical clusters to be tested. Physically adjacent areas form a community of shared destiny due to the continuity of the work process, and their performance fluctuations are transmissible. By capturing these potential risk units, the dynamic fluctuation of spatial turnover rate is used as a quantitative indicator of performance correlation. When the turnover fluctuation of the sorting area under test within the cluster is highly consistent with the average fluctuation of the core problem areas in the area of ​​concern, that is, the deviation is less than the preset turnover fluctuation deviation threshold, it indicates that they are in the same unstable system, and the risk has spread throughout the entire cluster, so the entire cluster needs to be corrected together. If the deviation is greater than the preset turnover fluctuation deviation threshold, it indicates that the risk is still limited to the original problem point. This realizes the upgrade from point-based early warning to cluster-based risk prevention and control, and accurately depicts the real impact range of the bottleneck effect.

[0102] Specifically, the anomaly determination module includes:

[0103] The key area determination unit is used to determine several key areas based on the preset artificial intelligence model and the queue growth rate of each of the sorting areas to be tested.

[0104] A ratio calculation unit, connected to the key point determination unit, is used to calculate the ratio of the number of key areas to the number of correction areas to obtain a quantity ratio.

[0105] An overlap calculation unit, connected to the ratio calculation unit, is used to count the number of overlapping areas in all the key areas and all the correction areas when the quantity ratio is within a preset ratio range, to obtain the overlap quantity; when the number of key areas is greater than the number of correction areas, to calculate the ratio of the overlap quantity to the number of key areas, to obtain the overlap degree; or, when the number of key areas is less than the number of correction areas, to calculate the ratio of the overlap quantity to the number of correction areas, to obtain the overlap degree.

[0106] An anomaly determination unit, which is connected to the overlap calculation unit, is used to determine several of the anomaly regions based on the overlap degree.

[0107] The preset ratio range is a standard value used to determine whether the ratio of the quantity in the key area to the quantity in the correction area is within a reasonable range. It depends on the complexity of the sorting task in the logistics center, the fault tolerance of the system, and its sensitivity to abnormal situations, and is usually set between 0.5 and 2.0. In this embodiment, it is set to 0.8 to 1.2, which can effectively balance the quantity difference between the key area and the correction area, ensure that the overlap calculation is performed when the quantity ratio is reasonable, and avoid misjudgment caused by excessive quantity difference.

[0108] By employing dual-source verification of prediction and diagnosis, intelligent trade-offs and precise decision-making in anomaly region identification are achieved. The key areas predicted by AI from a single source or the correction areas diagnosed by the system both contain uncertainties. However, the quantitative relationship and spatial overlap between the two reflect the degree of consensus on the system state. When the quantities of both are comparable, overlap analysis is initiated, embodying the principle of equitable trade-offs. Using the minimum value as the denominator to calculate the overlap ensures that this indicator always reflects the proportion of the overlapping portion in a smaller set, thus keenly capturing the core consensus region. The comparability of the dual-source data is quickly determined through the quantitative ratio, and the consistency of the two in spatial judgment is precisely quantified through geometric overlap. Finally, based on the consensus strength, the truly abnormal regions requiring intervention are intelligently identified, significantly improving the accuracy and reliability of the early warning results.

[0109] Specifically, the anomaly determination unit includes:

[0110] The overlap fluctuation calculation subunit is used to calculate the standard deviation of all the overlap degrees within a preset defined period to obtain the overlap fluctuation value;

[0111] An anomaly determination subunit, connected to the overlap fluctuation calculation subunit, is used to determine all the correction regions as the abnormal regions when the overlap fluctuation value is greater than a preset overlap fluctuation threshold, and to determine all the key regions and all the correction regions as abnormal regions when the overlap fluctuation value is less than or equal to the preset overlap fluctuation threshold, thereby identifying several abnormal regions.

[0112] The preset timeframe is the length of time used to calculate the standard deviation of overlap. It depends on the sorting task cycle of the logistics center, the parcel processing frequency, and the system's response speed requirements, and is usually set between 10 and 30 minutes. In this embodiment, it is set to 15 minutes, which can capture fluctuations in overlap in a timely manner while avoiding frequent misjudgments caused by too short a time, thereby improving the stability and reliability of the system.

[0113] The preset overlap fluctuation threshold is a standard value used to determine whether the overlap fluctuation is stable. It depends on the system's fault tolerance, the complexity of the sorting task, and its sensitivity to abnormal situations, and is usually set between 0.1 and 0.3. In this embodiment, it is set to 0.2 to ensure that an area is only identified as an abnormal area when the overlap fluctuation exceeds a certain range, thereby improving the accuracy and reliability of the system and reducing false alarms.

[0114] The reliability of the consensus of the entire judgment system is measured by the historical volatility of the overlap between the two judgment sources. When the volatility is high, it indicates that the prediction and diagnosis conclusions are inconsistent for a long time, and the system is in a state of cognitive divergence. In this case, it is more prudent and reliable to choose the correction area based on real-time systematic diagnosis as the final abnormal area. When the volatility is low, it indicates that the two judgment sources are consistent for a long time, and the system is in a stable state of cognitive consensus. In this case, it is certain that the judgments of the two sources are combined to expand the warning range. By introducing time series analysis into the decision arbitration process and using the overlap volatility value to judge consistency, the most reliable anomaly judgment strategy can be dynamically selected, thereby maintaining a high degree of credibility of the warning results in a complex and ever-changing production environment.

[0115] Specifically, the adjustment module includes:

[0116] The accuracy fluctuation calculation unit is used to calculate the standard deviation of the sorting accuracy of all the abnormal areas to obtain the accuracy fluctuation value.

[0117] An area adjustment unit, connected to the accuracy fluctuation calculation unit, is used to increase the preset area threshold based on the relative deviation between the accuracy fluctuation value and the preset accuracy fluctuation threshold and a preset adjustment coefficient when the accuracy fluctuation value is greater than the preset accuracy fluctuation threshold. Here, H'=H×(1+b×|K-K'| / K'), H' is the area threshold adjustment, H is the preset area threshold, b is the preset adjustment coefficient, K is the accuracy fluctuation value, and K' is the preset accuracy fluctuation threshold.

[0118] An accuracy change rate calculation unit, which is connected to the accuracy fluctuation calculation unit, is used to calculate the change rate of the average accuracy of all the abnormal regions within the preset adjustment time when the accuracy fluctuation value is less than or equal to the preset accuracy fluctuation threshold, and obtain the accuracy change rate.

[0119] A collaborative adjustment unit, connected to the accuracy change rate calculation unit, is used to increase the preset collaborative threshold based on the relative deviation between the accuracy change rate and the preset change rate threshold and a preset adjustment coefficient when the accuracy change rate is greater than a preset change rate threshold. Here, F' = F × (1 + a × |E-E'| / E'), F' adjusts the collaborative threshold, F is the preset collaborative threshold, a is the preset adjustment coefficient, E is the accuracy change rate, and E' is the preset change rate threshold.

[0120] The preset accuracy fluctuation threshold is a standard value used to determine whether the sorting accuracy fluctuation exceeds the normal range. It depends on the system's fault tolerance, the complexity of the sorting task, and its sensitivity to abnormal situations, and is usually set between 0.05 and 0.15. In this embodiment, it is set to 0.1 to ensure that the preset threshold is only adjusted when the sorting accuracy fluctuation exceeds a certain range, thereby improving the system's accuracy and reliability.

[0121] The preset adjustment coefficient is a factor used to control the adjustment range of the preset area threshold. It depends on the system's adjustment sensitivity and response speed to changes in sorting accuracy, and is usually set between 0.1 and 0.3. In this embodiment, it is set to 0.2, which can moderately adjust the preset area threshold, avoid over-adjustment that could lead to system instability, and ensure that the system can respond quickly to changes in sorting accuracy.

[0122] The preset rate of change threshold is a standard value used to determine whether the rate of change in sorting accuracy exceeds the normal range. It depends on the system's fault tolerance, the complexity of the sorting task, and its sensitivity to abnormal situations, and is usually set between 0.05 and 0.15. In this embodiment, it is set to 0.1 to ensure that the preset collaborative threshold is only adjusted when the rate of change in sorting accuracy exceeds a certain range, thereby improving the accuracy and reliability of the system.

[0123] The preset adjustment coefficient is a factor used to control the adjustment range of the preset coordination threshold. It depends on the system's adjustment sensitivity and response speed to changes in sorting accuracy, and is usually set between 0.1 and 0.3. In this embodiment, it is set to 0.2, which can moderately adjust the preset coordination threshold, avoid over-adjustment that could lead to system instability, and ensure that the system can respond quickly to changes in sorting accuracy.

[0124] By constructing a two-layer threshold adaptive mechanism based on performance feedback, the system's judgment criteria are precisely matched with the dynamic operating environment. The sorting accuracy rate, as the final performance indicator, is used as the gold standard to verify the effectiveness of the threshold. Precise adjustments are made for two typical failure modes. When the accuracy rate fluctuates drastically, it indicates that the current area threshold is too sensitive, causing the system to frequently misjudge between normal fluctuations and real risks. Therefore, the preset area threshold needs to be increased to improve system stability. When the accuracy rate fluctuates steadily but continues to deteriorate, it indicates that there are unidentified hidden bottlenecks in the system. The problem lies in the dynamic instability diagnosis stage. Therefore, the collaborative threshold needs to be increased to capture more subtle signs of loss of control. By using two different indicators, the accuracy rate fluctuation value and the accuracy rate change rate, the system can intelligently distinguish the types of interference faced by the system, forming a closed-loop control strategy that can self-optimize according to changes in the operating environment.

[0125] Specifically, the early warning module includes:

[0126] The risk index calculation unit is used to perform maximum-min normalization on the space stacking area to obtain an area normalized value, and to perform maximum-min normalization on the reciprocal of the space turnover rate to obtain a turnover rate normalized value, and to perform maximum-min normalization on the reciprocal of the sorting accuracy rate to obtain an accuracy normalized value, and to perform weighted summation on the area normalized value, the turnover rate normalized value, the accuracy normalized value, the preset area weight, the preset turnover rate weight, and the preset accuracy weight to obtain the risk index;

[0127] A high-risk determination unit, which is connected to the risk index calculation unit, is used to determine the abnormal area as the high-risk area when the risk index is greater than a preset risk index threshold, thereby obtaining several high-risk areas.

[0128] An early warning unit, which is connected to the high-risk determination unit, is used to issue early warnings for all the high-risk areas.

[0129] The preset area weight reflects the importance of the stacked space area in risk assessment. It depends on the impact of the stacked space area on sorting efficiency and logistics operations, as well as the system's sensitivity to space utilization, and is typically set between 0.2 and 0.5. In this embodiment, it is set to 0.3. This setting reasonably reflects the importance of the stacked space area in risk assessment, avoids excessive influence of area factors on the risk index, and thus improves the overall accuracy of the system's assessment.

[0130] The preset turnover rate weight reflects the importance of space turnover rate in risk assessment. It depends on the impact of space turnover rate on sorting efficiency and logistics operations, as well as the system's sensitivity to turnover efficiency, and is typically set between 0.2 and 0.5. In this embodiment, it is set to 0.3, which reasonably reflects the importance of space turnover rate in risk assessment, avoids excessive influence of turnover rate factors on the risk index, and thus improves the overall accuracy of the system's assessment.

[0131] The preset accuracy weight reflects the proportion of importance of sorting accuracy in risk assessment. It depends on the impact of sorting accuracy on logistics operations and the system's sensitivity to accuracy, and is typically set between 0.2 and 0.5. In this embodiment, the preset accuracy weight is set to 0.4, which reasonably reflects the importance of sorting accuracy in risk assessment, ensuring that accuracy has an appropriate proportion in the risk index, thereby improving the overall accuracy of the system's assessment.

[0132] The preset risk index threshold is a standard value used to determine whether the risk index exceeds the normal range. It depends on the system's fault tolerance, the complexity of the sorting task, and its sensitivity to abnormal situations, and is usually set between 0.6 and 0.9. In this embodiment, it is set to 0.7, which can effectively distinguish between normal risk and high risk, ensuring that an alert is only issued when the risk index is high, avoiding frequent false alarms, and improving the reliability and usability of the system.

[0133] By constructing a multi-dimensional risk fusion decision-making model, accurate mapping and efficient decision-making from multi-source heterogeneous data to a unified risk level are achieved. The operational risk of the sorting area is the comprehensive result of the combined effects of static load, dynamic efficiency, and final quality. Any single indicator can only reflect one aspect of the risk. First, the influence of different dimensions is eliminated by max-min normalization. Then, a comprehensive and quantitative risk index is generated by weighted summation with preset weights. This ensures that the system can accurately identify those high-risk areas that perform poorly in all three dimensions of load, efficiency, and quality from the abnormal areas screened through the aforementioned complex process, based on unified and objective standards. This allows for the issuance of accurate and effective early warnings, realizing the intelligent transformation from complex information to concise decision-making.

[0134] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automated system for self-learning artificial intelligence models, characterized in that, include: The acquisition module is used to obtain the spatial stacking area, queue growth rate, sorting rate, space turnover rate, and sorting accuracy rate of each sorting area to be tested in the logistics center within the past preset sorting cycle. The determination module is used to determine several temporary areas based on the spatial accumulation area and a preset area threshold. The attention determination module is used to determine several attention areas based on the queue growth rate, the sorting rate, and a preset collaboration threshold of each of the temporary areas. The correction module is used to correct the location of each of the interest areas, the spatial turnover rate, and the spatial turnover rate of each of the sorting areas to be tested, so as to obtain a number of correction areas. An anomaly determination module is used to determine several anomaly regions based on a preset artificial intelligence model, the queue growth rate of each of the sorting regions to be tested, and each of the correction regions. An adjustment module is used to adjust the preset area threshold or the preset coordination threshold according to the sorting accuracy of each abnormal area within a preset adjustment period. The early warning module is used to issue early warnings for several high-risk areas determined based on the spatial accumulation area, spatial turnover rate, and sorting accuracy of each of the abnormal areas, which are obtained again based on the adjusted preset area threshold or the preset collaborative threshold. The attention determination module includes: Focus on the fluctuation calculation unit, which is used to calculate the growth fluctuation value based on all the queue growth rates within a preset time period, and obtain several growth fluctuation values; and calculate the sorting fluctuation value based on all the sorting rates within a preset time period, and obtain several sorting fluctuation values. A focus determination unit, connected to the focus fluctuation calculation unit, is used to determine several focus areas based on all the growth fluctuation values ​​and all the sorting fluctuation values; The attention determination unit includes: The coordination degree calculation subunit is used to calculate the change coordination degree based on all the growth fluctuation values ​​and all the sorting fluctuation values; A focus determination subunit, connected to the coordination degree calculation subunit, is used to determine the temporary region as the focus region based on the comparison result of the changed coordination degree and the preset coordination threshold, thereby determining several focus regions; The correction module includes: The dispersion calculation unit is used to calculate the dispersion based on the location coordinates of each of the regions of interest. A correction unit, connected to the dispersion calculation unit, is used to make corrections based on the comparison results of the dispersion and the preset dispersion threshold, according to the spatial turnover rate of each of the areas of interest and the spatial turnover rate of each of the sorting areas to be tested, to obtain several correction areas.

2. The automated system for self-learning artificial intelligence models according to claim 1, characterized in that, The correction unit includes: The test cluster is determined by a sub-unit, which is used to cluster according to the location of each region of interest to obtain several test clusters; A turnover fluctuation calculation subunit, which is connected to the test cluster determination subunit, is used to calculate the average turnover fluctuation of the focus based on the total spatial turnover rate of each focus area in the test cluster within a preset correction time, and to calculate the test turnover fluctuation value based on the spatial turnover rate of each test sorting area in the same test cluster within a preset correction time. A correction subunit, connected to the turnover fluctuation calculation subunit, is used to calculate the turnover fluctuation deviation based on the turnover fluctuation value to be measured and the average turnover fluctuation value of concern, and to determine the sorting area to be measured and all the concern areas as correction areas based on a first comparison result of the turnover fluctuation deviation and a preset turnover fluctuation deviation threshold, and to determine all the concern areas as correction areas based on a second comparison result of the turnover fluctuation deviation and the preset turnover fluctuation deviation threshold.

3. The automated system for self-learning artificial intelligence models according to claim 2, characterized in that, The anomaly determination module includes: The key area determination unit is used to determine several key areas based on the preset artificial intelligence model and the queue growth rate of each of the sorting areas to be tested. A ratio calculation unit, connected to the key point determination unit, is used to calculate the ratio of the number of key areas to the number of correction areas to obtain a quantity ratio. An overlap calculation unit, connected to the ratio calculation unit, is used to count the number of overlapping areas in all the key areas and all the correction areas when the quantity ratio is within a preset ratio range, to obtain the overlap quantity; when the number of key areas is greater than the number of correction areas, to calculate the ratio of the overlap quantity to the number of key areas, to obtain the overlap degree; or, when the number of key areas is less than the number of correction areas, to calculate the ratio of the overlap quantity to the number of correction areas, to obtain the overlap degree. An anomaly determination unit, which is connected to the overlap calculation unit, is used to determine several of the anomaly regions based on the overlap degree.

4. The automated system for self-learning artificial intelligence models according to claim 3, characterized in that, The anomaly determination unit includes: The overlap fluctuation calculation subunit is used to calculate the overlap fluctuation value based on all the overlap within a preset period. An anomaly determination subunit, connected to the overlap fluctuation calculation subunit, is used to determine all the correction regions as the abnormal regions based on a first comparison result of the overlap fluctuation value and a preset overlap fluctuation threshold, and to determine all the key regions and all the correction regions as abnormal regions based on a second comparison result of the overlap fluctuation value and the preset overlap fluctuation threshold, thereby identifying several abnormal regions.

5. The automated system for self-learning artificial intelligence models according to claim 4, characterized in that, The adjustment module includes: An accuracy fluctuation calculation unit is used to calculate an accuracy fluctuation value based on the total sorting accuracy of all the abnormal areas. An area adjustment unit, which is connected to the accuracy fluctuation calculation unit, is used to adjust the preset area threshold according to the first comparison result of the accuracy fluctuation value and the preset accuracy fluctuation threshold. An accuracy change rate calculation unit, which is connected to the accuracy fluctuation calculation unit, is used to calculate the accuracy change rate based on the second comparison result of the accuracy fluctuation value and the preset accuracy fluctuation threshold, according to the average accuracy of all the abnormal regions within the preset adjustment time. A collaborative adjustment unit, connected to the accuracy change rate calculation unit, is used to adjust the preset collaborative threshold based on the comparison result between the accuracy change rate and the preset change rate threshold.

6. The automated system for self-learning artificial intelligence models according to claim 5, characterized in that, The early warning module includes: The risk index calculation unit is used to calculate the risk index based on the space stacking area, the space turnover rate, and the sorting accuracy rate. A high-risk determination unit, which is connected to the risk index calculation unit, is used to determine the abnormal area as the high-risk area based on the comparison result between the risk index and the preset risk index threshold, thereby obtaining several high-risk areas; An early warning unit, which is connected to the high-risk determination unit, is used to issue early warnings for all the high-risk areas.

7. The automated system for self-learning artificial intelligence models according to claim 6, characterized in that, The temporary area is determined based on the comparison between the spatial accumulation area and the preset area threshold.

Images